Humanoid robot starting method based on policy network trained by deep learning, humanoid robot, electronic device and computer readable storage medium

Abstract

The application relates to the technical field of robots, and discloses a humanoid robot starting method based on a strategy network trained by deep learning, a humanoid robot, an electronic device and a computer readable storage medium, the starting method comprising the following steps: after a user controller receives a current posture of a humanoid robot, the current posture of the humanoid robot is matched with a starting posture selected by a user; if the current posture matches the starting posture, a first starting instruction including the starting posture is triggered; after the first starting instruction is received by the humanoid robot, the humanoid robot starts to execute a starting operation corresponding to a suspended hoisting posture. Through the matching of the starting posture selected by the user and the current posture of the humanoid robot, the damage and failure of the humanoid robot caused by the execution of a starting operation inconsistent with the current posture due to the selection of a wrong starting posture by the user can be avoided through the matching operation.

Description

Technical Field

[0001] This application relates to the field of robotics, and more specifically to a method for initiating a humanoid robot based on a policy network trained by deep learning, a humanoid robot, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In existing technologies, certain humanoid robots need to be activated to enter a standing state before performing other actions. In current applications, before activation, a step is set up where the robot's activation posture is selected via an app. This means that the human eye first determines the robot's activation posture before the activation operation, and then the robot is manually selected via the app to perform the corresponding activation operation based on this determined posture.

[0003] However, when using humanoid robots, users may mistakenly select a different starting posture than the robot's current posture due to improper operation or lack of familiarity with the system. Alternatively, when users need to start the robot at a distance or in a different location, they may be unable to clearly see or determine the robot's current posture. This could lead to a discrepancy between the user-selected starting posture and the robot's current posture. In such cases, the robot might execute a starting operation inconsistent with its current posture, resulting in an erroneous start-up and potentially causing damage or malfunction. Summary of the Invention

[0004] This application provides a method for initiating a humanoid robot based on a policy network trained by deep learning, a humanoid robot, an electronic device, and a computer-readable storage medium, which can solve at least some of the above-mentioned technical problems.

[0005] Firstly, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, including: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the current posture of the humanoid robot is received. If they match, the first start command including the start posture is allowed to be triggered. The start posture is the suspended hoisting posture. Receive the first start command, including the start attitude, issued by the user controller; The motion control mode of the humanoid robot is switched sequentially from zero torque mode to damped mode, and then from damped mode to upright mode; Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0006] As mentioned earlier, the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture. Only when the match, or verification, is successful will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that is inconsistent with the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0007] Optionally, in some possible implementations of the first aspect, foot contact detection is performed in upright mode, including: In upright mode, the actual output torque of the motors corresponding to the leg joints of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Determine whether the sole of the foot is in contact with the ground based on the contact force of the foot.

[0008] Therefore, automatically detecting foot contact based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can prevent the humanoid robot from being unstable due to the user's misjudgment of foot contact with the ground by the naked eye, and thus prevent the humanoid robot from falling over due to instability.

[0009] Secondly, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applied to a user controller, comprising: In response to the selection operation, a first detection instruction is generated and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. Receive the current posture of the humanoid robot reported by the humanoid robot; The current posture of the humanoid robot is matched and verified with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is the suspended posture. A first start command including a start posture is sent to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot, after receiving the first start command including the start posture from the user controller, to sequentially switch the motion control mode of the humanoid robot from zero torque mode to damped mode, and then from damped mode to upright mode; and, in upright mode, to perform foot contact detection. When foot contact is detected, the motion control mode is switched to the first policy network.

[0010] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0011] Optionally, in some possible implementations of the second aspect, the user controller is located in an electronic device communicatively connected to the humanoid robot, allowing the triggering of a first initiation command including an initiation posture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After the first start command, including the start posture, is triggered, the electronic device is controlled to display the first start page. In response to the user's start trigger operation on the first start page, the first start command, including the start posture, is generated.

[0012] Therefore, after the first start command, including the start posture, is allowed to be triggered, the first start command is generated in a corresponding manner according to the settings of the electronic device, so that the humanoid robot can receive the first start command and execute the corresponding start operation.

[0013] Optionally, in some possible implementations of the second aspect, a first detection instruction is generated in response to a selection operation, including: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

[0014] Since the user controller only generates the first detection command in response to the selection operation after the user completes the selection operation of the start posture, and then the humanoid robot obtains its current posture after receiving the first detection command, the humanoid robot is restricted to performing subsequent start operations only after the user completes the selection operation of the start posture. This avoids the situation where the humanoid robot performs the start operation by mistake before the user completes the selection operation of the start posture.

[0015] Optionally, in some possible implementations of the second aspect, the user controller is located in an electronic device communicatively connected to the humanoid robot, and after sending a first initiation command, including an initiation posture, to the humanoid robot, the method further includes: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0016] Since users can perform the first operation of bringing the feet of the humanoid robot close to the ground by following the startup prompts displayed on the second startup page on the electronic device, users do not need to carry the instruction manual at all times or memorize the specific steps of the first operation. Instead, they can simply follow the startup prompts to perform the first operation. This improves the ease of use of starting the humanoid robot in a suspended posture, saves users' learning time, and improves operational efficiency.

[0017] Thirdly, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; The humanoid robot responds to the first detection command sent by the user controller, obtains the current posture of the humanoid robot, and reports it to the user controller; The user controller receives the current posture of the humanoid robot reported by the humanoid robot; The user controller matches and verifies the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is the suspended hoisting posture. The user controller sends a first start command, including the start posture, to the humanoid robot; The humanoid robot receives a first start command from the user controller, including the start posture; The humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0018] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0019] Fourthly, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, including: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the current posture of the humanoid robot is received. If they match, the first start command including the start posture is allowed to be triggered. The start posture is a lying posture. Receive the first start command, including the start attitude, issued by the user controller; The motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot is made to stand up. After the action is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0020] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0021] Fifthly, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, comprising: In response to the selection operation, a first detection instruction is generated and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. Receive the current posture of the humanoid robot reported by the humanoid robot; The current posture of the humanoid robot is matched and verified with the starting posture selected by the user. If they match, the first start command, which includes the starting posture, is allowed to be triggered. The starting posture is a lying posture. A first start command including the start posture is sent to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the second strategy network and perform the action of standing up after receiving the first start command including the start posture from the user controller. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0022] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0023] Optionally, in some possible implementations of the fifth aspect, the user controller is located in an electronic device communicatively connected to the humanoid robot, allowing the triggering of a first initiation command, including an initiation gesture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After the first start command, including the start posture, is triggered, the electronic device is controlled to display the first start page. In response to the user's start trigger operation on the first start page, the first start command, including the start posture, is generated.

[0024] Therefore, after the first start command, including the start posture, is allowed to be triggered, the first start command is generated in a corresponding manner according to the settings of the electronic device, so that the humanoid robot can receive the first start command and execute the corresponding start operation.

[0025] Optionally, in some possible implementations of the fifth aspect, a first detection instruction is generated in response to a selection operation, including: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

[0026] Since the user controller only generates the first detection command in response to the selection operation after the user completes the selection operation of the start posture, and then the humanoid robot obtains its current posture after receiving the first detection command, the humanoid robot is restricted to performing subsequent start operations only after the user completes the selection operation of the start posture. This avoids the situation where the humanoid robot performs the start operation by mistake before the user completes the selection operation of the start posture.

[0027] Sixthly, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; The humanoid robot responds to the first detection command sent by the user controller, obtains the current posture of the humanoid robot, and reports it to the user controller; The user controller receives the current posture of the humanoid robot reported by the humanoid robot; The user controller matches and verifies the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is a lying posture. The user controller sends a first start command, including the start posture, to the humanoid robot; The humanoid robot receives a first start command from the user controller, including the start posture; The humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0028] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0029] Seventhly, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applied to a humanoid robot, including: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the current posture of the humanoid robot is received. If they match, the first start command including the start posture is allowed to be triggered. The start posture is a squatting posture. Receive the first start command, including the start attitude, issued by the user controller; The motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the robot is made to stand up. After the action is completed, it is switched back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0030] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0031] Eighthly, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, comprising: In response to the selection operation, a first detection instruction is generated and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. Receive the current posture of the humanoid robot reported by the humanoid robot; The current posture of the humanoid robot is matched and verified with the starting posture selected by the user. If they match, the first start command, which includes the starting posture, is allowed to be triggered. The starting posture is a squatting posture. A first start command including the start posture is sent to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot to switch its motion control mode from zero torque mode to the third strategy network and perform the action of standing up after receiving the first start command including the start posture from the user controller. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0032] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0033] Optionally, in some possible implementations of the eighth aspect, the user controller is located in an electronic device communicatively connected to the humanoid robot, allowing the triggering of a first initiation command including an initiation posture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After the first start command, including the start posture, is triggered, the electronic device is controlled to display the first start page. In response to the user's start trigger operation on the first start page, the first start command, including the start posture, is generated.

[0034] Therefore, after the first start command, including the start posture, is allowed to be triggered, the first start command is generated in a corresponding manner according to the settings of the electronic device, so that the humanoid robot can receive the first start command and execute the corresponding start operation.

[0035] Optionally, in some possible implementations of the eighth aspect, a first detection instruction is generated in response to a selection operation, including: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

[0036] Therefore, the user controller will only generate the first detection command in response to the selection operation after the user completes the selection operation of the start posture. After receiving the first detection command, the humanoid robot obtains its current posture. This restricts the humanoid robot to perform subsequent start operations only after the user completes the selection operation of the start posture, thus avoiding the situation where the humanoid robot performs the start operation by mistake before the user completes the selection operation of the start posture.

[0037] Ninthly, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; The humanoid robot responds to the first detection command sent by the user controller, obtains the current posture of the humanoid robot, and reports it to the user controller; The user controller receives the current posture of the humanoid robot reported by the humanoid robot; The user controller matches and verifies the current posture of the humanoid robot with the starting posture selected by the user. If they match, the first start command, which includes the starting posture, is allowed to be triggered. The starting posture is a squatting posture. The user controller sends a first start command, including the start posture, to the humanoid robot; The humanoid robot receives a first start command from the user controller, including the start posture; The humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0038] Because the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture, and only when the match, that is, the verification is successful, does the humanoid robot execute the start operation corresponding to the start posture. Thus, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0039] In a tenth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: Receive a first start command including the start posture selected by the user, wherein the start posture is the suspended hoisting posture; Obtain the current pose of the humanoid robot; The current posture and the starting posture of the humanoid robot are matched and verified; If a match is found, the motion control mode of the humanoid robot will be switched sequentially from zero torque mode to damped mode, and then from damped mode to upright mode. Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0040] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0041] Optionally, in some possible implementations of aspect ten, foot contact detection is performed in upright mode, including: In upright mode, the actual output torque of the motors corresponding to the leg joints of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Determine whether the sole of the foot is in contact with the ground based on the contact force of the foot.

[0042] Therefore, automatically detecting foot contact based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can prevent the humanoid robot from being unstable due to the user's misjudgment of foot contact with the ground by the naked eye, and thus prevent the humanoid robot from falling over due to instability.

[0043] In its eleventh aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, comprising: In response to a selection operation, a first start command including the user-selected start posture is generated and sent to the humanoid robot. The start posture is a suspended posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain the current posture of the humanoid robot after receiving the first start command including the user-selected start posture. The current posture of the humanoid robot and the start posture are matched and verified. If they match, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. In the upright mode, foot contact detection is performed. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0044] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0045] Optionally, in some possible implementations of the eleventh aspect, the user controller is located in an electronic device communicatively connected to the humanoid robot, and after sending a first activation command, including a user-selected activation posture, to the humanoid robot, the method further includes: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0046] Therefore, users can perform the first operation, which allows the humanoid robot's feet to touch the ground, based on the startup prompts displayed on the second startup page on the electronic device. This eliminates the need for users to carry an instruction manual or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, improving the ease of use of starting the humanoid robot in a suspended posture, saving users' learning time, and increasing operational efficiency.

