Robot, control method and device thereof and storage medium

Abstract

The invention provides a biped robot, a control method and device thereof and a storage medium, and the control method comprises the steps: obtaining the motion information of a motor at a parallel joint when the robot moves, inputting the motion information into a pre-trained parallel motion model to obtain sole posture information, performing closed-loop control according to the sole attitude information and the expected attitude information of the robot to obtain a first control torque, performing closed-loop control based on the first control torque and the expected motion information of the parallel joints to obtain a second control torque, and controlling motors at the parallel joints to work according to the second control torque. In the embodiment of the invention, the plantar posture of the robot is estimated through the parallel motion model, and compared with a traditional numerical iterative algorithm, the algorithm process and the data volume are greatly simplified, and the control efficiency of the parallel robot is improved. And moreover, the motion stability and the motion precision of the robot are improved by adopting torque level control on the robot.

Description

Technical Field

[0001] The present disclosure relates to the technical field of intelligent robots, and in particular to a robot and a control method, device, and storage medium thereof. Background Art

[0002] Nowadays, robots are widely used in various life scenarios. Take bipedal robots as an example. Compared with quadruped robots, bipedal robots are closer to human movements, and bipedal robots can free the upper limbs to achieve more operations. Therefore, motion control of bipedal robots is one of the key research directions in the field of robotics. Summary of the invention

[0003] In order to improve the control stability and real-time performance of a robot, the embodiments of the present disclosure provide a robot control method, a device, a robot and a storage medium.

[0004] In a first aspect, an embodiment of the present disclosure provides a robot control method, comprising:

[0005] Obtain the motion information of the motors at the parallel joints when the robot moves;

[0006] Inputting the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model;

[0007] Performing closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate;

[0008] Performing closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain a second control torque of the parallel joint;

[0009] The operation of the motor at the parallel joint is controlled according to the second control torque.

[0010] In some embodiments, the motion information includes a motor angle and a motor angular velocity, and the plantar posture information includes a plantar posture angle and a plantar posture angular velocity;

[0011] The step of inputting the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model comprises:

[0012] Inputting the motor angle into a pre-trained parallel motion model to obtain the plantar posture angle of the robot output by the parallel motion model;

[0013] Based on the Jacobian matrix of the plantar plate of the robot and the parallel joint, the plantar posture angular velocity corresponding to the motor angular velocity is determined.

[0014] In some embodiments, the plantar posture information includes a plantar posture angle and a plantar posture angular velocity, and the desired posture information includes an desired posture angle, an desired posture angular velocity, and an expected feedforward torque of the plantar plate;

[0015] The step of performing closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate includes:

[0016] The first control torque is obtained by performing closed-loop control according to the difference between the plantar posture angle and the expected posture angle, the difference between the plantar posture angular velocity and the expected posture angular velocity, and the expected feedforward torque.

[0017] In some embodiments, the motion information includes a motor angle and a motor angular velocity, the desired posture information includes a desired posture angle and a desired posture angular velocity of the plantar plate, and the desired motion information includes a desired motor angle and a desired motor angular velocity of the parallel joint;

[0018] The method of performing closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain the second control torque of the parallel joint includes:

[0019] Determine the desired motor angle based on the desired posture angle, and determine the desired motor angular velocity based on the desired posture angular velocity;

[0020] Determine a motor feedforward torque corresponding to the first control torque based on a Jacobian matrix of the plantar plate of the robot and the parallel joint;

[0021] The second control torque is obtained by performing closed-loop control according to the difference between the desired motor angle and the motor angle, the difference between the desired motor angular velocity and the motor angular velocity, and the motor feedforward torque.

[0022] In some implementations, determining the desired motor angle based on the desired attitude angle, and determining the desired motor angular velocity based on the desired attitude angular velocity, includes:

[0023] Determine a first equation between the desired attitude angle and the desired motor angle, and a second equation between the desired attitude angular velocity and the desired motor angular velocity based on an inverse kinematics algorithm;

[0024] The corresponding expected motor angle is determined based on the expected posture angle and the first equation, and the corresponding expected motor angular velocity is determined based on the expected posture angular velocity and the second equation.

[0025] In some implementations, the training process of the parallel kinematic model includes:

[0026] Acquire a training data set, wherein the sample data in the training data set includes sample motion information of the motor at the parallel joint of the robot, and reference posture information of the plantar plate of the robot corresponding to the sample motion information;

[0027] Inputting the sample motion information into the parallel motion model to be trained, and obtaining the plantar posture information of the plantar plate of the robot output by the parallel motion model;

[0028] According to the difference between the plantar posture information and the reference posture information, the model parameters of the parallel motion model are adjusted until a convergence condition is met, thereby obtaining a trained parallel motion model.

[0029] In some embodiments, the parallel joint includes a knee joint motor and a connecting rod mechanism of the robot, and the knee joint motor drives the plantar plate of the robot to move through the connecting rod mechanism.

[0030] In a second aspect, an embodiment of the present disclosure provides a control device for a robot, comprising:

[0031] An information acquisition module is configured to acquire motion information of motors at parallel joints when the robot moves;

[0032] A motion estimation module, configured to input the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model;

[0033] A first torque module is configured to perform closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate;

[0034] A second torque module is configured to perform closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain a second control torque of the parallel joint;

[0035] The motion control module is configured to control the operation of the motor at the parallel joint according to the second control torque.

[0036] In some embodiments, the motion information includes a motor angle and a motor angular velocity, the plantar posture information includes a plantar posture angle and a plantar posture angular velocity; and the motion estimation module is configured to:

[0037] Inputting the motor angle into a pre-trained parallel motion model to obtain the plantar posture angle of the robot output by the parallel motion model;

[0038] Based on the Jacobian matrix of the plantar plate of the robot and the parallel joint, the plantar posture angular velocity corresponding to the motor angular velocity is determined.

[0039] In some embodiments, the plantar posture information includes a plantar posture angle and a plantar posture angular velocity, and the desired posture information includes an expected posture angle, an expected posture angular velocity, and an expected feedforward torque of the plantar plate; the first torque module is configured as follows:

[0040] The first control torque is obtained by performing closed-loop control according to the difference between the plantar posture angle and the expected posture angle, the difference between the plantar posture angular velocity and the expected posture angular velocity, and the expected feedforward torque.

