A safety protection method and system for a legged robot

Abstract

The application discloses a kind of foot type robot safety protection method and system, comprising: obtaining the desired input of robot body state control, and the state information of robot body is estimated;According to the current posture of robot, the state information of robot body estimated and current environmental information, according to priority, the output instruction of robot body state control is corrected according to the corresponding protection rule;While based on the weight calculation learning reward information of the protection rule called;Based on the desired input of robot body state control, the state information of robot body estimated and learning reward information, the fall risk of current robot is predicted using trained reinforcement learning neural network, and the desired input is corrected based on the prediction result.The application is based on the feedback control strategy that environmental perception protection control and reinforcement learning are combined, realizes the prediction and protection control of robot fall risk.

Description

Technical Field

[0001] This invention relates to the field of robot safety protection technology, and in particular to a safety protection method and system for legged robots. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Legged robots can walk in uneven and unstructured environments, and have better terrain adaptability and mobility compared to wheeled and tracked robots.

[0004] Existing control processes for legged robots, such as Figure 1 As shown, by inputting desired inputs (torso pose, gait, foot trajectory, foot landing point, etc.) into the controller, the controller, combined with the state estimator based on the current robot torso posture, leg joint motor position / velocity / torque data to obtain the robot's three-dimensional spatial posture, velocity, and whether the feet are in contact with the ground, uses control algorithms such as Virtual Model Control (VMC) and Model Predictive Control (MPC) to calculate the robot's leg joint motor outputs, which are then executed by the respective joint motor drivers. This achieves tracking of the desired inputs and enables movement under complex ground conditions. When the robot's movement is abnormal, a manual emergency stop input cuts off the motor drivers' control of the joint motors, thus protecting the robot.

[0005] However, current legged robots still face the following technical challenges in motion control:

[0006] (1) At present, the control system of legged robots does not consider the influence of environmental factors on the control process. Due to the complexity of the environment, when the robot faces continuous irregular steps or terrain, it is easy to have problems such as the foot slipping and standing up on the slope, which poses a great risk of falling and overturning, affecting the safety of the robot body and the equipment installed on it.

[0007] (2) Legged robots usually need to operate in different outdoor environments and face various adverse factors such as variable climate, strong electromagnetic interference, and mechanical vibration. These adverse factors also seriously affect the safe and stable operation of legged robots. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes a safety protection method and system for legged robots. By combining body protection based on environmental information and system protection based on mechanical and electrical systems, it is possible to predict the risk of fall during the operation of the legged robot and to provide safety protection for the robot body.

[0009] In some implementations, the following technical solutions are adopted:

[0010] A safety protection method for legged robots includes:

[0011] The desired input for robot body state control is obtained, and the robot body state information is estimated by combining the robot's current posture and leg joint motor data.

[0012] Based on the robot's current posture, the estimated state information of the robot body, and the current environmental information, the corresponding protection rules are invoked according to priority to correct the output instructions for the robot body's state control; at the same time, learning reward information is calculated based on the weights of the invoked protection rules.

[0013] Based on the expected input for robot body state control, the estimated state information of the robot body, and the learning reward information, a trained reinforcement learning neural network is used to predict the current risk of the robot falling, and the expected input is corrected based on the prediction results.

[0014] As a further solution, it also includes: controlling the robot to execute a pre-set fall righting action after it has fallen or when it is in a fall posture.

[0015] As a further approach, the reinforcement learning neural network adopts a transfer learning training strategy. First, the robot's dynamics are modeled in a simulation environment to obtain feedback information from various sensors under different simulation environments, thereby pre-training the network parameters. The parameters trained in the simulation environment are then loaded into the actual robot, and online learning is performed using data collected from the robot's real operating environment.

[0016] As a further solution, the robot collects current environmental images and point cloud information, extracts the environmental features of the robot's current location, classifies the environmental features, and obtains the robot's current environmental information.