[0047] In a twelfth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: In response to the selection operation, the user controller generates a first start command including the start posture selected by the user and sends the first start command including the start posture selected by the user to the humanoid robot. The start posture is a suspended posture. The humanoid robot receives a first start command, including the start posture selected by the user. The humanoid robot acquires its current posture; The humanoid robot performs a matching and verification process between its current posture and its initial posture. If matched, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode; The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0048] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0049] In a thirteenth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: Receives a first startup command including the user-selected startup posture, which is a lying posture. Obtain the current pose of the humanoid robot; The current posture and the starting posture of the humanoid robot are matched and verified; If a match is found, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot is made to stand up. After the action is completed, the robot is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0050] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0051] In a fourteenth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applied to a user controller, comprising: In response to the selection operation, a first start command including the user-selected start posture is generated and sent to the humanoid robot. The start posture is a lying posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain the current posture of the humanoid robot after receiving the first start command including the user-selected start posture. The current posture of the humanoid robot and the start posture are matched and verified. If they match, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the action of getting up is executed. After the execution is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0052] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0053] In a fifteenth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: In response to the selection operation, the user controller generates a first start command including the start posture selected by the user and sends the first start command including the start posture selected by the user to the humanoid robot. The start posture is a lying posture. The humanoid robot receives a first start command, including the start posture selected by the user. The humanoid robot acquires its current posture; The humanoid robot performs a matching and verification process between its current posture and its initial posture. If a match is found, the humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0054] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0055] In a sixteenth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: Receive a first startup command including the user-selected startup posture, which is a squatting posture; Obtain the current pose of the humanoid robot; The current posture and the starting posture of the humanoid robot are matched and verified; If a match is found, the motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the robot performs the action of standing up. After the action is completed, the robot is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0056] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0057] In a seventeenth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, comprising: In response to the selection operation, a first start command including the user-selected start posture is generated and sent to the humanoid robot. The start posture is a squatting posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain the current posture of the humanoid robot after receiving the first start command including the user-selected start posture. The current posture of the humanoid robot and the start posture are matched and verified. If they match, the motion control mode of the humanoid robot is switched from zero torque mode to the third policy network and the standing action is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0058] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0059] In its eighteenth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: In response to the selection operation, the user controller generates a first start command including the start posture selected by the user and sends the first start command including the start posture selected by the user to the humanoid robot. The start posture is a squatting posture. The humanoid robot receives a first start command, including the start posture selected by the user. The humanoid robot acquires its current posture; The humanoid robot performs a matching and verification process between its current posture and its initial posture. If a match is found, the humanoid robot will switch its motion control mode from zero torque mode to the third strategy network and perform the action of standing up. After the action is completed, it will switch back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0060] Because the humanoid robot matches the user-selected starting posture with the robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match or verification is successful, the verification operation can avoid damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a starting operation that does not match the current posture.

[0061] In its nineteenth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: In response to the second start command, detect the current posture of the humanoid robot; When the current posture of the humanoid robot is detected to be a suspended posture, the motion control mode of the humanoid robot is switched sequentially from zero torque mode to damped mode, and then from damped mode to upright mode. Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0062] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0063] Optionally, in some possible implementations of aspect nineteen, foot contact detection is performed in upright mode, including: In upright mode, the actual output torque of the motors corresponding to the leg joints of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Determine whether the sole of the foot is in contact with the ground based on the contact force of the foot.

[0064] Therefore, automatically detecting foot contact with the ground based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can avoid the problem of the humanoid robot being unstable due to the user's misjudgment of foot contact with the ground by relying solely on the naked eye.

[0065] In a twentieth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applied to a user controller, comprising: A second start command is generated in response to a trigger operation and sent to the humanoid robot. The second start command instructs the humanoid robot to detect its current posture. When the current posture of the humanoid robot is detected to be a suspended posture, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. In the upright mode, foot contact detection is performed. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0066] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0067] Optionally, in some possible implementations of the twentieth aspect, the user controller is located in an electronic device communicatively connected to the humanoid robot, and after sending the second start command to the humanoid robot, the method further includes: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0068] Therefore, users can perform the first operation, which allows the humanoid robot's feet to touch the ground, based on the startup prompts displayed on the second startup page on the electronic device. This eliminates the need for users to carry an instruction manual or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, improving the ease of use of starting the humanoid robot in a suspended posture, saving users' learning time, and increasing operational efficiency.

[0069] In its twentieth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; The humanoid robot responds to the second start command by detecting its current posture; When the current posture of the humanoid robot is detected to be a suspended posture, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0070] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0071] In a twenty-second aspect, this application provides a humanoid robot initiation method based on a policy network trained through deep learning, applicable to humanoid robots, comprising: In response to the second start command, detect the current posture of the humanoid robot; When the humanoid robot's current posture is a lying posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the second policy network and the action of getting up is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0072] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0073] In a twentieth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, comprising: A second start command is generated in response to the trigger operation and sent to the humanoid robot. The second start command is used to instruct the humanoid robot to detect the current posture of the humanoid robot in response to the second start command. When the current posture of the humanoid robot is a lying posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the second policy network and the action of getting up is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0074] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0075] In a twentieth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; The humanoid robot responds to the second start command by detecting its current posture; When the humanoid robot's current posture is a lying posture, the humanoid robot switches its motion control mode from zero torque mode to the second policy network and performs the action of getting up. After the action is completed, it switches back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0076] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0077] In its twentieth aspect, this application provides a humanoid robot initiation method based on a policy network trained through deep learning, applicable to humanoid robots, comprising: In response to the second start command, detect the current posture of the humanoid robot; When the humanoid robot's current posture is a squatting posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the third policy network and the action of standing up is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0078] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0079] In a twentieth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, comprising: A second start command is generated in response to the trigger operation and sent to the humanoid robot. The second start command is used to instruct the humanoid robot to detect the current posture of the humanoid robot in response to the second start command. When the current posture of the humanoid robot is a squatting posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the third policy network and the action of standing up is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0080] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0081] In its twenty-seventh aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; The humanoid robot responds to the second start command by detecting its current posture; When the humanoid robot's current posture is a squatting posture, the humanoid robot switches its motion control mode from zero torque mode to the third policy network and performs the action of standing up. After the action is completed, it switches back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0082] Therefore, without the user selecting a starting posture, the humanoid robot directly detects its current posture after receiving the second start command and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunctions to the humanoid robot caused by the user selecting the wrong starting posture and the robot executing a start operation that is inconsistent with the current posture.

[0083] In its twenty-eighth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is a suspended posture. In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0084] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0085] Optionally, in some possible implementations of aspect 28, foot contact detection is performed in upright mode, including: In upright mode, the actual output torque of the motors corresponding to the leg joints of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Determine whether the sole of the foot is in contact with the ground based on the contact force of the foot.

[0086] Therefore, automatically detecting foot contact with the ground based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can avoid the problem of the humanoid robot being unstable due to the user's misjudgment of foot contact with the ground by relying solely on the naked eye.

[0087] In its twenty-ninth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, the method comprising: In response to the first triggering operation, a second detection command is generated and sent to the humanoid robot. The second detection command is used to instruct the humanoid robot to acquire the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a suspended posture. Receive the current posture of the humanoid robot sent by the humanoid robot; In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command instructs the humanoid robot to sequentially switch its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode in response to the confirmation start command sent by the user controller; and, in upright mode, foot contact detection is performed, and when foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0088] Without requiring the user to select a starting posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. Once a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process prevents damage and malfunctions caused by the humanoid robot directly executing a start operation that does not match the current posture if the user selects the wrong starting posture.

[0089] Optionally, in some possible implementations of aspect twenty-nine, the user controller is located in an electronic device communicatively connected to the humanoid robot, and after sending a confirmation start command to the humanoid robot, the method further includes: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0090] Therefore, users can perform the first operation, which allows the humanoid robot's feet to touch the ground, based on the startup prompts displayed on the second startup page on the electronic device. This eliminates the need for users to carry an instruction manual or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, improving the ease of use of starting the humanoid robot in a suspended posture, saving users' learning time, and increasing operational efficiency.

[0091] Optionally, in some possible implementations of aspect 29, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: After receiving the current posture of the humanoid robot from the humanoid robot, the control electronic device displays the first startup page and generates a confirmation startup command in response to the user's startup trigger operation on the first startup page.

[0092] Therefore, a confirmation start command will only be generated when the user performs a start trigger operation on the first start page, so that the humanoid robot can perform the corresponding start operation. This allows the humanoid robot to avoid damage or malfunction caused by the user selecting the wrong start posture and the robot directly performing a start operation that does not match the current posture.

[0093] In a thirtieth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a suspended posture. The user controller receives the current posture of the humanoid robot sent by the humanoid robot; The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; In response to the confirmation start command sent by the user controller, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0094] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0095] In its thirty-first aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is a lying posture. In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0096] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0097] In a thirty-second aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, the method comprising: In response to the first triggering operation, a second detection command is generated and sent to the humanoid robot. The second detection command is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a lying posture. Receive the current posture of the humanoid robot sent by the humanoid robot; In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to switch its motion control mode from zero torque mode to the second strategy network and perform a standing action in response to the confirmation start command sent by the user controller. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0098] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0099] Optionally, in some possible implementations of aspect thirty-two, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: After receiving the current posture of the humanoid robot from the humanoid robot, the control electronic device displays the first startup page and generates a confirmation startup command in response to the user's startup trigger operation on the first startup page.

[0100] Therefore, a confirmation start command will only be generated when the user performs a start trigger operation on the first start page, so that the humanoid robot can perform the corresponding start operation. This allows the humanoid robot to avoid damage or malfunction caused by the user selecting the wrong start posture and the robot directly performing a start operation that does not match the current posture.

[0101] In its thirty-third aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a lying posture. The user controller receives the current posture of the humanoid robot sent by the humanoid robot; The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; In response to the confirmation start command sent by the user controller, the humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0102] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0103] In its thirty-fourth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applicable to humanoid robots, comprising: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is a squatting posture. In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the robot performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0104] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0105] In a thirty-fifth aspect, this application provides a humanoid robot initiation method based on a policy network trained by deep learning, applied to a user controller, comprising: In response to the first triggering operation, a second detection command is generated and sent to the humanoid robot. The second detection command is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a squatting posture. Receive the current posture of the humanoid robot sent by the humanoid robot; In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to switch its motion control mode from zero torque mode to the third strategy network and perform a standing action in response to the confirmation start command sent by the user controller. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0106] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0107] Optionally, in some possible implementations of aspect thirty-five, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: After receiving the current posture of the humanoid robot from the humanoid robot, the control electronic device displays the first startup page and generates a confirmation startup command in response to the user's startup trigger operation on the first startup page.

[0108] Therefore, a confirmation start command will only be generated when the user performs a start trigger operation on the first start page, so that the humanoid robot can perform the corresponding start operation. This allows the humanoid robot to avoid damage or malfunction caused by the user selecting the wrong start posture and the robot directly performing a start operation that does not match the current posture.

[0109] In its thirty-sixth aspect, this application provides a method for initiating a humanoid robot based on a policy network trained through deep learning, comprising: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a squatting posture. The user controller receives the current posture of the humanoid robot sent by the humanoid robot; The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; In response to the confirmation start command sent by the user controller, the humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0110] Therefore, without the user selecting a start posture, the humanoid robot directly obtains its current posture after receiving the second detection command and reports it to the user controller. The user then determines whether the humanoid robot's current posture matches the actual current posture. After a match is found and the user's verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. This verification process avoids situations where the humanoid robot directly executes a start operation that does not match the current posture if the user selects the wrong start posture, thus preventing damage or malfunctions to the humanoid robot.

[0111] In a thirty-seventh aspect, this application provides a humanoid robot, comprising: a first processor and a first memory, the first memory being connected to the first processor, the first memory storing a computer program, and the first processor executing the humanoid robot startup method based on a policy network trained by deep learning as described above.

[0112] In a thirty-eighth aspect, this application provides an electronic device, including: a second processor and a second memory, the second memory being connected to the second processor, the second memory storing a computer program, and the second processor executing the humanoid robot initiation method based on a policy network trained by deep learning as described above.