[0041] In some embodiments, the motion information includes a motor angle and a motor angular velocity, the desired posture information includes an expected posture angle and an expected posture angular velocity of the plantar plate, and the expected motion information includes an expected motor angle and an expected motor angular velocity of the parallel joint; the second torque module is configured as follows:

[0042] Determine the desired motor angle based on the desired posture angle, and determine the desired motor angular velocity based on the desired posture angular velocity;

[0043] Determine a motor feedforward torque corresponding to the first control torque based on a Jacobian matrix of the plantar plate of the robot and the parallel joint;

[0044] The second control torque is obtained by performing closed-loop control according to the difference between the desired motor angle and the motor angle, the difference between the desired motor angular velocity and the motor angular velocity, and the motor feedforward torque.

[0045] In some embodiments, the second torque module is configured to:

[0046] Determine a first equation between the desired attitude angle and the desired motor angle, and a second equation between the desired attitude angular velocity and the desired motor angular velocity based on an inverse kinematics algorithm;

[0047] The corresponding expected motor angle is determined based on the expected posture angle and the first equation, and the corresponding expected motor angular velocity is determined based on the expected posture angular velocity and the second equation.

[0048] In some embodiments, the motion estimation module is configured to:

[0049] Acquire a training data set, wherein each sample data in the training data set includes sample motion information of the motor at the parallel joint of the robot, and reference posture information of the plantar plate of the robot corresponding to the sample motion information;

[0050] Inputting the sample motion information into the parallel motion model to be trained, and obtaining the plantar posture information of the plantar plate of the robot output by the parallel motion model;

[0051] According to the difference between the plantar posture information and the reference posture information, the model parameters of the parallel motion model are adjusted until a convergence condition is met, thereby obtaining a trained parallel motion model.

[0052] In some embodiments, the parallel joint includes a knee joint motor and a connecting rod mechanism of the robot, and the knee joint motor drives the plantar plate of the robot to move through the connecting rod mechanism.

[0053] In a third aspect, the present disclosure provides a bipedal robot, comprising:

[0054] A plantar plate and a parallel joint, wherein the plantar plate is connected to a motor of the parallel joint via a connecting rod mechanism, and the motor drives the plantar plate to move via the connecting rod mechanism;

[0055] The controller comprises a processor and a memory, wherein the memory stores computer instructions, and the computer instructions are used to enable the processor to execute the method according to any embodiment of the first aspect.

[0056] In a fourth aspect, an embodiment of the present disclosure provides a storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the method according to any embodiment of the first aspect.

[0057] The control method of the disclosed embodiment includes obtaining the motion information of the motor at the parallel joint when the robot moves, inputting the motion information into a pre-trained parallel motion model to obtain the plantar posture information, performing closed-loop control based on the plantar posture information and the desired posture information of the robot to obtain a first control torque, performing closed-loop control based on the first control torque and the desired motion information of the parallel joint to obtain a second control torque, and controlling the operation of the motor at the parallel joint according to the second control torque. In the disclosed embodiment, the plantar posture of the robot is estimated by a parallel motion model, which greatly simplifies the algorithm process and data volume compared to the traditional numerical iterative algorithm, and improves the control efficiency of the parallel robot. Moreover, by adopting torque-level control for the robot, the robot's motion stability and motion accuracy are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the specific embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] Figure 1 It is a structural schematic diagram and a motion diagram of a biped robot in some embodiments of the present disclosure.

[0060] Figure 2 is a flow chart of a control method according to some embodiments of the present disclosure.

[0061] Figure 3 is a flow chart of a control method according to some embodiments of the present disclosure.

[0062] Figure 4 is a flow chart of a control method according to some embodiments of the present disclosure.

[0063] Figure 5 is a flow chart of a control method according to some embodiments of the present disclosure.

[0064] Figure 6 is a flow chart of a control method according to some embodiments of the present disclosure.

[0065] Figure 7 It is a structural block diagram of a control device in some embodiments of the present disclosure.

[0066] Figure 8 is a structural block diagram of a bipedal robot according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0067] The technical solution of the present disclosure will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described implementation is a part of the implementation of the present disclosure, rather than all the implementations. Based on the implementation in the present disclosure, all other implementations obtained by ordinary technicians in the field without creative work are within the scope of protection of the present disclosure. In addition, the technical features involved in the different implementations of the present disclosure described below can be combined with each other as long as they do not conflict with each other.

[0068] Nowadays, robots are widely used in various life scenarios. Take bipedal robots as an example. Compared with quadruped robots, bipedal robots are closer to human movements, and bipedal robots can free the upper limbs to achieve more operations. Therefore, the control of bipedal robots is one of the key research directions in the field of robotics.

[0069] The design goal of the bipedal robot leg structure is to adapt to human life and work scenarios by simulating human degrees of freedom and range of motion. At present, there are two main leg configurations of motor-driven bipedal robots, namely series type and parallel type.

[0070] The structural feature of the serial biped robot is that the motors are arranged strictly according to the degrees of freedom of the leg joints. Since the ankle joint has two degrees of freedom, two motors need to be arranged at the ankle joint and the topological structure is serial. The advantage of this structure is that the kinematic and dynamic algorithms can refer to the control theory of traditional serial manipulators. The algorithms are relatively mature and can easily realize the motion control of the biped robot. However, the disadvantage is that the ankle joint needs to be arranged with two motors, which makes the ankle joint structure bloated, and the mass of the lower leg is concentrated at the ankle joint. The leg inertia is large, which affects the motion performance of the robot itself.

[0071] In order to reduce the inertia of the robot's legs, in the parallel bipedal robot architecture, the two-degree-of-freedom motor of the ankle joint can be moved up to the knee joint or hip joint, and the ankle joint movement posture can be controlled by using multi-link, ball screw or belt drive. This method improves the leg stiffness of the bipedal robot to a certain extent, greatly reduces the leg rotation inertia, and improves the dynamic performance of the robot.

[0072] However, for parallel bipedal robots, the inability to use traditional kinematic and dynamic algorithms makes robot control more difficult. For multi-degree-of-freedom bipedal robots, the kinematic and dynamic solutions of the parallel structure will affect the real-time performance of the overall control, making it difficult to ensure the control stability and real-time performance of the robot.