[0017] As a further solution, it also includes: pre-setting the fall protection posture of the robot gimbal and robotic arm, and when it is predicted that the robot is at risk of falling, controlling the robot gimbal and robotic arm to move to the fall protection posture.

[0018] As a further option, it also includes: executing a soft shutdown procedure in response to a robot soft shutdown command.

[0019] As a further solution, the robot can automatically turn on the power to the sensors that need to work during the task execution, and automatically turn off the power to the sensors that do not need to work after the task is completed.

[0020] In other embodiments, the following technical solutions are adopted:

[0021] A safety protection system for a legged robot, comprising:

[0022] The robot state estimation module is used to obtain the expected input for robot body state control, and estimate the robot body state information by combining the robot's current posture and leg joint motor data.

[0023] The robot control command correction module is used to correct the output commands of the robot's state control by calling the corresponding protection rules according to priority based on the robot's current posture, the estimated state information of the robot body, and the current environment information; at the same time, it calculates learning reward information based on the weight of the protection rules.

[0024] The robot fall prediction module is used to predict the robot's current fall risk based on the expected input for robot body state control, the estimated state information of the robot body, and the learning reward information, using a trained reinforcement learning neural network, and to correct the expected input based on the prediction results.

[0025] As a further solution, it also includes: a robot fall protection module, used to pre-set the fall protection posture of the robot gimbal and robotic arm, and when it is predicted that the robot is at risk of falling, control the robot gimbal and robotic arm to move to the fall protection posture.

[0026] As a further embodiment, a fall-prevention buffer device is also included on the housing of the robot's leg unit, the fall-prevention buffer device including a fall-prevention bracket and a flexible fall-prevention ball disposed thereon.

[0027] As a further embodiment, a mechanical cavity is also included, the mechanical cavity comprising a bottom cavity and a mounting cover, wherein a sealing strip is provided at the position where the mounting cover contacts the bottom cavity, and a sealing plate is provided outside the sealing strip.

[0028] As a further solution, it also includes a soft shutdown module, which is used to execute the soft shutdown process after receiving a soft shutdown command.

[0029] As a further solution, it also includes a power control module, which automatically turns on the power to the sensors that need to work during the execution of the robot task, and automatically turns off the power to the sensors that do not need to work after the task is completed.

[0030] In other embodiments, the following technical solutions are adopted:

[0031] A legged robot includes: employing the above-described legged robot safety protection method; or, including the above-described legged robot safety protection system.

[0032] In other embodiments, the following technical solutions are adopted:

[0033] A terminal device includes a processor and a memory, wherein the processor is used to implement various instructions; and the memory is used to store multiple instructions, which are adapted to be loaded and executed by the processor to implement the above-described safety protection method for legged robots.

[0034] In other embodiments, the following technical solutions are adopted:

[0035] A computer-readable storage medium storing a plurality of instructions adapted to be loaded and executed by a processor of a terminal device for the above-described safety protection method for legged robots.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] (1) This invention proposes an all-round safety protection control method for legged robots and develops a feedback control algorithm based on environmental perception protection control and reinforcement learning. By dynamically correcting the input and output control commands of the robot, the prediction and protection control of the robot's fall risk is realized, which solves the problem that the robot is prone to fall when facing continuous irregular steps or terrain, and improves the stability and safety of the robot in complex environments.

[0038] (2) This invention proposes a method for protecting legged robots from falling. A fall protection buffer device is developed. The fall protection posture of the robot gimbal and the robotic arm is preset. When the risk of falling is predicted, the robot gimbal and the robotic arm are controlled to move to the fall protection posture to protect the robot body from falling. This reduces the damage to the robot's own equipment caused by falling, extends the service life of the equipment, and ensures the robot's adaptability to complex outdoor environments.

[0039] (3) This invention adds a robot soft shutdown control process and power management strategy to avoid the risk of hardware damage caused by hard shutdown, improve the reliability of robot operation, and effectively extend the robot's battery life and equipment lifespan.