[0113] In a thirty-ninth aspect, this application provides a computer-readable storage medium storing a computer program, which, when called by a processor, executes the humanoid robot initiation method based on a policy network trained by deep learning as described above. Attached Figure Description

[0114] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced. Obviously, the accompanying drawings described below are only some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0115] Figure 1 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the first embodiment of this application. Figure 2 This is a schematic diagram of the startup process of the humanoid robot in this application, whose current posture is a suspended posture and whose feet are not touching the ground; Figure 3 This is a schematic diagram of the startup process of the humanoid robot in this application, whose current posture is a suspended posture and whose feet have touched the ground; Figure 4 This is a schematic diagram of the foot contact detection process of the humanoid robot initiation method based on a policy network trained by deep learning in the first embodiment of this application. Figure 5 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the second embodiment of this application. Figure 6 This is a schematic diagram of the selected starting posture as the suspended hoisting posture in the first embodiment of this application; Figure 7 This is a schematic diagram of the first startup page in the first embodiment of this application; Figure 8 This is a schematic diagram of the second startup page in the first embodiment of this application; Figure 9 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the third embodiment of this application. Figure 10 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the fourth embodiment of this application. Figure 11 This is a schematic diagram of the startup process of the humanoid robot in this application, whose current posture is a lying posture; Figure 12 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the fifth embodiment of this application. Figure 13 This is a schematic diagram of the selected starting posture as a lying posture in the fifth embodiment of this application; Figure 14 This is a schematic diagram of the first startup page in the fifth embodiment of this application; Figure 15 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the sixth embodiment of this application. Figure 16 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the seventh embodiment of this application. Figure 17 This is a schematic diagram of the humanoid robot in this application, whose current posture is a squatting posture and whose startup process is inside the box; Figure 18 This is a schematic diagram of the humanoid robot in this application, whose current posture is a squatting posture and the startup process is on flat ground; Figure 19 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the eighth embodiment of this application. Figure 20 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the ninth embodiment of this application. Figure 21This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the tenth embodiment of this application. Figure 22 This is a schematic diagram of the foot contact detection process of the humanoid robot initiation method based on a policy network trained by deep learning in the tenth embodiment of this application. Figure 23 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the eleventh embodiment of this application. Figure 24 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twelfth embodiment of this application. Figure 25 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the thirteenth embodiment of this application. Figure 26 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the fourteenth embodiment of this application. Figure 27 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the fifteenth embodiment of this application. Figure 28 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the sixteenth embodiment of this application. Figure 29 This is a schematic diagram of the foot contact detection process of the humanoid robot initiation method based on a policy network trained by deep learning in the sixteenth embodiment of this application. Figure 30 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the seventeenth embodiment of this application. Figure 31 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the eighteenth embodiment of this application. Figure 32 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the nineteenth embodiment of this application. Figure 33 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twentieth embodiment of this application. Figure 34 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-first embodiment of this application. Figure 35This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-second embodiment of this application. Figure 36 This is a schematic diagram of the foot contact detection process of the humanoid robot initiation method based on a policy network trained by deep learning in the twenty-second embodiment of this application. Figure 37 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-third embodiment of this application. Figure 38 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-fourth embodiment of this application. Figure 39 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-fifth embodiment of this application. Figure 40 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-sixth embodiment of this application. Figure 41 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-seventh embodiment of this application. Figure 42 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-eighth embodiment of this application. Figure 43 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the twenty-ninth embodiment of this application. Figure 44 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained by deep learning in the thirtieth embodiment of this application. Figure 45 This is a schematic diagram of a humanoid robot according to one embodiment of this application; Figure 46 This is a schematic diagram of an electronic device according to an embodiment of this application.

[0116] Icon labels: Humanoid robot-31; First processor-311; First memory-312; Electronic device-32; Second processor-321; Second memory-322. Detailed Implementation

[0117] In the description of this application, terms such as “center,” “longitudinal,” “transverse,” “length,” “width,” “thickness,” “upper,” “lower,” “front,” “rear,” “left,” “right,” “vertical,” “horizontal,” “clockwise,” “counterclockwise,” “axial,” and “radial” indicate orientation or positional relationships based on the accompanying drawings. They are used only for ease of description and simplification and do not imply that the device or component must have a specific orientation or be constructed and operated in a specific orientation. They should not be regarded as limitations on this application.

[0118] The term "zero torque mode" refers to the working state in which the robot joint actuator does not actively output driving torque or the active output torque is zero or close to zero. The term "damped mode" refers to the working state in which the robot joint actuator outputs damping torque opposite to the direction of joint movement based on the trend of joint movement speed, angular velocity, or displacement. The term "upright mode" refers to the working state in which the robot joint actuator actively outputs torque to counteract gravity, stabilize posture, and move in the opposite direction of joint movement in order to maintain the robot's upright static state. The term "policy network" refers to the policy model or execution network in a reinforcement learning model.

[0119] The terms "first" and "second" are used descriptively only and do not indicate relative importance or imply the number of technical features. Therefore, the feature referred to as "first" or "second" may explicitly or implicitly include at least one of those features. "A plurality of" means at least two, unless otherwise expressly defined.

[0120] Unless otherwise specified, the terms “installation,” “connection,” “linking,” “fixing,” etc., should be interpreted broadly. They can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections, electrical connections, or connections that can communicate with each other; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components or the interaction between two components.

[0121] In this application, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," and "on top" can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "beneath" can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0122] Please see Figure 1 , Figure 1 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the first embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning is applied to a humanoid robot and includes: S101: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the current posture of the humanoid robot is received. If they match, the first start command including the start posture is allowed to be triggered. The start posture is the suspended hoisting posture. S102: Receive the first start command, including the start attitude, issued by the user controller; S103: Sequentially switch the motion control mode of the humanoid robot from zero torque mode to damped mode, and then switch it from damped mode to upright mode; S104: Perform foot contact detection in upright mode. When foot contact is detected, switch the motion control mode to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0123] In this embodiment, the humanoid robot receives and responds to the first detection command sent by the user controller, and then obtains the current posture of the humanoid robot. After receiving the current posture of the humanoid robot, the user controller performs a matching verification between the current posture of the humanoid robot and the start posture selected by the user. If they match, the first start command including the start posture is allowed to be triggered. After receiving the first start command, the humanoid robot begins to execute the start operation corresponding to the suspended posture, so as to finally stand stably on the ground after its feet touch the ground and complete the start.

[0124] As mentioned earlier, the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture. Only when the match, or verification, is successful will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that is inconsistent with the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0125] Furthermore, when the starting posture is a suspended hoisting posture, after receiving the first start command from the user controller, which includes the starting posture, the humanoid robot does not directly switch its motion control mode from zero torque mode to upright mode. Instead, it sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. This gradual mode switching of "zero torque mode → damped mode → upright mode" can prevent the hoisting frame of the humanoid robot in a suspended hoisting posture from being impacted and swaying during the release process.

[0126] In some embodiments, if there is a mismatch, the user controller outputs an alarm message to remind the user that the selected start posture is incorrect, or the user controller sends an alarm signal to the humanoid robot to control the humanoid robot to issue an alarm message. In this case, the humanoid robot remains stationary because it has not received the first start command.

[0127] In some embodiments, the output of alarm information may specifically be through voice broadcast, text output, vibration alert, etc.

[0128] In some embodiments, the humanoid robot includes an IMU (Inertial Measurement Unit) and a joint detection module. There may be at least two IMUs; when there are two IMUs, one is located at the robot's chest and the other at its hip to detect the robot's overall posture (e.g., torso tilt). Joint sensors are used to detect the angles and torques of each joint. Therefore, based on the posture detected by the IMUs and the detection data from the joint sensors, the current posture of the humanoid robot can be detected. Furthermore, the robot's gait can be adjusted based on the posture detected by the IMUs and the detection data from the joint sensors to maintain balance during working states (e.g., walking, bending over, etc.).

[0129] In some embodiments, when the humanoid robot is in a suspended posture, the IMU acquires acceleration and angular velocity information, and calculates the chest posture angle and hip posture angle based on the acceleration and angular velocity information. Then, it calculates the relative posture difference between the chest posture angle and the hip posture angle. When the relative posture difference is less than a preset relative posture threshold, and the duration of the humanoid robot's current posture reaches a preset time, it is determined that the humanoid robot is in an upright mode.

[0130] Specifically, the first strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up, which specifically means stopping and standing up during walking.

[0131] Please see Figure 2 , Figure 2 This is a schematic diagram of the startup process of the humanoid robot in this application, where the current posture is a suspended posture and the soles of its feet are not touching the ground. Figure 2 (2a)-(2f) in the text are the startup process when the current posture of the humanoid robot is suspended and the feet are not touching the ground. In the initial state, the current posture of the humanoid robot is suspended and the feet are not touching the ground. Then, after the humanoid robot receives the first startup command including the startup posture issued by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to damped mode, and then from damped mode to upright mode. The user gradually lowers the humanoid robot so that the feet touch the ground and achieve stable standing. Figure 2 (2e) and (2f) are schematic diagrams of a humanoid robot walking after it has stood stably.

[0132] Please see Figure 3 , Figure 3 This is a schematic diagram of the startup process of the humanoid robot in this application, whose current posture is a suspended posture and whose feet have touched the ground. Figure 3 (3a)-(3d) in the text are the start-up process of the humanoid robot in the current posture of being suspended and with its feet touching the ground. In the initial state, the current posture of the humanoid robot is suspended and with its feet touching the ground. After the humanoid robot receives the first start command, which includes the start posture, from the user controller, the motion control mode of the humanoid robot can be switched from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot can stand stably without the user having to gradually lower it down. Figure 3 (3c) and (3d) are schematic diagrams of a humanoid robot walking after it has stood stably.

[0133] In this application, "foot contact" specifically refers to the humanoid robot's feet being in complete contact with the ground so that the humanoid robot can stand stably on the ground.

[0134] In some embodiments, when the humanoid robot is in a suspended posture, it first determines the preset target position and posture to reach the upright mode, calculates the joint angle that each joint needs to rotate compared to the initial zero torque mode based on the preset target position and posture, and rotates each joint by the corresponding joint angle to finally switch to the upright mode.

[0135] In some embodiments, when the humanoid robot is in a working state (e.g., walking, bending over, etc.), a preset value for reaching the upright mode (at this time, since the humanoid robot is working on the ground, the upright mode can also be regarded as the standing mode) is first determined, and the humanoid robot is restored to the standard upright mode based on the preset value through RL (Reinforcement Learning).

[0136] Please see Figure 4 , Figure 4 This is a flowchart illustrating the foot contact detection process of the humanoid robot initiation method based on a policy network trained through deep learning, as described in the first embodiment of this application. Foot contact detection in upright mode includes: S105: Obtain the actual output torque of the motor corresponding to the leg joint of the humanoid robot in upright mode; S106: Calculate the torque difference between the actual output torque and the corresponding preset output torque; S107: The plantar contact force on the sole of the foot is obtained based on the torque difference; S108: Determine whether the sole of the foot is in contact with the ground based on the contact force of the sole.

[0137] Therefore, automatically detecting foot contact based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can prevent the humanoid robot from being unstable due to the user's misjudgment of foot contact with the ground by the naked eye, and thus prevent the humanoid robot from falling over due to instability.

[0138] Specifically, the humanoid robot calculates the preset output torque of the motors corresponding to its leg joints. This preset output torque represents the torque at which the robot's feet are not touching the ground. Then, in upright mode, the actual output torque of the motors corresponding to the humanoid robot's leg joints is obtained, and the torque difference is calculated. This torque difference is then mapped into a foot contact force using a preset mapping matrix. Finally, whether the foot is touching the ground is determined based on whether the foot contact force meets preset conditions. The humanoid robot's leg joints include at least one of the hip, knee, and ankle joints.

[0139] In some embodiments, the preset mapping matrix is ​​the foot Jacobian matrix, or it can be the contact constraint Jacobian matrix.

[0140] In some embodiments, the method further includes: The humanoid robot is controlled by a first-strategy network to perform standing and / or walking actions.

[0141] In some embodiments, a startup success command is sent when the humanoid robot has been standing on the ground for a preset duration after startup.

[0142] In some embodiments, if an abnormality occurs in the joints of the humanoid robot during the startup process, an abnormality prompt message is output.

[0143] Please see Figure 5 , Figure 5 This is a flowchart illustrating the humanoid robot startup method based on a policy network trained using deep learning, as described in the second embodiment of this application. The humanoid robot startup method based on a policy network trained using deep learning is applied to a user controller and includes: S201: In response to the selection operation, a first detection instruction is generated and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. S202: Receive the current posture of the humanoid robot reported by the humanoid robot; S203: Match and verify the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, including the start posture, is allowed to be triggered. The start posture is the suspended posture. S204: Send a first start command including a start posture to the humanoid robot. The first start command including a start posture is used to instruct the humanoid robot, after receiving the first start command including a start posture from the user controller, to sequentially switch the motion control mode of the humanoid robot from zero torque mode to damped mode, and then from damped mode to upright mode; and, in upright mode, perform foot contact detection, and when foot contact is detected, switch the motion control mode to the first policy network.