[0073] In the related art, some kinematic algorithms for parallel biped robots use numerical iteration to solve the parallel joint angles, that is, the plantar posture is calculated given the parallel joint angles, and the parallel joint angles in the next iteration are adjusted according to the difference between the plantar posture and the expected posture until the results converge. In this control method, a closed-loop control process may require multiple iterations to ensure the result convergence, the calculation efficiency is very poor, the real-time performance of the results cannot be guaranteed, and only the parallel joint angle results are considered, resulting in poor robot motion stability.

[0074] In view of the above-mentioned defects, the embodiments of the present disclosure provide a robot control method, device, robot and storage medium, aiming to improve the control stability and real-time performance of the robot and ensure the robot's motion performance.

[0075] Figure 1 A schematic diagram of the lower leg structure and a simplified motion diagram of a parallel biped robot in some embodiments of the present disclosure are shown.

[0076] exist Figure 1 In the example, the calf structure of the biped robot includes knee joint motors M1 and M2. A 2-DOF motor is no longer provided at the ankle joint where the calf rod 200 and the plantar plate 300 are hinged. Instead, two sets of connecting rod mechanisms are used to connect the plantar plate 300 to the knee joint motors M1 and M2.

[0077] like Figure 1 As shown, the first link mechanism includes a first horizontal rod 110 and a first vertical rod 210, one end of the first horizontal rod 110 is connected to the output shaft of the motor M2, and the other end is connected to one end of the first vertical rod 210, and the other end of the first vertical rod 210 is connected to the foot plate 300. The second link mechanism includes a second horizontal rod 120 and a second vertical rod 220, one end of the second horizontal rod 120 is connected to the output shaft of the motor M1, and the other end is connected to one end of the second vertical rod 220, and the other end of the second vertical rod 220 is connected to the foot plate 300.

[0078] Combination Figure 1 As can be understood from the motion diagram shown, the motor M1 and the motor M2 can drive the two sets of connecting rod mechanisms to move, thereby driving the sole plate 300 to generate a 2-degree-of-freedom motion, which are the roll angle θ around the y-axis and the roll angle θ. roll and the pitch angle θ around the x-axis pitc h, that is, the posture of the sole plate 300 can be expressed by {θ roll ,θ pitc h} to indicate.

[0079] Understandably, Figure 1 Only one leg structure of the biped robot is shown in the figure, and the other leg structure is exactly the same, which will not be described in detail in this disclosure. Based on the understanding of the basic structure of the above parallel biped robot, the robot control method of the embodiment of the present disclosure is described below.

[0080] like Figure 2 As shown, in some embodiments, the control method of the robot of the present disclosure example includes:

[0081] S210, obtaining movement information of motors at parallel joints when the robot moves.

[0082] In the embodiment of the present disclosure, the joint connected to the sole plate through the transmission mechanism is defined as a parallel joint, for example Figure 1 In the example, the sole plate 300 is connected to the motor at the knee joint through a connecting rod mechanism, and is driven to move by the knee joint motors M1 and M2, so that the knee joint in this example is a parallel joint.

[0083] It is worth noting that for a parallel biped robot, the sole plate is not limited to being connected to the knee joint for transmission. For example, in some embodiments, the movement of the sole plate can be connected to the hip joint of the robot through a transmission mechanism. In this case, the hip joint is a parallel joint. Those skilled in the art can understand this, and this disclosure will not elaborate on it.

[0084] In the disclosed embodiment, when the robot moves, the motor at the parallel joint can drive the foot plate to move, and the movement information of the motor at the parallel joint can include the motor angle and / or the motor angular velocity. Figure 1 In the example, during the movement of the robot, the rotation of the motors M1 and M2 can drive the posture of the sole plate 300 to change, so the rotation angle and / or angular velocity of the motors M1 and M2 is the movement information described in the present disclosure.

[0085] The motion information of the motor can be collected in real time through the motor encoder. For example, during the movement of the robot, the motor encoder can collect information such as the motor angle, motor angular velocity, and motor angular acceleration of each motor.

[0086] For example, in some embodiments, in the examples of the present disclosure, the motion information of the motor includes the motor angle and the motor angular velocity. Figure 1 As an example, the motor angle of motor M1 is expressed as θ1, and the motor angular velocity is expressed as The motor angle of motor M2 is denoted as θ2, and the motor angular velocity is denoted as That is, in this example, the motion information of the motor at the parallel joint includes the motor angles θ1, θ2 and the motor angular velocity

[0087] S220, inputting the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model.

[0088] As mentioned above, the biped robot with parallel mechanism cannot adopt traditional kinematic and dynamic algorithms, so there is no analytical expression for the correct kinematic solution, and it is impossible to directly solve the posture of the plantar plate based on the motion information of the parallel joint motor.

[0089] In the disclosed embodiment, a parallel motion model based on deep neural networks (DNN) is used to estimate the plantar posture information of the robot's plantar plate according to the motion information of the motor.

[0090] In some embodiments, a parallel motion model can be pre-built and trained, the input of the parallel motion model is the motion information of the motor, and the output is the sole posture information of the sole plate. Figure 1In this example, the motion information of the motors includes the motor angles θ1 and θ2 of the motors M1 and M2. The motor angles θ1 and θ2 are input into the parallel motion model to obtain the plantar posture information {θ roll ,θ pitch}, θ roll represents the roll angle of the sole plate 300, θ pitch Indicates the pitch angle of the sole plate 300 .

[0091] The model structure and model training process of the parallel motion model are explained in the following of this disclosure and will not be described in detail here.

[0092] It is worth noting that since there is no analytical expression for the kinematics of the parallel mechanism, the related art algorithm uses a numerical iteration method to estimate the plantar posture, and the calculation results are difficult to converge, and a closed-loop control process may require multiple iterations, the calculation efficiency is very poor, and the real-time performance of the results cannot be ensured. In the embodiment of the present disclosure, a neural network model is used to determine the plantar posture information of the robot, without the need for a complex iterative process, thereby improving the efficiency and stability of the plantar posture solution.

[0093] S230, performing closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate.

[0094] In the disclosed embodiment, after obtaining the sole posture information, it is necessary to control the posture of the sole plate according to the sole posture information and the desired posture information so that the sole plate moves to the desired posture.

[0095] The expected posture information refers to the expected posture of the plantar plate movement. The expected posture information is generally determined by the control task. For example, when the robot walks on flat ground, the expected posture of the plantar plate should be level with the ground. Therefore, the expected posture information is the posture information that makes the plantar plate level.