[0040] (4) This invention proposes a rigid-flexible coupling collision protection method for legged robots, designs a robot protective sealed shell, realizes waterproof and dustproof robot body and overall electromagnetic protection; and sets an anti-fall buffer device in the leg unit, without the need for special modification of the joint of the leg motion module, the anti-fall buffer device can be easily installed on the leg motion module, thus realizing protection and buffering when falling.

[0041] Other features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0042] Figure 1This is a schematic diagram of the control process of a legged robot in the existing technology;

[0043] Figure 2 This is a control flowchart of the safety protection method for legged robots in an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the legged robot structure in an embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of the mechanical cavity structure in an embodiment of the present invention. Detailed Implementation

[0046] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0048] Example 1

[0049] In one or more embodiments, a safety protection method for legged robots is disclosed, combining... Figure 2 Specifically, it includes the following processes:

[0050] (1) Obtain the desired input for robot body state control, and estimate the robot body state information by combining the robot's current posture and leg joint motor data;

[0051] In this embodiment, the expected inputs for robot body state control include: expected input quantities such as torso pose, gait, foot trajectory and foot landing point; combined with the state estimator based on the current robot torso posture, leg joint motor position / velocity / torque data, etc., the state estimator estimates the robot body's three-dimensional spatial posture, velocity, whether the foot is in contact with the ground and other state information.

[0052] (2) Based on the robot's current posture, the estimated state information of the robot body, and the current environment information, the corresponding protection rules are called according to priority to correct the output instructions of the robot body state control; at the same time, the learning reward information is calculated based on the weight of the protection rules.

[0053] In this embodiment, priority protection rules for different environments are manually configured, including (but not limited to): abnormal threshold of posture sensor data when falling in different environments, threshold of foot force output of controller in different environments, safe gait to be executed when the robot falls in different environments, etc., and a rule calling priority matrix is ​​obtained based on the robot's current posture and the estimated state information of the robot body.

[0054] The rule matrix consists of environmental features and protection rules corresponding to robot state anomaly detection. Assume the identified N environmental features are represented as a set E = {e...} i}, i=1…N(e i The set of M protection rules for the robot, representing the identified environmental features, is defined as A = {a}. j},j=1…M(a j (representing protection rules), combined with the robot's state (body posture, velocity, angular velocity, etc.) corresponding to the anomaly detection threshold set Z = {z k},k=1…K(z k B = {b} represents the K protection rules corresponding to the threshold of the corresponding state. k},k=1…K(b k (Indicates protection rules)

[0055] The rule priority matrix W is denoted as , where w xiyj This indicates the priority weight value.

[0056]

[0057] Then, by combining the robot's current posture Z, the estimated state S of the robot body, and environmental features e, the rules within the matrix are activated. By judging the execution priority of the rules, on the one hand, the highest priority rule is used to correct the controller output to achieve a rapid response to the robot's fall (to quickly return the leg joints to a safe position); on the other hand, the average weight of the activated (called) rules is used as the learning reward to achieve dynamic correction of the robot's expected gait during training of the reinforcement learning module and runtime.

[0058] The process of acquiring current environmental information is as follows: using laser, vision and other sensors installed on the robot, information such as current environmental images and point clouds is collected. Through feature recognition and classification algorithms, the environmental features (surface texture, normal vector, elevation, etc.) of the current robot are extracted to classify the robot's current environmental features for dynamic invocation of internal preset protection rules.

[0059] (3) Based on the expected input for robot body state control, the estimated state information of the robot body and the learning reward information, the trained reinforcement learning neural network is used to predict the current risk of robot falling, and the expected input is corrected based on the prediction results.

[0060] In this embodiment, the robot controller's expected input, estimated state information of the robot body, and learning reward information generated by the protection control module are used to dynamically assess the current robot fall risk through the reinforcement learning neural network inside the module. By real-time correction of the expected input, the robot body and environmental influencing factors are pre-incorporated into the robot body control process, improving the predictability and adaptability of the robot body control to external disturbances, and reducing the robot fall risk from the source of control.