[0144] In this embodiment, the user controller matches the user-selected start posture with the current posture of the humanoid robot, that is, it verifies the user-selected start posture. Only when the match, that is, the verification is successful, will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture due to the user selecting the wrong start posture, which would cause damage or failure to the humanoid robot.

[0145] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, allowing the triggering of a first initiation command including an initiation gesture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After the first start command, including the start posture, is triggered, the electronic device is controlled to display the first start page. In response to the user's start trigger operation on the first start page, the first start command, including the start posture, is generated.

[0146] Therefore, after the first start command, including the start posture, is allowed to be triggered, the first start command is generated in a corresponding manner according to the settings of the electronic device, so that the humanoid robot can receive the first start command and execute the corresponding start operation.

[0147] Specifically, after the user controller allows the triggering of a first start command including a start posture, it directly generates a first start command including a start posture; or, after the user controller allows the triggering of a first start command including a start posture, the user performs a start triggering operation on the first start page displayed on the electronic device, thereby generating a first start command including a start posture through user operation.

[0148] In some embodiments, after the first startup command is generated, the electronic device displays a success page or outputs a voice message indicating that the first startup command has been successfully generated.

[0149] In some embodiments, the electronic device is a tablet computer, smartphone, smartwatch, etc.

[0150] In some embodiments, a first detection instruction is generated in response to a selection operation, including: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

[0151] Therefore, the user controller will only generate the first detection command in response to the selection operation after the user completes the selection operation of the start posture. After receiving the first detection command, the humanoid robot obtains its current posture. This restricts the humanoid robot to perform subsequent start operations only after the user completes the selection operation of the start posture, thus avoiding the situation where the humanoid robot performs the start operation by mistake before the user completes the selection operation of the start posture.

[0152] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and after sending a first activation command including an activation gesture to the humanoid robot, the method further includes: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0153] Therefore, users can perform the first operation, which brings the feet of the humanoid robot in a suspended posture close to the ground, based on the startup prompts displayed on the second startup page of the electronic device. This eliminates the need for users to carry the instruction manual at all times or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, which improves the ease of use of starting the humanoid robot in a suspended posture, saves users' learning time, and improves operational efficiency.

[0154] In some embodiments, a robot startup page is displayed on the electronic device. The user selects a startup posture, clicks the confirmation button on the page to generate a first detection command, and sends the first detection command to the humanoid robot to execute subsequent method steps.

[0155] Please see Figure 6 , Figure 6 This is a schematic diagram of the selected starting posture as a suspended hoisting posture in the first embodiment of this application. After entering the robot startup page, a selection page for choosing the starting posture of the humanoid robot will first be displayed, such as... Figure 6 As shown, the selection page displays multiple startup postures, such as the suspended hoisting posture (for...). Figure 6 (Suspension start-up), reclining posture (for) Figure 6 (The supine start) and squatting position (not in) Figure 6 As shown in the diagram, but can be added to the selectable startup postures as needed), the user selects one of multiple startup postures. In some embodiments, after the user selects one of multiple startup postures, a white border will appear around the edge of the area where the selected startup posture is located, and a checkmark symbol representing the selection will appear in the lower right corner (e.g., ...). Figure 6 (The selected suspended hoisting posture) After confirming the selected start posture, the user controller responds to the selection operation by generating posture information including the user-selected start posture, and generates the first detection command after generating the posture information.

[0156] In some embodiments, after the current posture and the starting posture of the humanoid robot are matched, the user controller allows the triggering of a first start command including the starting posture. After the user controller allows the triggering of the first start command including the starting posture, the electronic device is controlled to display a first start page, which is a page corresponding to the starting posture.

[0157] For details, please refer to Figure 7 , Figure 7 This is a schematic diagram of the first startup page in the first embodiment of this application. When the startup posture is a suspended hoisting posture, the first startup page is the startup page corresponding to the suspended hoisting posture. When the user clicks the "Start" button on the first startup page, that is, the user performs a startup trigger operation on the first startup page, the user controller responds to the user's startup trigger operation on the first startup page, generates a first startup command including the startup posture, and then sends the first startup command to the humanoid robot.

[0158] like Figure 7 As shown, when the starting posture is the suspended hoisting posture, the first startup page will display a schematic diagram of the humanoid robot in the suspended hoisting posture, and the first startup page will also display the prompt message "Click 'Start', at which time the robot indicator light will flash green".

[0159] In other embodiments, the user controller can directly generate the first start command. In this case, the information displayed on the first start page is the light effect of the indicator light when the humanoid robot is in upright mode. After the user confirms that everything is correct, he / she can click the "Start" button on the first start page to enter the second start page.

[0160] In some embodiments, after sending a first start command, including a start gesture, to the humanoid robot, the control electronics display a second start page. See also... Figure 8 , Figure 8This is a schematic diagram of the second startup page in the first embodiment of this application. The second startup page at this time corresponds to the suspended posture. The second startup page displays startup prompt information to prompt the user to perform the first operation, causing the suspending frame of the humanoid robot to lower the humanoid robot vertically, so that the soles of the humanoid robot's feet are close to the ground. Figure 8 The startup prompts are as follows: "Please slowly lower the robot until its feet touch the ground smoothly" and "Remove the back strap from the hanger". The startup prompts also include "Click 'Confirm' to complete the startup". When the user clicks the "Confirm" button on the second startup page, the user controller is notified to complete the startup of the humanoid robot. After that, the electronic device will jump to the control page so that the user can control the humanoid robot on the electronic device as needed.

[0161] In some embodiments, after the user clicks the "Confirm" button on the second startup page, the user controller generates and sends a confirmation command to the humanoid robot so that the humanoid robot confirms that the startup is complete, which facilitates the execution of subsequent actions.

[0162] Please see Figure 9 , Figure 9 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the third embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning includes: S301: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; S302: The humanoid robot responds to the first detection command sent by the user controller, obtains the current posture of the humanoid robot, and reports it to the user controller; S303: The user controller receives the current posture of the humanoid robot reported by the humanoid robot; S304: The user controller matches and verifies the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is the suspended posture. S305: The user controller sends a first start command, including the start posture, to the humanoid robot; S306: The humanoid robot receives the first start command, including the start posture, issued by the user controller; S307: The humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode; S308: The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0163] In this embodiment, the user controller matches the user-selected start posture with the current posture of the humanoid robot, that is, it verifies the user-selected start posture. Only when the match, that is, the verification is successful, will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture due to the user selecting the wrong start posture, which would cause damage or failure to the humanoid robot.

[0164] Please see Figure 10 , Figure 10 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the fourth embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning is applied to a humanoid robot and includes: S401: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the current posture of the humanoid robot is received. If they match, the first start command including the start posture is allowed to be triggered. The start posture is a lying posture. S402: Receive the first start command, including the start attitude, issued by the user controller; S403: Switch the motion control mode of the humanoid robot from zero torque mode to the second strategy network and execute the action of standing up. After the execution is completed, switch back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0165] In this embodiment, the humanoid robot receives and responds to the first detection command sent by the user controller, and then obtains the current posture of the humanoid robot. After receiving the current posture of the humanoid robot, the user controller performs a matching verification between the current posture of the humanoid robot and the start posture selected by the user. If they match, the first start command including the start posture is allowed to be triggered. After receiving the first start command, the humanoid robot begins to execute the start operation corresponding to the lying posture, so as to finally stand stably on the ground after its feet touch the ground, thus completing the start of the humanoid robot.

[0166] As mentioned earlier, the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture. Only when the match, or verification, is successful will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that is inconsistent with the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0167] Please see Figure 11 , Figure 11 This is a schematic diagram of the startup process of the humanoid robot in this application, whose current posture is a lying posture. Figure 11 (11a)-(11i) are the startup process of the humanoid robot's current posture being a lying posture, specifically switching the humanoid robot's motion control mode from zero torque mode to the second policy network and performing the action of getting up.

[0168] In some embodiments, when the humanoid robot's current posture is a lying posture, after switching the humanoid robot's motion control mode from zero torque mode to the second policy network and performing the action of getting up, the IMU obtains acceleration information and angular velocity information, and calculates the chest posture angle and hip posture angle based on the acceleration information and angular velocity information. Then, it calculates the relative posture difference between the chest posture angle and the hip posture angle. When the relative posture difference is less than a preset relative posture threshold, and the duration of the humanoid robot's current posture reaches a preset time, it is determined that the humanoid robot has completed getting up in the lying posture.

[0169] Please see Figure 12 , Figure 12 This is a flowchart illustrating the humanoid robot startup method based on a policy network trained using deep learning, as described in the fifth embodiment of this application. The humanoid robot startup method based on a policy network trained using deep learning is applied to a user controller and includes: S501: In response to the selection operation, a first detection instruction is generated and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to respond to the first detection instruction sent by the user controller, obtain the current posture of the humanoid robot and report it to the user controller. S502: Receive the current posture of the humanoid robot reported by the humanoid robot; S503: Match and verify the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, including the start posture, is allowed to be triggered. The start posture is a lying posture. S504: Send a first start command including the start posture to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the second strategy network and perform the action of getting up after receiving the first start command including the start posture from the user controller. After the action is completed, switch back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0170] In this embodiment, the user controller matches the user-selected start posture with the current posture of the humanoid robot, that is, it verifies the user-selected start posture. Only when the match, that is, the verification is successful, will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture due to the user selecting the wrong start posture, which would cause damage or failure to the humanoid robot.

[0171] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, allowing the triggering of a first initiation command including an initiation gesture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After the first start command, including the start posture, is triggered, the electronic device is controlled to display the first start page. In response to the user's start trigger operation on the first start page, the first start command, including the start posture, is generated.

[0172] Therefore, after the first start command, including the start posture, is allowed to be triggered, the first start command is generated in a corresponding manner according to the settings of the electronic device, so that the humanoid robot can receive the first start command and execute the corresponding start operation.

[0173] In some embodiments, a first detection instruction is generated in response to a selection operation, including: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

[0174] Therefore, the user controller will only generate the first detection command in response to the selection operation after the user completes the selection operation of the start posture. After receiving the first detection command, the humanoid robot obtains its current posture. This restricts the humanoid robot to perform subsequent start operations only after the user completes the selection operation of the start posture, thus avoiding the situation where the humanoid robot performs the start operation by mistake before the user completes the selection operation of the start posture.

[0175] Please see Figure 13 , Figure 13 This is a schematic diagram of the selected starting posture as a lying posture in the fifth embodiment of this application. After entering the robot startup page, a selection page for choosing the starting posture of the humanoid robot will first be displayed, such as... Figure 13 As shown, the selection page displays multiple startup postures, such as the suspended hoisting posture (for...). Figure 13 (Suspension start-up), reclining posture (for) Figure 13 (The supine start) and squatting position (not in) Figure 13 As shown in the diagram, but can be added to the selectable startup postures as needed), when a user selects one of multiple startup postures, a white border will appear around the selected posture, and a checkmark will appear in the lower right corner to indicate that it has been selected (e.g., ...). Figure 13 (The selected lying posture) After confirming the selected start posture, the user controller responds to the selection operation by generating posture information including the user-selected start posture after clicking confirm, and generates the first detection command after generating the posture information.

[0176] In some embodiments, after the current posture and the starting posture of the humanoid robot are matched, the user controller allows the triggering of a first start command including the starting posture. After the user controller allows the triggering of the first start command including the starting posture, the electronic device is controlled to display a first start page, which is a page corresponding to the starting posture.

[0177] Please see Figure 14 , Figure 14 This is a schematic diagram of the first startup page in the fifth embodiment of this application. When the startup posture is a lying posture, the first startup page is the startup page corresponding to the lying posture. When the user clicks the "Start" button on the first startup page, that is, the user performs a startup trigger operation on the first startup page. The user controller responds to the user's startup trigger operation on the first startup page, generates a first startup command including the startup posture, and then sends the first startup command to the humanoid robot.

[0178] like Figure 14As shown, when the robot is in a lying position upon startup, the first startup page displays an illustration of the humanoid robot in this position. The page also displays the messages: "Please ensure the space around the robot is open and no one is near," and "Click 'Start,' and the robot will automatically stand up; the indicator light will flash green." After the user clicks the start button, the electronic device redirects to the control page, allowing the user to control the humanoid robot as needed.