[0096] In some embodiments, the desired posture information includes the desired posture angle, the desired posture angular velocity, and the desired feedforward torque of the sole plate. The desired posture angle includes the desired roll angle of the sole plate. and the desired pitch angle The desired attitude angular velocity includes the desired rolling angular velocity of the sole plate and the desired pitch angular velocity The desired feedforward torque includes the desired rolling angular torque of the foot plate and the desired pitch moment

[0097] In the embodiments of the present disclosure, PD combined with feedforward control can be used to implement closed-loop control of the robot kinematics, so as to calculate the control torque for the posture of the plantar plate according to the plantar posture information and the expected posture information during the actual movement of the robot, and the control torque is the first control torque. The first control torque can be understood as the movement posture of the plantar plate can reach the expected posture when the first control torque is provided to the plantar plate. The process of PD closed-loop control is described in the following of the present disclosure and will not be described in detail here.

[0098] S240, performing closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain a second control torque of the parallel joint.

[0099] It should be understood that after calculating the first control torque for the plantar posture, for a traditional serial robot, since its ankle joint is equipped with a 2-DOF motor, the ankle joint motor can be controlled to output the first control torque, so that the plantar posture can reach the desired posture.

[0100] However, in the embodiment of the present disclosure, Figure 1 The parallel mechanism shown in the figure cannot directly output the first control torque at the ankle joint position because the motor is no longer arranged at the ankle joint. Instead, the first control torque for the ankle joint needs to be resolved to the controllable parallel joint motor, that is, the second control torque at the parallel joint position is calculated based on the first control torque.

[0101] In some embodiments, PD combined with feedforward control can also be used to achieve closed-loop control of the parallel joint torque. The control quantity of the closed-loop control is the second control torque at the parallel joint, and the control law is the first control torque of the robot plantar plate and the desired motion information of the parallel joint.

[0102] The expected motion information of the parallel joint refers to the expected motion state of the parallel joint motor, which can be understood as the motion information of the parallel joint motor required to move the robot's foot plate to the expected posture. For example, in some embodiments, the expected motion information of the parallel joint includes the expected motor angle, expected motor angular velocity, and motor feedforward torque of the parallel joint motor.

[0103] For example Figure 1 In this example, the expected motor angle includes the expected motor angle of motor M1. and the desired motor angle of motor M2 The desired motor angular velocity includes the desired motor angular velocity of motor M1 and the desired motor angular velocity of motor M2 The motor feedforward torque includes the feedforward torque of motor M1 and the feedforward torque of motor M2

[0104] In the example of the present disclosure, the desired motor angle and the desired motor angular velocity can be obtained by solving the kinematic inverse solution, and the motor feedforward torque can be obtained by solving the first control solution. The specific solution process is described below in the present disclosure.

[0105] In the embodiment of the present disclosure, closed-loop control is performed according to the desired motion information of the parallel joint and the first control torque of the plantar plate of the robot to obtain the second control torque. The process of closed-loop control is described below in the present disclosure and will not be described in detail here.

[0106] S250, controlling the operation of the motor at the parallel joint according to the second control torque.

[0107] Combined with the above, it can be known that the second control torque refers to the torque at the parallel joint, which can be understood as the torque that needs to be applied to the parallel joint motor in order to move the plantar plate to the desired posture. Therefore, in the embodiment of the present disclosure, after obtaining the second control torque, the parallel joint motor can be controlled to output the corresponding torque according to the second control torque, thereby realizing robot control.

[0108] For example Figure 1 In the exemplary implementation, the second control torque includes the control torque of the motor M1 and the control torque of the motor M2. The sole of the robot can be kept in a desired posture by controlling the motors M1 and M2 respectively according to the control torques.

[0109] In the disclosed embodiments, the plantar posture of the robot is estimated by a parallel motion model, which greatly simplifies the algorithm process and data volume compared to the traditional numerical iterative algorithm, and improves the control efficiency of the parallel robot.

[0110] In addition, it is worth mentioning that in the embodiment of the present disclosure, torque-level control is used for robot kinematics, that is, the motor output is controlled based on the calculated torque. Compared with the traditional motor position-based control algorithm, the control accuracy and motion stability of the robot motion can be effectively improved.

[0111] For example, in a conventional control algorithm based on motor position, the control quantity finally solved is the motor angle, thereby controlling the angle corresponding to the motor output. In the embodiment of the present disclosure, the control quantity finally solved is the torque, thereby controlling the torque corresponding to the motor output.

[0112] Taking the bipedal robot walking scene as an example, if the robot's sole steps on an obstacle such as a stone, the control algorithm based on the motor position in the traditional solution will output a large torque to ensure that the sole plate remains horizontal in order to rotate the motor to the target angle, so that the fuselage is prone to tilt or even rollover. In the embodiment of the present disclosure, when the robot also steps on an obstacle such as a stone, it only needs to control the motor to output the corresponding torque. Even if the sole plate is affected by the obstacle and is not level with the ground, the fuselage can still remain stable, greatly reducing the risk of tilting or rollover, and the robot has better dynamic performance.

[0113] From the above, it can be seen that in the embodiment of the present disclosure, the robot's plantar posture is estimated by a parallel kinematic model, which greatly simplifies the algorithm process and data volume compared to the traditional numerical iteration algorithm, and improves the control efficiency of the parallel robot. Moreover, by adopting torque level control for the robot, the robot's motion stability and motion accuracy are improved.

[0114] For the convenience of further understanding and explanation, in the following embodiments of the present disclosure, the parallel biped robot will be Figure 1 The parallel architecture shown is illustrated.

[0115] In the embodiments of the present disclosure, a neural network architecture of a parallel motion model may be pre-constructed, and then the constructed parallel motion model may be trained using sample data to obtain a trained parallel motion model. The network construction and training process of the parallel motion model is described below.

[0116] In some embodiments, the parallel motion model can adopt a two-layer or multi-layer feedforward neural network, whose hidden layer uses sigmoid neurons and the output layer uses linear neurons. With sufficient training data and neurons, the model can arbitrarily fit multi-dimensional input and output problems. For example, in some embodiments, the input of the parallel motion model is the motion information of the motor at the parallel joint, and the output is the plantar posture information of the robot's plantar plate.

[0117] After constructing the parallel kinematic model, Figure 3 The method process shown in the figure is used to train the parallel motion model. Figure 3 Provide explanation.