[0061] To achieve efficient and rapid training of reinforcement learning neural networks, a transfer learning training strategy was adopted. First, the robot's dynamics were modeled in a simulation environment. By loading different ground environments and collecting feedback information from various sensors in the simulation environment, the network parameters were pre-trained. Then, the parameters trained in the simulation environment were loaded into the actual robot, and online learning was performed using data collected from the robot's real operating environment to achieve environmental adaptive tuning of the network parameters, ensuring that the robot is well adapted to the current operating environment.

[0062] As an alternative implementation, after the robot falls, the learning network can be trained to output control commands based on the robot's state, thereby achieving automatic righting control after the fall.

[0063] As an optional implementation, fall protection postures for the robot gimbal and robotic arm are pre-set. When a fall risk is predicted, the robot gimbal and robotic arm are controlled to move to the fall protection posture. For example, when a robot falls, it usually rolls sideways. In the fall protection posture, the gimbal rotates back to face forward, and the robotic arm curls up along the longitudinal axis of the robot's body to reduce the contact area and impact.

[0064] As an optional implementation, this embodiment addresses the problem that the commonly used power-off shutdown (hard shutdown) of robots may damage the control system's storage devices, thereby causing the robot control system to malfunction. The safety protection method in this embodiment responds to the robot's soft shutdown command and executes a soft shutdown procedure. By adding a soft shutdown procedure, the risk of hardware damage caused by hard shutdown can be avoided, improving the reliability of robot operation.

[0065] As an optional implementation method, in order to effectively extend the robot's battery life and equipment lifespan, the robot automatically turns on the power of the sensors that need to work during the task execution, and automatically turns off the power of the sensors that do not need to work after the task is completed.

[0066] This embodiment of the legged robot safety protection method adds a feedback control loop based on environmental perception protection control and reinforcement learning to the original robot control system. By dynamically correcting the input and output of the robot controller, it realizes the prediction and protection control of the robot's fall risk, thereby improving the stability and safety of the robot in complex environments.

[0067] Example 2

[0068] In one or more embodiments, a safety protection system for a legged robot is disclosed, specifically including:

[0069] The robot state estimation module is used to obtain the expected input for robot body state control, and estimate the robot body state information by combining the robot's current posture and leg joint motor data.

[0070] The robot control command correction module is used to correct the output commands of the robot's state control by calling the corresponding protection rules according to priority based on the robot's current posture, the estimated state information of the robot body, and the current environment information; at the same time, it calculates learning reward information based on the weight of the protection rules.

[0071] The robot fall prediction module is used to predict the robot's current fall risk based on the expected input for robot body state control, the estimated state information of the robot body, and the learning reward information, using a trained reinforcement learning neural network, and to correct the expected input based on the prediction results.

[0072] The specific implementation methods of the above modules have been described in detail in Example 1, and will not be repeated here.

[0073] As an optional implementation, the safety protection system in this embodiment features an integrated protective shell for mechanical protection, giving the robot excellent waterproof, dustproof, and electromagnetic interference resistance capabilities; specifically, combined with Figure 3The protective shell mainly consists of leg units and cavity units. The leg units primarily comprise joint motors, transmission mechanisms, and anti-fall buffer devices. Each joint motor itself possesses waterproof and electrical protection functions. The anti-fall buffer device is installed on the outer shell of the leg units. In this embodiment, the anti-fall buffer device consists of an anti-fall ball and an anti-fall bracket. The anti-fall ball is fixed to the anti-fall bracket and can be made of flexible cushioning materials such as rubber, polyurethane, or silicone. The anti-fall bracket can be made of materials such as aluminum alloy or steel. Installing the anti-fall buffer device on the outside of the leg units provides flexible cushioning when the robot falls, and can also serve as a handle to help the robot up after a fall and as an assistive handle during transport.