[0179] Please see Figure 15 , Figure 15 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the sixth embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning includes: S601: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; S602: The humanoid robot responds to the first detection command sent by the user controller, obtains the current posture of the humanoid robot, and reports it to the user controller; S603: The user controller receives the current posture of the humanoid robot reported by the humanoid robot; S604: The user controller performs a match verification between the current posture of the humanoid robot and the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is a lying posture. S605: The user controller sends a first start command, including the start posture, to the humanoid robot; S606: The humanoid robot receives the first start command, including the start posture, issued by the user controller; S607: The humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0180] In this embodiment, the user controller matches the user-selected start posture with the current posture of the humanoid robot, that is, it verifies the user-selected start posture. Only when the match, that is, the verification is successful, will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture due to the user selecting the wrong start posture, which would cause damage or failure to the humanoid robot.

[0181] Please see Figure 16 , Figure 16 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the seventh embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S701: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the current posture of the humanoid robot is received. If they match, the first start command including the start posture is allowed to be triggered. The start posture is a squatting posture. S702: Receives the first start command, including the start attitude, issued by the user controller; S703: Switch the motion control mode of the humanoid robot from zero torque mode to the third strategy network and execute the action of standing up. After the execution is completed, switch back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0182] In this embodiment, the humanoid robot receives and responds to the first detection command sent by the user controller, and then obtains the current posture of the humanoid robot. After receiving the current posture of the humanoid robot, the user controller performs a matching verification between the current posture of the humanoid robot and the start posture selected by the user. If they match, the first start command including the start posture is allowed to be triggered. After receiving the first start command, the humanoid robot begins to execute the start operation corresponding to the squatting posture, so as to finally stand stably on the ground after its feet touch the ground, thereby completing the start of the humanoid robot.

[0183] As mentioned earlier, the user controller matches the user-selected start posture with the humanoid robot's current posture, that is, it verifies the user-selected start posture. Only when the match, or verification, is successful will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that is inconsistent with the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0184] In some embodiments, when the humanoid robot's current posture is a squatting posture, after switching the humanoid robot's motion control mode from zero torque mode to the third policy network and performing the action of standing up, the IMU obtains acceleration information and angular velocity information, and calculates the chest posture angle and hip posture angle based on the acceleration information and angular velocity information. Then, it calculates the relative posture difference between the chest posture angle and the hip posture angle. When the relative posture difference is less than a preset relative posture threshold, and the duration of the humanoid robot's current posture reaches a preset time, it is determined that the humanoid robot has completed standing up in the squatting posture.

[0185] Please see Figure 17 , Figure 17 This is a schematic diagram of the humanoid robot in this application, whose current posture is a squatting posture and the startup process inside the box. Figure 17 (17a)-(17f) are respectively the startup process of the humanoid robot in the box when the current posture is a squatting posture. Specifically, the robot switches the motion control mode of the humanoid robot from the zero torque mode to the third policy network and performs the action of standing up in order to finally achieve the operation of leaving the box and standing up.

[0186] Please see Figure 18 , Figure 18 This is a schematic diagram of the humanoid robot in this application, whose current posture is a squatting posture and the startup process is on flat ground. Figure 18 (18a)-(18i) are, respectively, the startup process of the humanoid robot when its current posture is a squatting posture and it is on flat ground. Specifically, the robot first adjusts from a squatting posture to a lying posture on flat ground, and then performs a startup operation with the current posture as a lying posture to finally complete the startup.

[0187] Please see Figure 19 , Figure 19 This is a flowchart illustrating the humanoid robot startup method based on a policy network trained using deep learning, as described in the eighth embodiment of this application. The humanoid robot startup method based on a policy network trained using deep learning is applied to a user controller and includes: S801: In response to the selection operation, a first detection instruction is generated and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to respond to the first detection instruction sent by the user controller, obtain the current posture of the humanoid robot and report it to the user controller. S802: Receives the current posture of the humanoid robot reported by the humanoid robot; S803: Match and verify the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, including the start posture, is allowed to be triggered. The start posture is a squatting posture. S804: Send a first start command including the start posture to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the third strategy network and perform the action of getting up after receiving the first start command including the start posture from the user controller. After the action is completed, switch back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0188] In this embodiment, the user controller matches the user-selected start posture with the current posture of the humanoid robot, that is, it verifies the user-selected start posture. Only when the match, that is, the verification is successful, will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture due to the user selecting the wrong start posture, which would cause damage or failure to the humanoid robot.

[0189] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, allowing the triggering of a first initiation command including an initiation gesture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After the first start command, including the start posture, is triggered, the electronic device is controlled to display the first start page. In response to the user's start trigger operation on the first start page, the first start command, including the start posture, is generated.

[0190] Therefore, after the first start command, including the start posture, is allowed to be triggered, the first start command is generated in a corresponding manner according to the settings of the electronic device, so that the humanoid robot can receive the first start command and execute the corresponding start operation.

[0191] In some embodiments, a first detection instruction is generated in response to a selection operation, including: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

[0192] Therefore, the user controller will only generate the first detection command in response to the selection operation after the user completes the selection operation of the start posture. After receiving the first detection command, the humanoid robot obtains its current posture. This restricts the humanoid robot to perform subsequent start operations only after the user completes the selection operation of the start posture, thus avoiding the situation where the humanoid robot performs the start operation by mistake before the user completes the selection operation of the start posture.

[0193] When the starting posture is a squatting posture, the first startup page and trigger operation displayed by the electronic device are similar to those when the starting posture is a lying posture, except that the diagram of the lying posture on the first startup page is replaced with a diagram of the squatting posture.

[0194] Please see Figure 20 , Figure 20 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the ninth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning includes: S901: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; S902: The humanoid robot responds to the first detection command sent by the user controller, obtains the current posture of the humanoid robot, and reports it to the user controller; S903: The user controller receives the current posture of the humanoid robot reported by the humanoid robot; S904: The user controller matches and verifies the current posture of the humanoid robot with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is a squatting posture. S905: The user controller sends a first start command, including the start posture, to the humanoid robot; S906: The humanoid robot receives the first start command, including the start posture, issued by the user controller; S907: The humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0195] In this embodiment, the user controller matches the user-selected start posture with the current posture of the humanoid robot, that is, it verifies the user-selected start posture. Only when the match, that is, the verification is successful, will the humanoid robot execute the start operation corresponding to the start posture. In this way, the verification operation can avoid the situation where the humanoid robot executes a start operation that does not match the current posture due to the user selecting the wrong start posture, which would cause damage or failure to the humanoid robot.

[0196] In this application, regardless of whether the humanoid robot's current posture is a suspended posture, a lying posture, or a squatting posture, the same set of matching and verification logic between the humanoid robot's current posture and the user-selected starting posture can be applied. That is, the matching and verification logic is not affected by the humanoid robot's current posture. Therefore, the same set of matching and verification logic can be adapted to multiple starting postures of the humanoid robot, so the matching and verification logic in this application is applicable.

[0197] Please see Figure 21 , Figure 21 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the tenth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S1001: Receive the first start command including the start posture selected by the user, the start posture is the suspended hoisting posture; S1002: Obtain the current posture of the humanoid robot; S1003: Perform a matching verification between the current posture and the starting posture of the humanoid robot; S1004: If matched, sequentially switch the motion control mode of the humanoid robot from zero torque mode to damped mode, and then switch from damped mode to upright mode. S1005: Perform foot contact detection in upright mode. When foot contact is detected, switch the motion control mode to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0198] In this embodiment, the humanoid robot receives and responds to a first start command sent by the user controller, which includes the start posture selected by the user (the start posture at this time is the suspended posture). After receiving the first start command, the humanoid robot obtains its current posture. Then, the humanoid robot performs a matching verification between its current posture and the start posture selected by the user. If they match, the robot begins to execute the start operation corresponding to the suspended posture, so as to finally stand stably on the ground after its feet touch the ground and complete the start.

[0199] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0200] Please see Figure 22 , Figure 22 This is a flowchart illustrating the foot contact detection process of the humanoid robot initiation method based on a policy network trained through deep learning, as described in the tenth embodiment of this application. Foot contact detection in upright mode includes: S1006: Obtain the actual output torque of the motor corresponding to the leg joint of the humanoid robot in upright mode; S1007: Calculate the torque difference between the actual output torque and the corresponding preset output torque; S1008: The plantar contact force on the sole of the foot is obtained based on the torque difference; S1009: Determine whether the sole of the foot is in contact with the ground based on the contact force of the sole.

[0201] Therefore, automatically detecting foot contact based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can prevent the humanoid robot from being unstable due to the user's misjudgment of foot contact with the ground by the naked eye, and thus prevent the humanoid robot from falling over due to instability.

[0202] In some embodiments, the humanoid robot initiation method based on a policy network trained by deep learning is applied to the user controller, including: In response to a selection operation, a first start command including the user-selected start posture is generated and sent to the humanoid robot. The start posture is a suspended posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain the current posture of the humanoid robot after receiving the first start command including the user-selected start posture. The current posture of the humanoid robot and the start posture are matched and verified. If they match, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. In the upright mode, foot contact detection is performed. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0203] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0204] The page for selecting the startup posture is as follows: Figure 6 As shown, in some embodiments, after the user performs a selection operation, and the starting posture matches the current posture of the humanoid robot, the humanoid robot sends a jump command to the user controller, causing the electronic device including the user controller to display the corresponding first startup page. At this time, the first startup page displays reminder information related to the upright mode to remind the user that the current motion control mode of the humanoid robot is in the upright mode, and enters the second startup page after the user performs a confirmation operation.

[0205] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes, after sending a first activation command, including a user-selected activation posture, to the humanoid robot: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0206] Therefore, users can perform the first operation, which allows the humanoid robot's feet to touch the ground, based on the startup prompts displayed on the second startup page on the electronic device. This eliminates the need for users to carry an instruction manual or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, improving the ease of use of starting the humanoid robot in a suspended posture, saving users' learning time, and increasing operational efficiency.

[0207] Please see Figure 23 , Figure 23 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained using deep learning, as described in the eleventh embodiment of this application. The humanoid robot activation method based on a policy network trained using deep learning includes: S1101: The user controller generates a first start command including the start posture selected by the user in response to the selection operation, and sends the first start command including the start posture selected by the user to the humanoid robot. The start posture is a suspended posture. S1102: The humanoid robot receives a first start command including the start posture selected by the user; S1103: The humanoid robot acquires its current posture; S1104: The humanoid robot performs a matching and verification between its current posture and its starting posture; S1105: If matched, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. S1106: The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0208] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0209] Please see Figure 24 , Figure 24 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twelfth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S1201: Receive a first startup command including the user-selected startup posture, wherein the startup posture is a lying posture; S1202: Obtain the current posture of the humanoid robot; S1203: Perform a matching verification between the current posture and the starting posture of the humanoid robot; S1204: If matched, switch the motion control mode of the humanoid robot from zero torque mode to the second strategy network and execute the action of getting up. After execution, switch back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0210] In this embodiment, the humanoid robot receives and responds to a first start command sent by the user controller, which includes the start posture selected by the user (the start posture at this time is a lying posture). After receiving the first start command, the humanoid robot obtains its current posture. Then, the humanoid robot performs a matching verification between its current posture and the start posture selected by the user. If they match, the robot begins to execute the start operation corresponding to the lying posture, so as to finally stand stably on the ground after its feet touch the ground and complete the start.

[0211] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0212] In some embodiments, the humanoid robot initiation method based on a policy network trained by deep learning is applied to the user controller, including: In response to the selection operation, a first start command including the user-selected start posture is generated and sent to the humanoid robot. The start posture is a lying posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain the current posture of the humanoid robot after receiving the first start command including the user-selected start posture. The current posture of the humanoid robot and the start posture are matched and verified. If they match, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the action of getting up is executed. After the execution is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0213] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0214] The page for selecting the startup posture is as follows: Figure 13As shown, in some embodiments, after the user performs a selection operation, and after the startup posture matches the current posture of the humanoid robot, the humanoid robot sends a jump command to the user controller, causing the electronic device including the user controller to display as shown. Figure 14 The first startup page shown.