[0118] like Figure 3 As shown, in some embodiments, in the control method of the example of the present disclosure, the training process of the parallel kinematic model includes:

[0119] S310: Obtain a training data set.

[0120] In the disclosed embodiment, the training data set refers to a sample set used to train the parallel motion model, which includes a large amount of sample data, wherein each sample data includes sample motion information of the motor at the parallel joint of the robot, and reference posture information of the plantar plate of the robot corresponding to the sample motion information.

[0121] by Figure 1 As shown in the figure, in a sample data, the sample motion information of the parallel joint includes the motor angle θ1 of the knee joint motor M1 and the motor angle θ2 of the motor M2. The reference posture information of the sole plate includes the roll angle θ roll and the pitch angle θ pitch .

[0122] In some embodiments, the sample data can be obtained by measuring in a three-dimensional modeling software, that is, the rolling angle θ of the sole plate of any given robot in the three-dimensional modeling software roll and the pitch angle θ pitch , and the corresponding motor angles θ1 and θ2 are measured.

[0123] In other embodiments, the sample data can also be obtained by inverse kinematics solution of the parallel robot, that is, given the roll angle θ of the plantar plate of the robot roll and the pitch angle θ pitch In the case of, the corresponding motor angles θ1 and θ2 can be calculated by inverse kinematics solution. The algorithm process of inverse kinematics solution is described below in this disclosure.

[0124] In the above manner, multiple sample data can be obtained, and these sample data can be constructed as a training data set. For example, in one example, the training data set may include 250,000 sample data in total. During the model training process, 75% of the sample data in the training data set can be used as a training set, and 25% of the sample data can be used as a validation set to verify the generalization ability of the model.

[0125] S320, inputting the sample motion information into the parallel motion model to be trained, and obtaining the plantar posture information of the plantar plate of the robot output by the parallel motion model.

[0126] In the embodiment of the present disclosure, taking a sample data as an example, the sample motion information θ1 and θ2 of the sample data are input into the parallel motion model, and the plantar posture information θ′ predicted and output by the parallel motion model can be obtained. roll and θ′ pitch .

[0127] It can be understood that the plantar posture information θ′ roll and θ′ pitch represents the predicted value of the parallel motion model, and the sample data includes the reference posture information θ roll and θpitch It represents the true value (GT, Ground Truth), so the difference between the two can reflect the performance of the parallel motion model.

[0128] S330. According to the difference between the plantar posture information and the reference posture information, the model parameters of the parallel motion model are adjusted until the convergence condition is met, thereby obtaining the trained parallel motion model.

[0129] Taking the aforementioned sample data as an example, the plantar posture information θ′ output by the parallel kinematic model is roll and θ′ pitch , and the reference posture information θ included in the sample data roll and θ pitch The difference between the two is the loss value of model training. Based on the loss value, the model parameters of the parallel motion model are optimized using the back propagation algorithm to complete an iterative training process.

[0130] In some implementations, the Levenberg-Marquardt back-propagation algorithm may be used for model training.

[0131] The above is only an example of one sample data. For multiple sample data in the training data set, the above training process can be repeated in sequence to continuously iteratively tune the model parameters of the parallel motion model until the convergence conditions are met. The model training process can be completed to obtain the trained parallel motion model. The trained parallel motion model can be exported as a function and stored in the robot for subsequent calls.

[0132] After completing the construction and model training of the parallel motion model through the above process, the trained parallel motion model can be used to implement the control method disclosed in the present invention, which is described in detail below.

[0133] Combination Figure 1 As shown, during the movement of the robot, the motors M1 and M2 at the parallel joints can collect movement information. In some embodiments, the movement information includes motor angles and motor angular velocities. The motor angles include the motor angle θ1 of the motor M1 and the motor angle θ2 of the motor M2. The motor angular velocities include the motor angular velocities of the motor M1. and the motor angular velocity of motor M2

[0134] like Figure 4 As shown, in some embodiments, the control method of the example of the present disclosure, the process of obtaining the plantar posture information of the robot includes:

[0135] S410, inputting the motor angle into the pre-trained parallel motion model to obtain the plantar posture angle of the robot output by the parallel motion model.

[0136] S420, based on the Jacobian matrix of the robot's foot plate and the parallel joint, determine the foot posture angular velocity corresponding to the motor angular velocity.

[0137] In the embodiment of the present disclosure, according to the above, the motor angles θ1 and θ2 can be input into the pre-trained parallel motion model to obtain the plantar posture angle output by the parallel motion model, and the plantar posture angle is expressed as θ roll and θ pitch .

[0138] At the same time, based on the following formula (1), according to the motor angular velocity and Calculate the angular velocity of the plantar posture and

[0139]

[0140] In formula (1), J represents the Jacobian matrix of the sole plate and the parallel joint in the parallel mechanism. The calculation process of the Jacobian matrix will be described below.

[0141] The plantar posture angle θ can be obtained through the above parallel motion model. roll and θ pitch , and according to formula (1), the angular velocity of the plantar posture can be calculated and In the disclosed embodiment, the plantar posture information of the plantar plate of the robot includes the plantar posture angle and the plantar posture angular velocity.

[0142] After obtaining the plantar posture information θ roll ,θ pitch , and After that, a closed-loop control can be constructed based on the plantar posture information and the desired posture information. In the example disclosed in this disclosure, a closed-loop controller is constructed by combining PD with feedforward control. The closed-loop controller is expressed as:

[0143]

[0144] In formula (2), τ roll , τ pitch Represents the first control torque. Kp roll , K ppitch is the foot posture angle error gain, Kd roll , Kd pitch is the foot posture angular velocity error gain. is the desired attitude angle, is the desired attitude angular velocity, is the expected feedforward torque. In the disclosed example, the expected posture information includes the aforementioned expected posture angle Expected attitude angular velocity and the expected feedforward torque

[0145] The closed-loop control process of formula (2) represents closed-loop control based on the difference between the plantar posture angle and the desired posture angle, the difference between the plantar posture angular velocity and the desired posture angular velocity, and the desired feedforward torque. According to the closed-loop control process of formula (2), the first control torque τ of the robot ankle joint can be obtained: roll , τ pitch τ roll represents the torque that controls the roll angle of the robot's foot plate, τ pitch Represents the torque that controls the pitch angle of the robot's foot plate.