[0074] In addition, the mechanical cavity, robot motion control and power supply components are all installed inside the mechanical cavity, combined with Figure 4 The mechanical cavity consists of a bottom cavity, a mounting cover (including a protective bracket for the upper sensor), an external sealing plate, and a waterproof sealing strip. The bottom cavity is integrally machined, with the waterproof sealing strip installed on top of it. The opening provides access for the leg motion module and upper sensor components, as well as a connection point between the mounting cover and the bottom cavity. Waterproof silicone strips are installed on the top of the bottom cavity and the mounting cover, and then attached to the inside of the external sealing plate using techniques such as bonding or interference fitting. The external sealing plate can be fixed to the mounting cover and the bottom cavity with screws.

[0075] In this embodiment, the top cover can be designed as a cover plate to cover the top of the bottom cavity, changing the traditional installation position and method of the waterproof strip, improving the robot's waterproof performance, and solving problems such as aging and deformation of the waterproof strip that are detrimental to waterproofing.

[0076] As an optional implementation, the safety protection system in this embodiment addresses the problem that the current common practice of power-off shutdown (hard shutdown) of robots may damage the control system's storage devices, thereby causing the robot to malfunction. By adding a robot soft shutdown switch, once the robot detects and confirms that the switch is pressed, it will execute the software shutdown process, avoiding the risk of hardware damage caused by hard shutdown and improving the reliability of robot operation.

[0077] In addition, by adding peripheral power management functions to the robot, the sensor power can be automatically turned on during the task and automatically turned off after the task is completed, which can effectively extend the robot's battery life and equipment lifespan.

[0078] Example 3

[0079] In one or more embodiments, a terminal device is disclosed, including a server. The server includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the legged robot safety protection method of Embodiment 1. For the sake of brevity, further details are omitted here.

[0080] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0081] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0082] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0083] Example 4

[0084] In one or more embodiments, a computer-readable storage medium is disclosed, wherein a plurality of instructions are stored, the instructions being adapted to be loaded by a processor of a terminal device and executed to perform the legged robot safety protection method described in Embodiment 1.

[0085] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A safety protection method for a legged robot, characterized by, The method comprises: obtaining expected input for robot body state control, combining the current posture of the robot and the leg joint motor data to estimate the state information of the robot body; obtaining a rule priority matrix based on the current posture of the robot, the estimated state information of the robot body, the rule priority matrix giving the priority weight values corresponding to different protection rules under different environmental characteristics and robot state anomaly detection thresholds; according to the current posture of the robot, the state information of the robot body, and the current environmental information, the corresponding protection rules are called according to the priority, and the highest priority rule is used to correct the output instruction of the robot body state control, so as to realize the rapid response to the falling state of the robot; at the same time, the learning reward information is calculated based on the weight of the called protection rule, so as to realize the training of the reinforcement learning module and the dynamic correction of the expected gait of the robot in operation; based on the expected input, the state information of the robot body and the learning reward information, using the trained reinforcement learning neural network, predicting the falling risk of the current robot, and correcting the expected input based on the prediction result.

2. The safety protection method for a foot robot according to claim 1, wherein Further comprising: after detecting the falling of the robot or the falling posture, controlling the execution of the pre-set falling and turning over action.

3. The method of claim 1, wherein the step of detecting a hazard comprises: determining a distance between the robot and the hazard; and determining a velocity of the robot. The reinforcement learning neural network adopts a transfer learning training strategy, first performs dynamic modeling on the robot in a simulation environment, obtains feedback information of various sensors in different simulation environments, and realizes pre-training of network parameters; the parameters trained in the simulation environment are loaded to the actual robot, and online learning is performed through the data collected in the real running environment of the robot.

4. The method of claim 1, wherein the step of detecting a hazard comprises: detecting a hazard in the environment of the robot; and determining a safe path for the robot to travel to avoid the hazard. The robot collects current environmental images and point cloud information, extracts environmental characteristics of the robot, classifies the environmental characteristics, and obtains current environmental information of the robot.