[0215] Please see Figure 25 , Figure 25 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the thirteenth embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning includes: S1301: The user controller generates a first start command including the start posture selected by the user in response to the selection operation, and sends the first start command including the start posture selected by the user to the humanoid robot. The start posture is a lying posture. S1302: The humanoid robot receives a first start command including the start posture selected by the user; S1303: The humanoid robot acquires its current posture; S1304: The humanoid robot performs a matching and verification between its current posture and its starting posture; S1305: If matched, the humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs the action of standing up. After execution, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0216] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0217] Please see Figure 26 , Figure 26 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the fourteenth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S1401: Receive a first start command including the start posture selected by the user, wherein the start posture is a squatting posture; S1402: Obtain the current posture of the humanoid robot; S1403: Perform a matching verification between the current posture and the starting posture of the humanoid robot; S1404: If matched, switch the motion control mode of the humanoid robot from zero torque mode to the third strategy network and execute the action of getting up. After execution, switch back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0218] In this embodiment, the humanoid robot receives and responds to a first start command sent by the user controller, which includes the start posture selected by the user (the start posture at this time is a squatting posture). After receiving the first start command, the humanoid robot obtains its current posture. Then, the humanoid robot performs a matching verification between its current posture and the start posture selected by the user. If they match, the robot begins to execute the start operation corresponding to the squatting posture, so as to finally stand stably on the ground after its feet touch the ground and complete the start.

[0219] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0220] In some embodiments, the humanoid robot initiation method based on a policy network trained by deep learning is applied to the user controller, including: In response to the selection operation, a first start command including the user-selected start posture is generated and sent to the humanoid robot. The start posture is a squatting posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain the current posture of the humanoid robot after receiving the first start command including the user-selected start posture. The current posture of the humanoid robot and the start posture are matched and verified. If they match, the motion control mode of the humanoid robot is switched from zero torque mode to the third policy network and the standing action is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0221] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0222] When the starting posture is a squatting posture, the first startup page and trigger operation displayed by the electronic device are similar to those of the first startup page and trigger operation when the starting posture is a lying posture, except that the diagram of the lying posture is replaced with the diagram of the squatting posture.

[0223] Please see Figure 27 , Figure 27 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the fifteenth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning includes: S1501: The user controller generates a first start command including the start posture selected by the user in response to the selection operation, and sends the first start command including the start posture selected by the user to the humanoid robot. The start posture is a squatting posture. S1502: The humanoid robot receives a first start command including the start posture selected by the user; S1503: The humanoid robot acquires its current posture; S1504: The humanoid robot performs a matching and verification between its current posture and its starting posture; S1505: If matched, the humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs the action of standing up. After execution, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0224] In this embodiment, the humanoid robot matches the user-selected starting posture with the humanoid robot's current posture, that is, it verifies the user-selected starting posture, and only executes the starting operation corresponding to the starting posture when the match, that is, the verification, is successful. In this way, the verification operation can avoid the situation where the humanoid robot executes a starting operation that does not match the current posture after the user selects the wrong starting posture, which would cause damage or failure to the humanoid robot.

[0225] Please see Figure 28 , Figure 28 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the sixteenth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S1601: In response to the second start command, detect the current posture of the humanoid robot; S1602: When the current posture of the humanoid robot is detected to be a suspended posture, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. S1603: Perform foot contact detection in upright mode. When foot contact is detected, switch the motion control mode to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0226] In this embodiment, the humanoid robot responds to the second start command. The second start command is a single start command and does not include the start posture. After receiving the second start command, the humanoid robot detects its current posture. When it detects that the current posture of the humanoid robot is a suspended posture, it executes the start operation corresponding to the suspended posture, so as to finally stand stably on the ground after its feet touch the ground and complete the start.

[0227] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0228] In some embodiments, the second start command may be issued by the user controller or may be triggered directly on the humanoid robot, such as by pressing a button on the humanoid robot.

[0229] Please see Figure 29 , Figure 29 This is a schematic flowchart illustrating the foot contact detection process of the humanoid robot initiation method based on a policy network trained through deep learning, as described in the sixteenth embodiment of this application. Foot contact detection in upright mode includes: S1604: Obtain the actual output torque of the motor corresponding to the leg joint of the humanoid robot in upright mode; S1605: Calculate the torque difference between the actual output torque and the corresponding preset output torque; S1606: The plantar contact force on the sole of the foot is obtained based on the torque difference; S1607: Determine whether the sole of the foot is in contact with the ground based on the contact force of the sole.

[0230] Therefore, automatically detecting foot contact with the ground based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can avoid the problem of the humanoid robot being unstable due to the user's misjudgment of foot contact with the ground by relying solely on the naked eye.

[0231] In some embodiments, the humanoid robot initiation method based on a policy network trained by deep learning is applied to the user controller, including: A second start command is generated in response to a trigger operation and sent to the humanoid robot. The second start command instructs the humanoid robot to detect its current posture. When the current posture of the humanoid robot is detected to be a suspended posture, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. In the upright mode, foot contact detection is performed. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0232] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0233] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes, after sending a second start command to the humanoid robot: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0234] Therefore, users can perform the first operation, which allows the humanoid robot's feet to touch the ground, based on the startup prompts displayed on the second startup page on the electronic device. This eliminates the need for users to carry an instruction manual or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, improving the ease of use of starting the humanoid robot in a suspended posture, saving users' learning time, and increasing operational efficiency.

[0235] In some embodiments, when the humanoid robot detects that its current posture is a suspended, hoisted posture, it can issue a start prompt command to the user controller, causing the electronic device including the user controller to display something like... Figure 8 The second startup page is shown.

[0236] Please see Figure 30 , Figure 30 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the seventeenth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning includes: S1701: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; S1702: The humanoid robot responds to the second start command and detects the current posture of the humanoid robot; S1703: When the current posture of the humanoid robot is detected to be a suspended posture, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. S1704: The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0237] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0238] Please see Figure 31 , Figure 31 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the eighteenth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S1801: In response to the second start command, detect the current posture of the humanoid robot; S1802: When the humanoid robot's current posture is a lying posture, switch the humanoid robot's motion control mode from zero torque mode to the second policy network and execute the action of getting up. After execution, switch back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0239] In this embodiment, the humanoid robot responds to the second start command. The second start command is a single start command and does not include the start posture. After receiving the second start command, the humanoid robot detects its current posture. When it detects that the current posture of the humanoid robot is a lying posture, it executes the start operation corresponding to the lying posture so as to finally stand stably on the ground and complete the start.

[0240] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0241] In some embodiments, the humanoid robot initiation method based on a policy network trained by deep learning is applied to the user controller, including: A second start command is generated in response to the trigger operation and sent to the humanoid robot. The second start command is used to instruct the humanoid robot to detect the current posture of the humanoid robot in response to the second start command. When the current posture of the humanoid robot is a lying posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the second policy network and the action of getting up is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0242] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0243] Please see Figure 32 , Figure 32 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the nineteenth embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning includes: S1901: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; S1902: The humanoid robot responds to the second start command and detects the current posture of the humanoid robot; S1903: When the humanoid robot's current posture is a lying posture, the humanoid robot switches its motion control mode from zero torque mode to the second policy network and performs the action of getting up. After the action is completed, it switches back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0244] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0245] Please see Figure 33 , Figure 33 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twentieth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S2001: In response to the second start command, detect the current posture of the humanoid robot; S2002: When the current posture of the humanoid robot is a squatting posture, switch the motion control mode of the humanoid robot from the zero torque mode to the third policy network and execute the action of standing up. After the execution is completed, switch back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0246] In this embodiment, the humanoid robot responds to the second start command. The second start command is a single start command and does not include the start posture. After receiving the second start command, the humanoid robot detects its current posture. When it detects that the current posture of the humanoid robot is a squatting posture, it executes the start operation corresponding to the squatting posture, so as to finally stand stably on the ground and complete the start.

[0247] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0248] In some embodiments, the humanoid robot initiation method based on a policy network trained by deep learning is applied to the user controller, including: A second start command is generated in response to the trigger operation and sent to the humanoid robot. The second start command is used to instruct the humanoid robot to detect the current posture of the humanoid robot in response to the second start command. When the current posture of the humanoid robot is a squatting posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the third policy network and the action of standing up is executed. After the execution is completed, it is switched back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0249] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0250] Please see Figure 34 , Figure 34 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twenty-first embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning includes: S2101: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; S2102: The humanoid robot responds to the second start command and detects the current posture of the humanoid robot; S2103: When the humanoid robot's current posture is a squatting posture, the humanoid robot switches its motion control mode from zero torque mode to the third policy network and performs the action of standing up. After the action is completed, it switches back to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0251] In this embodiment, the humanoid robot does not need to select a starting posture. After receiving the second start command, it directly detects the current posture and executes the corresponding start operation based on the detected current posture. This avoids damage and malfunction of the humanoid robot caused by the user selecting the wrong starting posture and the humanoid robot executing a start operation that is inconsistent with the current posture.

[0252] Please see Figure 35 , Figure 35 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twenty-second embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S2201: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is a suspended posture. S2202: In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. S2203: Perform foot contact detection in upright mode. When foot contact is detected, switch the motion control mode to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0253] In this embodiment, the humanoid robot responds to the second detection command issued by the user controller and then obtains the current posture of the humanoid robot. After receiving the current posture of the humanoid robot (the current posture of the humanoid robot here is the suspended posture), the user controller confirms the current posture of the humanoid robot and, after confirmation, executes a trigger operation to generate a confirmation start command. The humanoid robot receives and responds to the confirmation start command and executes the start operation corresponding to the suspended posture, so as to finally stand stably on the ground and complete the start.

[0254] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0255] Please see Figure 36 , Figure 36 This is a schematic flowchart illustrating the foot contact detection process of the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twenty-second embodiment of this application. Foot contact detection in upright mode includes: S2204: Obtain the actual output torque of the motor corresponding to the leg joint of the humanoid robot in upright mode; S2205: Calculate the torque difference between the actual output torque and the corresponding preset output torque; S2206: The plantar contact force on the sole of the foot is obtained based on the torque difference; S2207: Determine whether the sole of the foot is in contact with the ground based on the contact force of the sole.

[0256] Therefore, automatically detecting foot contact with the ground based on torque difference in upright mode can improve the stability of the humanoid robot when it is put down and stands on the ground after being suspended in the air. This can avoid the problem of the humanoid robot being unstable due to the user's misjudgment of foot contact with the ground by relying solely on the naked eye.

[0257] Please see Figure 37 , Figure 37 This is a flowchart illustrating the humanoid robot startup method based on a policy network trained using deep learning, as described in the twenty-third embodiment of this application. The humanoid robot startup method based on a policy network trained using deep learning is applied to a user controller, and the method includes: S2301: In response to the first triggering operation, a second detection instruction is generated and sent to the humanoid robot. The second detection instruction is used to instruct the humanoid robot to acquire the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a suspended posture. S2302: Receive the current posture of the humanoid robot sent by the humanoid robot; S2303: In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to sequentially switch the motion control mode of the humanoid robot from zero torque mode to damped mode, and then from damped mode to upright mode in response to the confirmation start command sent by the user controller; and, in upright mode, foot contact detection is performed, and when foot contact is detected, the motion control mode is switched to the first strategy network; The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0258] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0259] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes, after sending a confirmation start command to the humanoid robot: The control electronic device displays a second startup page, which displays startup prompts for the user to perform a first operation, causing the gantry that lifts the humanoid robot to descend vertically, bringing the soles of the robot's feet close to the ground.

[0260] Therefore, users can perform the first operation, which allows the humanoid robot's feet to touch the ground, based on the startup prompts displayed on the second startup page on the electronic device. This eliminates the need for users to carry an instruction manual or memorize the specific steps of the first operation. Instead, users can simply follow the startup prompts to perform the first operation, improving the ease of use of starting the humanoid robot in a suspended posture, saving users' learning time, and increasing operational efficiency.

[0261] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: After receiving the current posture of the humanoid robot from the humanoid robot, the control electronic device displays the first startup page and generates a confirmation startup command in response to the user's startup trigger operation on the first startup page.

[0262] Therefore, a confirmation start command will only be generated when the user performs a start trigger operation on the first start page, so that the humanoid robot can perform the corresponding start operation. This allows the humanoid robot to avoid damage or malfunction caused by the user selecting the wrong start posture and the robot directly performing a start operation that does not match the current posture.

[0263] In this embodiment, the first startup page can display the current posture of the humanoid robot and a confirmation button. When the user confirms that the current posture of the humanoid robot is correct, they can click the confirmation button to execute the startup trigger operation and generate a confirmation startup command.