[0146] According to the above, we can know that when we get the first control torque τ roll , τ pitch After that, the first control torque τ of the ankle joint needs to be roll , τ pitch Solve to the parallel joint, and obtain the second control torque of the parallel joint through closed-loop control. The closed-loop control process of the second control torque can be expressed as:

[0147]

[0148] In formula (3), τ1 and τ2 represent the second control torque. Kp1 and Kp2 are the joint motor angle error gains, Kd1 and Kd2 are the joint motor angular velocity error gains. θ1 and θ2 are the motor angles, is the motor angular velocity, which is obtained through the motion information collected above.

[0149] The unknowns in formula (3) include: represents the desired motor angle, represents the desired motor angular velocity, Represents the motor feedforward torque. The expected motion information of the parallel joint described in the embodiment of the present disclosure includes the aforementioned expected motor angle, expected motor angular velocity and motor feedforward torque.

[0150] For the calculation process of the unknown quantity in formula (3), the present disclosure is combined with Figure 5 The implementation method is described.

[0151] like Figure 5 As shown, in some embodiments, the control method of the example of the present disclosure, the process of obtaining the second control torque of the parallel joint includes:

[0152] S510: Determine an expected motor angle based on an expected posture angle, and determine an expected motor angular velocity based on an expected posture angular velocity.

[0153] In the embodiment of the present disclosure, the desired motor angle According to the above expected posture angle Calculate the expected attitude angle Refers to the planned foot plate attitude angle, which is a known quantity, so according to the expected attitude angle Perform inverse kinematics to get the desired motor angle

[0154] Similarly, the expected motor angular velocity According to the above expected attitude angular velocity Calculated, the expected attitude angular velocity It represents the planned angular velocity of the plantar plate, which is a known quantity. Perform inverse kinematics to get the desired motor angular velocity Combine the following Figure 6 The process of inverse solution of the kinematics of the parallel mechanism is explained.

[0155] like Figure 6 As shown, in some embodiments, the control method of the present disclosure example includes:

[0156] S511. Determine a first equation of a desired attitude angle and a desired motor angle, and a second equation of a desired attitude angular velocity and a desired motor angular velocity based on a kinematic inverse solution algorithm.

[0157] S512: Determine the corresponding expected motor angle based on the expected attitude angular velocity and the first equation, and determine the corresponding expected motor angular velocity based on the expected attitude angular velocity and the second equation.

[0158] Combination Figure 1 As shown, the parallel mechanism includes two closed chain structures. Taking the closed chain on one side where the motor M1 is located as an example, the expression of points A, B, and C in the coordinate system O-xyz is:

[0159]

[0160] The coordinates of point B B It can be calculated based on the coordinates of point A.

[0161]

[0162] The length of the first horizontal rod 110 is l AB, the rotation angle is θ1, which represents the angle between the first horizontal rod 110 and the horizontal plane. C According to the foot posture angle θ roll ,θ pitch Calculated

[0163]

[0164] In formula (6), r C′ Indicates the foot posture angle θ roll ,θ pitch The coordinates of point C when both are 0. The length l of the first vertical rod 210 BC It can be calculated based on the coordinates of point B and point C.

[0165]

[0166] Then substitute formula (5) into formula (7), and we get

[0167]

[0168] Then formula (7) can be simplified to

[0169] -acosθ1+bsinθ1=c (9)

[0170] In formula (9), a=(x C -x A ),b=(z C -z A ),

[0171] According to the auxiliary angle formula, the rotation angle θ1 of the first horizontal rod 110 can be obtained as:

[0172]

[0173] Formula (10) is the first equation of the desired posture angle and the desired motor angle described in the present disclosure. By formula (10), the relationship between the plantar posture angle of the plantar plate of the robot and the motor angle of the parallel joint can be expressed. The above description is only based on a closed chain structure where the motor M1 is located. The same is true for another closed chain structure where the motor M2 is located. Referring to the above process, the relationship between the plantar posture angle of another closed chain structure and the motor angle of the parallel joint can be expressed.

[0174] In S510, the desired attitude angle Substituting into formula (10), the corresponding desired motor angle can be calculated

[0175] For the expected attitude angular velocity The process of inverse kinematics solution is as follows:

[0176] The speed of point B r B It can be obtained by taking the derivative of the position of point B.

[0177]

[0178] The velocity of point C can be obtained by taking the derivative of the position of point C.

[0179]

[0180] Where S(ω) represents the antisymmetric matrix of vector ω, which is in the following form:

[0181]

[0182] Since the length of the first vertical rod 210 is unchanged, the projection of the velocity of point B on the first vertical rod 210 is equal to the projection of the velocity of point C on the first vertical rod 210, that is:

[0183]

[0184] According to the dot product operation, the angular velocity of the first horizontal rod 110 can be obtained by expanding formula (14):

[0185]

[0186] Formula (15) is the second equation of the desired posture angular velocity and the desired motor angular velocity described in the present disclosure. Through formula (15), the relationship between the plantar posture angular velocity of the plantar plate of the robot and the motor angular velocity of the parallel joint can be expressed. The above description is only based on a closed chain structure where the motor M1 is located. The same is true for another closed chain structure where the motor M2 is located. Referring to the above process, the relationship between the plantar posture angular velocity of another closed chain structure and the motor angular velocity of the parallel joint can be expressed.

[0187] In S510, the desired attitude angular velocity Substituting into formula (15), the corresponding expected motor angular velocity can be calculated

[0188] S520: Determine the motor feedforward torque corresponding to the first control torque based on the Jacobian matrix of the plantar plate and the parallel joint of the robot.

[0189] In the embodiment of the present disclosure, in combination with the above formula (3), it can be known that when the desired motor angle is determined by the process S510, and the desired motor angular velocity After that, it is necessary to solve the motor feedforward torque Motor feedforward torque According to the first control torque τ roll , τ pitch Calculated.