5. The method of claim 1, wherein the step of detecting a hazard comprises: detecting a hazard in the environment of the robot; and determining a distance between the robot and the hazard. Further comprising: pre-setting the falling protection posture of the robot holder and the mechanical arm, and when it is predicted that the current robot has a falling risk, controlling the robot holder and the mechanical arm to move to the falling protection posture.

6. The method of claim 1, wherein the step of detecting a hazard comprises: detecting a hazard in the environment of the robot; and determining a distance between the robot and the hazard. Further comprising: in response to the instruction of soft shutdown of the robot, executing the soft shutdown process.

7. The method of claim 1, wherein the step of detecting a hazard comprises: detecting a hazard in the environment of the robot. Further comprising: the robot automatically turns on the power of the sensor needed to work during the task execution according to the different tasks, and automatically turns off the power of the sensor not needed to work after the task execution is completed.

8. A safety shield system for a foot-based robot, comprising: The method comprises: a robot state estimation module for obtaining expected input for robot body state control, combining the current posture of the robot and the leg joint motor data to estimate the state information of the robot body; The robot control instruction correction module obtains a rule priority matrix based on the current posture of the robot and the estimated state information of the robot body, the rule priority matrix giving corresponding priority weight values of different protection rules under different environmental characteristics and different state anomaly detection thresholds of the robot, for calling corresponding protection rules according to the current posture of the robot, the estimated state information of the robot body and the current environmental information in priority, and using the highest priority rule to correct the output instruction of the robot body state control, so as to realize fast response to the falling state of the robot; and learning reward information is calculated based on the weight of the protection rule, so as to realize training of the reinforcement learning module and dynamic correction of the expected gait of the robot during operation. The robot falling prediction module is configured to predict the falling risk of the current robot based on the expected input of the robot body state control, the estimated state information of the robot body and the learning reward information, and utilize the trained reinforcement learning neural network to correct the expected input based on the prediction result.

9. A safety guard system for a foot-based robot as claimed in claim 8, wherein, Further comprising: The robot falling protection module is configured to pre-set a falling protection posture of the robot holder and the robot arm, and control the robot holder and the robot arm to move to the falling protection posture when it is predicted that the current robot has a falling risk.

10. A safety guard system for a foot robot as claimed in claim 8, wherein, Further comprising a falling prevention buffer device arranged on the robot leg unit housing, the falling prevention buffer device comprising a falling prevention support and a flexible falling prevention ball arranged thereon.

11. A safety guard system for a foot robot as defined in claim 8, wherein Further comprising a mechanical cavity, the mechanical cavity comprising a bottom cavity and a mounting upper cover, a sealing rubber strip being arranged at a position where the mounting upper cover contacts the bottom cavity, and a sealing plate being arranged outside the sealing rubber strip.

12. A safety guard system for a foot robot as claimed in claim 8, wherein, Further comprising: The soft shutdown module is configured to execute a soft shutdown process after receiving a soft shutdown instruction.

13. A safety guard system for a foot robot as defined in claim 8, wherein Further comprising: The power control module is configured to automatically turn on the power of the sensor that needs to work during the execution of the robot task, and automatically turn off the power of the sensor that does not need to work after the execution of the task.

14. A legged robot characterized by comprising: The method comprises: The method comprises:

15. A terminal device comprising a processor and a memory, the processor configured to implement instructions; the memory configured to store a plurality of instructions, the terminal device characterized by, The instructions are adapted to be loaded and executed by the processor to perform the foot robot safety protection method according to any one of claims 1-7.

16. A computer-readable storage medium having stored therein a plurality of instructions, the instructions being executable by a processor to: The instructions are adapted to be loaded and executed by the processor of the terminal device to perform the foot robot safety protection method according to any one of claims 1-7. The instructions are adapted to be loaded and executed by the processor of the terminal device to perform the foot robot safety protection method according to any one of claims 1-7.

Images