[0264] Please see Figure 38 , Figure 38 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained using deep learning, as described in the twenty-fourth embodiment of this application. The humanoid robot initiation method based on a policy network trained using deep learning includes: S2401: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; S2402: The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a suspended posture. S2403: The user controller receives the current posture of the humanoid robot sent by the humanoid robot; S2404: The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; S2405: In response to the confirmation start command sent by the user controller, the humanoid robot sequentially switches the motion control mode of the humanoid robot from zero torque mode to damped mode, and then from damped mode to upright mode; S2406: The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

[0265] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0266] Please see Figure 39 , Figure 39 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twenty-fifth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S2501: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is a lying posture. S2502: In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the standing action is executed. After the execution is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0267] In this embodiment, the humanoid robot responds to the second detection command issued by the user controller and then obtains the current posture of the humanoid robot. After receiving the current posture of the humanoid robot (the current posture of the humanoid robot here is a lying posture), the user controller confirms the current posture of the humanoid robot and, after confirmation, executes a trigger operation to generate a confirmation start command. The humanoid robot receives and responds to the confirmation start command and executes the start operation corresponding to the lying posture, so as to finally stand stably on the ground and complete the start.

[0268] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0269] Please see Figure 40 , Figure 40 This is a flowchart illustrating the humanoid robot startup method based on a policy network trained using deep learning, as described in the twenty-sixth embodiment of this application. The humanoid robot startup method based on a policy network trained using deep learning is applied to a user controller, and the method includes: S2601: In response to the first triggering operation, a second detection instruction is generated and sent to the humanoid robot. The second detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a lying posture. S2602: Receive the current posture of the humanoid robot sent by the humanoid robot; S2603: In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the second strategy network and perform the action of standing up in response to the confirmation start command sent by the user controller. After the execution is completed, it switches to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0270] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0271] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: After receiving the current posture of the humanoid robot from the humanoid robot, the control electronic device displays the first startup page and generates a confirmation startup command in response to the user's startup trigger operation on the first startup page.

[0272] Therefore, a confirmation start command will only be generated when the user performs a start trigger operation on the first start page, so that the humanoid robot can perform the corresponding start operation. This allows the humanoid robot to avoid damage or malfunction caused by the user selecting the wrong start posture and the robot directly performing a start operation that does not match the current posture.

[0273] Please see Figure 41 , Figure 41 This is a flowchart illustrating the humanoid robot activation method based on a policy network trained using deep learning, as described in the twenty-seventh embodiment of this application. The humanoid robot activation method based on a policy network trained using deep learning includes: S2701: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; S2702: The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a lying posture. S2703: The user controller receives the current posture of the humanoid robot sent by the humanoid robot; S2704: The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; S2705: In response to the confirmation start command sent by the user controller, the humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

[0274] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0275] Please see Figure 42 , Figure 42 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the twenty-eighth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning is applied to a humanoid robot and includes: S2801: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is obtained and reported to the user controller. The current posture of the humanoid robot is a squatting posture. S2802: In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the standing action is executed. After the execution is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0276] In this embodiment, the humanoid robot responds to the second detection command issued by the user controller and then obtains the current posture of the humanoid robot. After receiving the current posture of the humanoid robot (the current posture of the humanoid robot in this case is a squatting posture), the user controller confirms the current posture of the humanoid robot and, after confirmation, executes a trigger operation to generate a confirmation start command. The humanoid robot receives and responds to the confirmation start command and executes the start operation corresponding to the squatting posture, so as to finally stand stably on the ground and complete the start.

[0277] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0278] Please see Figure 43 , Figure 43 This is a flowchart illustrating the humanoid robot startup method based on a policy network trained using deep learning, as described in the twenty-ninth embodiment of this application. The humanoid robot startup method based on a policy network trained using deep learning is applied to a user controller and includes: S2901: In response to the first triggering operation, a second detection instruction is generated and sent to the humanoid robot. The second detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a squatting posture. S2902: Receive the current posture of the humanoid robot sent by the humanoid robot; S2903: In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the third strategy network and perform the action of getting up in response to the confirmation start command sent by the user controller. After the execution is completed, it switches to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0279] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0280] In some embodiments, the user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: After receiving the current posture of the humanoid robot from the humanoid robot, the control electronic device displays the first startup page and generates a confirmation startup command in response to the user's startup trigger operation on the first startup page.

[0281] Therefore, a confirmation start command will only be generated when the user performs a start trigger operation on the first start page, so that the humanoid robot can perform the corresponding start operation. This allows the humanoid robot to avoid damage or malfunction caused by the user selecting the wrong start posture and the robot directly performing a start operation that does not match the current posture.

[0282] Please see Figure 44 , Figure 44 This is a flowchart illustrating the humanoid robot initiation method based on a policy network trained through deep learning, as described in the thirtieth embodiment of this application. The humanoid robot initiation method based on a policy network trained through deep learning includes: S3001: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; S3002: The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a squatting posture. S3003: The user controller receives the current posture of the humanoid robot sent by the humanoid robot; S3004: The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; S3005: In response to the confirmation start command sent by the user controller, the humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

[0283] In this embodiment, the humanoid robot does not need to select a start posture. After receiving the second detection command, the humanoid robot directly obtains its current posture and reports it to the user controller. The user determines whether the current posture of the humanoid robot matches the actual current posture. After the match is confirmed and the user verification is successful, a confirmation start command is generated to enable the humanoid robot to execute the corresponding start operation. In this way, the verification operation can avoid the situation where the humanoid robot directly executes a start operation that does not match the current posture after the user selects the wrong start posture, which would cause damage or failure to the humanoid robot.

[0284] Please see Figure 45 , Figure 45 This is a schematic diagram of a humanoid robot 31 according to an embodiment of this application. The humanoid robot 31 includes a first processor 311 and a first memory 312. The first memory 312 is connected to the first processor 311 and stores a computer program. The first processor 311 executes the humanoid robot startup method based on a policy network trained by deep learning, as described above.

[0285] Please see Figure 46 , Figure 46 This is a schematic diagram of an electronic device 32 according to an embodiment of this application. The electronic device 32 includes a second processor 321 and a second memory 322. The second memory 322 is connected to the second processor 321 and stores a computer program. The second processor 321 executes the humanoid robot initiation method based on a policy network trained by deep learning and applied to a user controller as described above.

[0286] This application provides a computer-readable storage medium storing a computer program, which is invoked by a processor to execute the humanoid robot initiation method based on a policy network trained by deep learning as described above.

[0287] Finally, the above preferred embodiments are only used to illustrate the technical solutions of this application and are not restrictive. Although this application has been described in detail, those skilled in the art should understand that changes in form and detail can be made without departing from the scope defined by the claims of this application. The dimensions in the drawings are not related to the specific physical object, and the physical object dimensions can be arbitrarily changed.

Claims

1. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is acquired and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the user controller receives the current posture of the humanoid robot. If they match, the first start command including the start posture is allowed to be triggered. The start posture is the suspended hoisting posture. Receive a first startup command, including a startup posture, issued by the user controller; The motion control mode of the humanoid robot is switched sequentially from zero torque mode to damped mode, and then from damped mode to upright mode; Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

2. The startup method according to claim 1, characterized in that, The foot contact detection in upright mode includes: In the upright mode, the actual output torque of the motor corresponding to the leg joint of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Based on the aforementioned foot contact force, determine whether the foot is in contact with the ground.

3. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that: include: A first detection instruction is generated in response to a selection operation and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. Receive the current posture of the humanoid robot reported by the humanoid robot; The current posture of the humanoid robot is matched and verified with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is the suspended posture. The first start command, including the start posture, is sent to the humanoid robot. The first start command, including the start posture, is used to instruct the humanoid robot, after receiving the first start command, including the start posture, from the user controller, to sequentially switch the motion control mode of the humanoid robot from zero torque mode to damped mode, and then from damped mode to upright mode; and, in upright mode, to perform foot contact detection, and when foot contact is detected, to switch the motion control mode to the first strategy network.

4. The startup method according to claim 3, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot, and the first start command that allows triggering includes a start posture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After a first startup command including a startup posture is triggered, the electronic device is controlled to display a first startup page. In response to the user's startup trigger operation on the first startup page, a first startup command including a startup posture is generated.

5. The startup method according to claim 3, characterized in that, The generation of the first detection instruction in response to the selection operation includes: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

6. The startup method according to claim 3, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot. After sending the first start command, including the start posture, to the humanoid robot, the method further includes: The electronic device is controlled to display a second startup page, which displays startup prompt information. The startup prompt information is used to prompt the user to perform a first operation, so that the gantry that lifts the humanoid robot lowers the humanoid robot vertically, so that the soles of the humanoid robot's feet are close to the ground.

7. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; In response to a first detection command sent by the user controller, the humanoid robot obtains its current posture and reports it to the user controller. The user controller receives the current posture of the humanoid robot reported by the humanoid robot; The user controller performs a matching and verification between the current posture of the humanoid robot and the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is a suspended posture. The user controller sends the first start command, including the start posture, to the humanoid robot; The humanoid robot receives a first start command, including a start posture, issued by the user controller; The humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

8. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is acquired and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the user controller receives the current posture of the humanoid robot. If they match, the first start command including the start posture is allowed to be triggered. The start posture is a lying posture. Receive a first startup command, including a startup posture, issued by the user controller; The motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot performs the action of standing up. After the action is completed, it is switched back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

9. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: A first detection instruction is generated in response to a selection operation and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. Receive the current posture of the humanoid robot reported by the humanoid robot; The current posture of the humanoid robot is matched and verified with the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is a lying posture. Send the first start command including the start posture to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the second strategy network and perform the action of getting up after receiving the first start command including the start posture from the user controller. After the action is completed, switch back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

10. The startup method according to claim 9, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot, and the first start command that allows triggering includes a start posture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After a first startup command including a startup posture is triggered, the electronic device is controlled to display a first startup page. In response to the user's startup trigger operation on the first startup page, a first startup command including a startup posture is generated.

11. The startup method according to claim 9, characterized in that, The generation of the first detection instruction in response to the selection operation includes: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

12. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; In response to a first detection command sent by the user controller, the humanoid robot obtains its current posture and reports it to the user controller. The user controller receives the current posture of the humanoid robot reported by the humanoid robot; The user controller performs a matching and verification between the current posture of the humanoid robot and the start posture selected by the user. If they match, the first start command, which includes the start posture, is allowed to be triggered. The start posture is a lying posture. The user controller sends the first start command, including the start posture, to the humanoid robot; The humanoid robot receives a first start command, including a start posture, issued by the user controller; The humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs a standing action. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

13. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the first detection command sent by the user controller, the current posture of the humanoid robot is acquired and reported to the user controller. The current posture of the humanoid robot is used by the user controller to match and verify the current posture of the humanoid robot with the start posture selected by the user when the user controller receives the current posture of the humanoid robot. If they match, the first start command including the start posture is allowed to be triggered. The start posture is a squatting posture. Receive a first startup command, including a startup posture, issued by the user controller; The motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the robot performs the action of standing up. After the action is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

14. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: A first detection instruction is generated in response to a selection operation and sent to the humanoid robot. The first detection instruction is used to instruct the humanoid robot to obtain the current posture of the humanoid robot and report it to the user controller in response to the first detection instruction sent by the user controller. Receive the current posture of the humanoid robot reported by the humanoid robot; The current posture of the humanoid robot is matched and verified with the starting posture selected by the user. If they match, the first starting command, which includes the starting posture, is allowed to be triggered. The starting posture is a squatting posture. Send the first start command including the start posture to the humanoid robot. The first start command including the start posture is used to instruct the humanoid robot to switch the motion control mode of the humanoid robot from the zero torque mode to the third strategy network and perform the action of getting up after receiving the first start command including the start posture from the user controller. After the action is completed, switch back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

15. The startup method according to claim 14, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot, and the first start command that allows triggering includes a start posture, comprising: After triggering the first start command including the start attitude, a first start command including the start attitude is generated; or, After a first startup command including a startup posture is triggered, the electronic device is controlled to display a first startup page. In response to the user's startup trigger operation on the first startup page, a first startup command including a startup posture is generated.

16. The startup method according to claim 14, characterized in that, The generation of the first detection instruction in response to the selection operation includes: In response to a selection operation, attitude information including the user-selected start posture is generated, and a first detection command is generated after the attitude information is generated.

17. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a first detection command in response to the selection operation and sends the first detection command to the humanoid robot; In response to a first detection command sent by the user controller, the humanoid robot obtains its current posture and reports it to the user controller. The user controller receives the current posture of the humanoid robot reported by the humanoid robot; The user controller performs a matching and verification between the current posture of the humanoid robot and the starting posture selected by the user. If they match, the first starting command, which includes the starting posture, is allowed to be triggered. The starting posture is a squatting posture. The user controller sends the first start command, including the start posture, to the humanoid robot; The humanoid robot receives a first start command, including a start posture, issued by the user controller; The humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs a standing action. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

18. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: Receive a first start command including a start posture selected by the user, wherein the start posture is a suspended hoisting posture; Obtain the current pose of the humanoid robot; The current posture and the starting posture of the humanoid robot are matched and verified. If a match is found, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

19. The startup method according to claim 18, characterized in that, The foot contact detection in upright mode includes: In the upright mode, the actual output torque of the motor corresponding to the leg joint of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Based on the aforementioned foot contact force, determine whether the foot is in contact with the ground.

20. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that... include: In response to a selection operation, a first start command including a user-selected start posture is generated and sent to the humanoid robot. The start posture is a suspended posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain its current posture after receiving the first start command including the user-selected start posture, and to perform a matching verification between the current posture and the start posture. If a match is found, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. In the upright mode, foot contact detection is performed. When foot contact is detected, the motion control mode is switched to the first policy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

21. The startup method according to claim 20, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot. After sending the first start command, including the user-selected start posture, to the humanoid robot, the method further includes: The electronic device is controlled to display a second startup page, which displays startup prompt information. The startup prompt information is used to prompt the user to perform a first operation, so that the gantry that lifts the humanoid robot lowers the humanoid robot vertically, so that the soles of the humanoid robot's feet are close to the ground.

22. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: In response to a selection operation, the user controller generates a first start command including the start posture selected by the user, and sends the first start command including the start posture selected by the user to the humanoid robot, wherein the start posture is a suspended posture. The humanoid robot receives a first start command including the start posture selected by the user; The humanoid robot acquires its current posture; The humanoid robot performs a matching and verification between its current posture and its starting posture; If a match is found, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

23. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: Receive a first startup command including a user-selected startup posture, wherein the startup posture is a lying posture; Obtain the current pose of the humanoid robot; The current posture and the starting posture of the humanoid robot are matched and verified. If a match is found, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot performs the action of standing up. After the action is completed, the robot switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

24. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: In response to a selection operation, a first start command including a user-selected start posture is generated and sent to the humanoid robot. The start posture is a lying posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain its current posture after receiving the first start command including the user-selected start posture, and to perform a matching verification between the current posture and the start posture. If they match, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and a standing action is performed. After the action is completed, the mode is switched back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

25. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: In response to a selection operation, the user controller generates a first start command including the start posture selected by the user and sends the first start command including the start posture selected by the user to the humanoid robot, wherein the start posture is a lying posture. The humanoid robot receives a first start command including the start posture selected by the user; The humanoid robot acquires its current posture; The humanoid robot performs a matching and verification between its current posture and its starting posture; If a match is found, the humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs a standing action. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

26. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: Receive a first start command including a user-selected start posture, wherein the start posture is a squatting posture; Obtain the current pose of the humanoid robot; The current posture and the starting posture of the humanoid robot are matched and verified. If a match is found, the motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the robot performs the action of standing up. After the action is completed, the robot switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

27. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: In response to a selection operation, a first start command including a user-selected start posture is generated and sent to the humanoid robot. The start posture is a squatting posture. The first start command including the user-selected start posture is used to instruct the humanoid robot to obtain its current posture after receiving the first start command including the user-selected start posture, and to perform a matching verification between the current posture and the start posture. If they match, the motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and a standing action is performed. After the action is completed, the mode is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

28. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: In response to a selection operation, the user controller generates a first start command including the start posture selected by the user and sends the first start command including the start posture selected by the user to the humanoid robot, wherein the start posture is a squatting posture. The humanoid robot receives a first start command including the start posture selected by the user; The humanoid robot acquires its current posture; The humanoid robot performs a matching and verification between its current posture and its starting posture; If a match is found, the humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs a standing action. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

29. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the second start command, the current posture of the humanoid robot is detected; When the current posture of the humanoid robot is detected to be a suspended posture, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

30. The startup method according to claim 29, characterized in that, The foot contact detection in upright mode includes: In the upright mode, the actual output torque of the motor corresponding to the leg joint of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Based on the aforementioned foot contact force, determine whether the foot is in contact with the ground.

31. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: A second start command is generated in response to a trigger operation and sent to the humanoid robot. The second start command instructs the humanoid robot to detect its current posture in response to the second start command. When the current posture of the humanoid robot is detected to be a suspended posture, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. In addition, in upright mode, foot contact detection is performed. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

32. The startup method according to claim 31, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot. After sending the second start command to the humanoid robot, the method further includes: The electronic device is controlled to display a second startup page, which displays startup prompt information. The startup prompt information is used to prompt the user to perform a first operation, so that the gantry that lifts the humanoid robot lowers the humanoid robot vertically, so that the soles of the humanoid robot's feet are close to the ground.

33. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; The humanoid robot responds to the second start command by detecting its current posture; When the current posture of the humanoid robot is detected to be a suspended posture, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

34. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the second start command, the current posture of the humanoid robot is detected; When the current posture of the humanoid robot is a lying posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the second strategy network and the action of getting up is executed. After the execution is completed, it is switched back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

35. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: A second start command is generated in response to a trigger operation and sent to the humanoid robot. The second start command is used to instruct the humanoid robot to detect its current posture in response to the second start command. When the current posture of the humanoid robot is a lying posture, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot performs a standing up action. After the action is completed, the robot switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

36. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; The humanoid robot responds to the second start command by detecting the current posture of the humanoid robot; When the current posture of the humanoid robot is a lying posture, the humanoid robot switches the motion control mode of the humanoid robot from the zero torque mode to the second strategy network and performs the action of getting up. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

37. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the second start command, the current posture of the humanoid robot is detected; When the current posture of the humanoid robot is a squatting posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the third strategy network and the action of standing up is executed. After the execution is completed, it is switched back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

38. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: A second start command is generated in response to a trigger operation and sent to the humanoid robot. The second start command is used to instruct the humanoid robot to detect its current posture in response to the second start command. When the current posture of the humanoid robot is a squatting posture, the motion control mode of the humanoid robot is switched from the zero torque mode to the third strategy network and the robot performs a standing up action. After the action is completed, the robot switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

39. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a second start command in response to the trigger operation and sends the second start command to the humanoid robot; The humanoid robot responds to the second start command by detecting the current posture of the humanoid robot; When the current posture of the humanoid robot is a squatting posture, the humanoid robot switches the motion control mode of the humanoid robot from the zero torque mode to the third strategy network and performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

40. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is acquired and reported to the user controller. The current posture of the humanoid robot is a suspended posture. In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is sequentially switched from zero torque mode to damped mode, and then from damped mode to upright mode. Foot contact detection is performed in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

41. The startup method according to claim 40, characterized in that, The foot contact detection in upright mode includes: In the upright mode, the actual output torque of the motor corresponding to the leg joint of the humanoid robot is obtained; Calculate the torque difference between the actual output torque and the corresponding preset output torque; The contact force on the sole of the foot is obtained based on the torque difference. Based on the aforementioned foot contact force, determine whether the foot is in contact with the ground.

42. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, The method includes: In response to the first triggering operation, a second detection instruction is generated and sent to the humanoid robot. The second detection instruction is used to instruct the humanoid robot to acquire the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a suspended posture. Receive the current posture of the humanoid robot sent by the humanoid robot; In response to a second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command instructs the humanoid robot to sequentially switch its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode in response to the confirmation start command sent by the user controller. In upright mode, foot contact detection is performed, and when foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

43. The startup method according to claim 42, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot. After sending the confirmation start command to the humanoid robot, the method further includes: The electronic device is controlled to display a second startup page, which displays startup prompt information. The startup prompt information is used to prompt the user to perform a first operation, so that the gantry that lifts the humanoid robot lowers the humanoid robot vertically, so that the soles of the humanoid robot's feet are close to the ground.

44. The startup method according to claim 42, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: Upon receiving the current posture of the humanoid robot sent by the humanoid robot, the electronic device is controlled to display a first startup page, and in response to the user's startup trigger operation on the first startup page, the confirmation startup command is generated.

45. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a suspended posture. The user controller receives the current posture of the humanoid robot sent by the humanoid robot; The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; In response to the confirmation start command sent by the user controller, the humanoid robot sequentially switches its motion control mode from zero torque mode to damped mode, and then from damped mode to upright mode. The humanoid robot performs foot contact detection in upright mode. When foot contact is detected, the motion control mode is switched to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions.

46. ​​A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is acquired and reported to the user controller. The current posture of the humanoid robot is a lying posture. In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to the second strategy network and the robot performs a standing action. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

47. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, The method includes: In response to the first triggering operation, a second detection instruction is generated and sent to the humanoid robot. The second detection instruction is used to instruct the humanoid robot to acquire the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a lying posture. Receive the current posture of the humanoid robot sent by the humanoid robot; In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to switch its motion control mode from zero torque mode to the second strategy network and perform a standing action in response to the confirmation start command sent by the user controller. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

48. The startup method according to claim 47, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: Upon receiving the current posture of the humanoid robot sent by the humanoid robot, the electronic device is controlled to display a first startup page, and in response to the user's startup trigger operation on the first startup page, the confirmation startup command is generated.

49. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a lying posture. The user controller receives the current posture of the humanoid robot sent by the humanoid robot; The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; In response to the confirmation start command sent by the user controller, the humanoid robot switches its motion control mode from zero torque mode to the second strategy network and performs a standing action. After the action is completed, it switches back to the first strategy network. The first policy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The second policy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a lying position.

50. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a humanoid robot, characterized in that, include: In response to the second detection command issued by the user controller, the current posture of the humanoid robot is acquired and reported to the user controller. The current posture of the humanoid robot is a squatting posture. In response to the confirmation start command sent by the user controller, the motion control mode of the humanoid robot is switched from zero torque mode to the third strategy network and the robot performs the action of standing up. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

51. A method for initiating a humanoid robot based on a policy network trained through deep learning, applied to a user controller, characterized in that, include: In response to the first triggering operation, a second detection instruction is generated and sent to the humanoid robot. The second detection instruction is used to instruct the humanoid robot to acquire the current posture of the humanoid robot and report it to the user controller. The current posture of the humanoid robot is a squatting posture. Receive the current posture of the humanoid robot sent by the humanoid robot; In response to the second trigger operation, a confirmation start command is generated and sent to the humanoid robot. The confirmation start command is used to instruct the humanoid robot to switch its motion control mode from zero torque mode to the third strategy network and perform a standing action in response to the confirmation start command sent by the user controller. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

52. The startup method according to claim 51, characterized in that, The user controller is located in an electronic device communicatively connected to the humanoid robot, and the method further includes: Upon receiving the current posture of the humanoid robot sent by the humanoid robot, the electronic device is controlled to display a first startup page, and in response to the user's startup trigger operation on the first startup page, the confirmation startup command is generated.

53. A method for initiating a humanoid robot based on a policy network trained through deep learning, characterized in that, include: The user controller generates a second detection command in response to the first trigger operation and sends the second detection command to the humanoid robot; The humanoid robot acquires its current posture and reports it to the user controller. The current posture of the humanoid robot is a squatting posture. The user controller receives the current posture of the humanoid robot sent by the humanoid robot; The user controller generates a confirmation start command in response to the second trigger operation and sends the confirmation start command to the humanoid robot; In response to the confirmation start command sent by the user controller, the humanoid robot switches its motion control mode from zero torque mode to the third strategy network and performs a standing action. After the action is completed, it switches back to the first strategy network. The first strategy network is trained using deep learning technology and is used to control the humanoid robot to perform standing and / or walking actions. The third strategy network is trained using deep learning technology and is used to control the humanoid robot to stand up from a squatting position.

54. A humanoid robot, characterized in that, include: A first processor and a first memory, the first memory being connected to the first processor, the first memory storing a computer program, the first processor executing the humanoid robot initiation method based on a policy network trained by deep learning as described in any one of claims 1, 2, 8, 13, 18, 19, 23, 26, 29, 30, 34, 37, 40, 41, 46, and 50.

55. An electronic device, characterized in that, include: A second processor and a second memory, the second memory being connected to the second processor, the second memory storing a computer program, the second processor executing the humanoid robot initiation method based on a policy network trained by deep learning as described in any one of claims 3-6, 9-11, 14-16, 20, 21, 24, 27, 31, 32, 35, 38, 42-44, 47, 48, 51, and 52.

56. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when called by a processor, executes the humanoid robot initiation method based on a policy network trained by deep learning as described in any one of claims 1-53.

Images