[0190] Specifically, consider Figure 1 In the parallel mechanism shown, the mass of the first vertical rod 210 is often very light and the movement speed is not high, so the gravity, inertia force, etc. of the first vertical rod 210 can be ignored during the movement. At this time, the mapping relationship from the first control torque of the plantar plate to the motor feedforward torque of the parallel joint can be expressed by the Jacobian matrix:

[0191]

[0192] In formula (16), is the motor feedforward torque, τ roll , τ pitch represents the first control torque, J T Represents the transpose of the Jacobian matrix. The calculation process of the Jacobian matrix J is as follows:

[0193] Expanding and simplifying formula (12) yields

[0194]

[0195] In formula (12)

[0196] m1=-x c′ sinθ pitch +y c′ cosθ pitch sinθ roll +z c′ cosθ pitch cosθ roll

[0197] m2=y c′ cosθ roll sinθ pitch -z c′ sinθ pitch sinθ roll

[0198] n1=-(z c′ cosθ roll +y c′ sinθ roll )

[0199] p1=-x c′ cosθ pitch -y c′ sinθ pitch sinθ roll-z c′ cosθ roll sinθ pitch

[0200] p2=y c′ cosθ pitch cosθ roll -z c′ cosθ pitch sinθ roll

[0201] The components of the Jacobian matrix can be extracted from formula (12), expressed as:

[0202]

[0203] Similarly, the above description is based on the closed-chain mechanism on one side where the motor M1 is located. Similarly, the closed-chain mechanism on the other side where the motor M3 is located can be calculated to obtain the component J of the remaining Jacobian matrix: 21 ,J 22 According to the definition of Jacobian matrix, it can be expressed as:

[0204]

[0205] Substitute formula (19) into formula (16), and combine it with the first control torque τ obtained by the above calculation roll , τ pitc h, the motor feedforward torque can be calculated

[0206] S530, performing closed-loop control according to the difference between the desired motor angle and the motor angle, the difference between the desired motor angular velocity and the motor angular velocity, and the motor feedforward torque to obtain a second control torque.

[0207] In the embodiment of the present disclosure, combined with the closed-loop control equation of the above formula (3), through the above S510-S520 method process, the desired motor angle in formula (3) is Expected motor angular velocity And the motor feedforward torque All are obtained by calculation, so the right side of the equal sign of formula (3) are all known quantities, so the second control torques τ1 and τ2 can be calculated. The second control torques τ1 and τ2 represent the control torques of motors M1 and M2.

[0208] Therefore, in the embodiment of the present disclosure, after obtaining the second control torques τ1 and τ2, the parallel joint motor can be controlled to output a corresponding torque according to the second control torques τ1 and τ2, thereby realizing robot control.

[0209] From the above, it can be seen that in the embodiment of the present disclosure, the plantar posture of the robot is estimated through a parallel motion model, which greatly simplifies the algorithm process and data volume compared to the traditional numerical iterative algorithm, and improves the control efficiency of the parallel robot. Moreover, based on the parallel leg structure, the inertia of the robot legs is reduced, the stiffness of the robot legs is improved, and the robot has better dynamic performance. Furthermore, by adopting torque-level control for the robot, the robot's motion stability and motion accuracy are improved. In addition, an analytical expression of the Jacobian matrix of the robot's ankle joint and the parallel joint is provided, which can effectively map the plantar posture control torque to the control torque of the parallel joint motor, and realize the dynamic control of the parallel mechanism.

[0210] like Figure 7 As shown, in some embodiments, the present disclosure provides a control device for a robot, including:

[0211] The information acquisition module 10 is configured to acquire the motion information of the motors at the parallel joints when the robot moves;

[0212] The motion estimation module 20 is configured to input the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model;

[0213] A first torque module 30 is configured to perform closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate;

[0214] A second torque module 40 is configured to perform closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain a second control torque of the parallel joint;

[0215] The motion control module 50 is configured to control the operation of the motor at the parallel joint according to the second control torque.

[0216] In some embodiments, the motion information includes a motor angle and a motor angular velocity, and the plantar posture information includes a plantar posture angle and a plantar posture angular velocity; the motion estimation module 20 is configured as follows:

[0217] Inputting the motor angle into a pre-trained parallel motion model to obtain the plantar posture angle of the robot output by the parallel motion model;

[0218] Based on the Jacobian matrix of the plantar plate of the robot and the parallel joint, the plantar posture angular velocity corresponding to the motor angular velocity is determined.

[0219] In some embodiments, the plantar posture information includes a plantar posture angle and a plantar posture angular velocity, and the desired posture information includes an expected posture angle, an expected posture angular velocity, and an expected feedforward torque of the plantar plate; the first torque module 30 is configured as follows:

[0220] The first control torque is obtained by performing closed-loop control according to the difference between the plantar posture angle and the expected posture angle, the difference between the plantar posture angular velocity and the expected posture angular velocity, and the expected feedforward torque.

[0221] In some embodiments, the motion information includes a motor angle and a motor angular velocity, the desired posture information includes an expected posture angle and an expected posture angular velocity of the plantar plate, and the expected motion information includes an expected motor angle and an expected motor angular velocity of the parallel joint; the second torque module 40 is configured as follows:

[0222] Determine the desired motor angle based on the desired posture angle, and determine the desired motor angular velocity based on the desired posture angular velocity;

[0223] Determine a motor feedforward torque corresponding to the first control torque based on a Jacobian matrix of the plantar plate of the robot and the parallel joint;

[0224] The second control torque is obtained by performing closed-loop control according to the difference between the desired motor angle and the motor angle, the difference between the desired motor angular velocity and the motor angular velocity, and the motor feedforward torque.

[0225] In some embodiments, the second torque module 40 is configured as follows:

[0226] Determine a first equation between the desired attitude angle and the desired motor angle, and a second equation between the desired attitude angular velocity and the desired motor angular velocity based on an inverse kinematics algorithm;

[0227] The corresponding expected motor angle is determined based on the expected posture angle and the first equation, and the corresponding expected motor angular velocity is determined based on the expected posture angular velocity and the second equation.

[0228] In some embodiments, the motion estimation module 20 is configured to:

[0229] Acquire a training data set, wherein each sample data in the training data set includes sample motion information of the motor at the parallel joint of the robot, and reference posture information of the plantar plate of the robot corresponding to the sample motion information;

[0230] Inputting the sample motion information into the parallel motion model to be trained, and obtaining the plantar posture information of the plantar plate of the robot output by the parallel motion model;

[0231] According to the difference between the plantar posture information and the reference posture information, the model parameters of the parallel motion model are adjusted until a convergence condition is met, thereby obtaining a trained parallel motion model.

[0232] In some embodiments, the parallel joint includes a knee joint motor and a connecting rod mechanism of the robot, and the knee joint motor drives the plantar plate of the robot to move through the connecting rod mechanism.

[0233] From the above, it can be seen that in the embodiment of the present disclosure, the plantar posture of the robot is estimated through a parallel motion model, which greatly simplifies the algorithm process and data volume compared to the traditional numerical iterative algorithm, and improves the control efficiency of the parallel robot. Moreover, based on the parallel leg structure, the inertia of the robot legs is reduced, the stiffness of the robot legs is improved, and the robot has better dynamic performance. Furthermore, by adopting torque-level control for the robot, the robot's motion stability and motion accuracy are improved. In addition, an analytical expression of the Jacobian matrix of the robot's ankle joint and the parallel joint is provided, which can effectively map the plantar posture control torque to the control torque of the parallel joint motor, and realize the dynamic control of the parallel mechanism.

[0234] In some embodiments, the present disclosure provides a robot comprising:

[0235] A plantar plate and a parallel joint, wherein the plantar plate is connected to a motor of the parallel joint via a connecting rod mechanism, and the motor drives the plantar plate to move via the connecting rod mechanism;

[0236] The controller includes a processor and a memory, wherein the memory stores computer instructions, and the computer instructions are used to enable the processor to execute the method described in any of the above embodiments.

[0237] In some embodiments, the present disclosure provides a storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the method described in any of the aforementioned embodiments.

[0238] Specifically, Figure 8 A schematic diagram of the structure of a robot 600 suitable for implementing the method disclosed in the present invention is shown. Figure 8 The robot shown can realize the corresponding functions of the above-mentioned processor and storage medium.

[0239] like Figure 8As shown, the robot 600 includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a memory 602 or a program loaded from a storage part 608 into the memory 602. In the memory 602, various programs and data required for the operation of the robot 600 are also stored. The processor 601 and the memory 602 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0240] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage section 608 as needed.

[0241] In particular, according to an embodiment of the present disclosure, the above method process can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 609, and / or installed from a removable medium 611.

[0242] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and a module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0243] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the embodiments. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A robot control method, characterized in that: include: Obtain the motion information of the motors at the parallel joints when the robot moves; Inputting the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model; Performing closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate; Performing closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain a second control torque of the parallel joint; The operation of the motor at the parallel joint is controlled according to the second control torque.

2. The method according to claim 1, characterized in that The motion information includes a motor angle and a motor angular velocity, and the plantar posture information includes a plantar posture angle and a plantar posture angular velocity; The step of inputting the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model comprises: Inputting the motor angle into a pre-trained parallel motion model to obtain the plantar posture angle of the robot output by the parallel motion model; Based on the Jacobian matrix of the plantar plate of the robot and the parallel joint, the plantar posture angular velocity corresponding to the motor angular velocity is determined.

3. The method according to claim 1, characterized in that The plantar posture information includes a plantar posture angle and a plantar posture angular velocity, and the desired posture information includes an desired posture angle, an desired posture angular velocity, and an desired feedforward torque of the plantar plate; The step of performing closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate includes: The first control torque is obtained by performing closed-loop control according to the difference between the plantar posture angle and the expected posture angle, the difference between the plantar posture angular velocity and the expected posture angular velocity, and the expected feedforward torque.

4. The method according to claim 1, characterized in that: The motion information includes a motor angle and a motor angular velocity, the expected posture information includes an expected posture angle and an expected posture angular velocity of the sole plate, and the expected motion information includes an expected motor angle and an expected motor angular velocity of the parallel joint; The method of performing closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain the second control torque of the parallel joint includes: Determine the desired motor angle based on the desired posture angle, and determine the desired motor angular velocity based on the desired posture angular velocity; Determine a motor feedforward torque corresponding to the first control torque based on a Jacobian matrix of the plantar plate of the robot and the parallel joint; The second control torque is obtained by performing closed-loop control according to the difference between the desired motor angle and the motor angle, the difference between the desired motor angular velocity and the motor angular velocity, and the motor feedforward torque.

5. The method according to claim 4, characterized in that The step of determining the expected motor angle based on the expected posture angle and determining the expected motor angular velocity based on the expected posture angular velocity comprises: Determine a first equation between the desired attitude angle and the desired motor angle, and a second equation between the desired attitude angular velocity and the desired motor angular velocity based on an inverse kinematics algorithm; The corresponding expected motor angle is determined based on the expected posture angle and the first equation, and the corresponding expected motor angular velocity is determined based on the expected posture angular velocity and the second equation.

6. The method according to any one of claims 1 to 5, characterized in that: The training process of the parallel kinematic model includes: Acquire a training data set, wherein the sample data in the training data set includes sample motion information of the motor at the parallel joint of the robot, and reference posture information of the plantar plate of the robot corresponding to the sample motion information; Inputting the sample motion information into the parallel motion model to be trained, and obtaining the plantar posture information of the plantar plate of the robot output by the parallel motion model; According to the difference between the plantar posture information and the reference posture information, the model parameters of the parallel motion model are adjusted until a convergence condition is met, thereby obtaining a trained parallel motion model.

7. The method according to claim 1, characterized in that The parallel joint includes a knee joint motor and a connecting rod mechanism of the robot, and the knee joint motor drives the plantar plate of the robot to move through the connecting rod mechanism.

8. A robot control device, characterized in that: include: An information acquisition module is configured to acquire motion information of motors at parallel joints when the robot moves; A motion estimation module, configured to input the motion information into a pre-trained parallel motion model to obtain the plantar posture information of the robot output by the parallel motion model; A first torque module is configured to perform closed-loop control according to the sole posture information and the desired posture information of the sole plate of the robot to obtain a first control torque of the sole plate; A second torque module is configured to perform closed-loop control based on the first control torque of the plantar plate and the desired motion information of the parallel joint to obtain a second control torque of the parallel joint; The motion control module is configured to control the operation of the motor at the parallel joint according to the second control torque.

9. A robot, characterized in that: include: A plantar plate and a parallel joint, wherein the plantar plate is connected to a motor of the parallel joint via a connecting rod mechanism, and the motor drives the plantar plate to move via the connecting rod mechanism; A controller comprises a processor and a memory, wherein the memory stores computer instructions, and the computer instructions are used to enable the processor to execute the method according to any one of claims 1 to 7.

10. A storage medium, characterized in that: Computer instructions are stored, and the computer instructions are used to cause a processor of a robot to execute the method according to any one of claims 1 to 7.

Images