A method and system for motion design and complex behavior generation of a quadruped robot

Through visual animation model and action function design, the problem of high professional threshold and low efficiency of action design and behavior synthesis of four-legged robots is solved, the flexibility of action design and the efficiency of behavior generation is realized, the action design process is simplified and the action design process is improved. The operational capabilities of four-legged robots in complex environments.

CN118114709BActive Publication Date: 2024-11-05SHANDONG GONGYAN ZHICHUANG EDUCATION TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410248990.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-05
Publication Date
2024-11-05
Estimated Expiration
2044-03-05

AI Technical Summary

Technical Problem

The movement design and behavior synthesis of four-legged robots in complex environments have problems of high professional requirements and low efficiency. The threshold for movement design and behavior synthesis is high, making it difficult to achieve flexible and efficient action generation.

Method used

The visual animation model is used for action design, and the first function and power function are selected as the trajectory parameter equations. The action design and optimization are carried out through the action function, the action library is established, and the behavioral trajectory data is combined in the timeline to control the four-legged robot.

Benefits of technology

It lowers the threshold for action design, improves the efficiency and flexibility of action design, realizes the efficiency and visualization of behavior generation, simplifies the action design process, and improves the stability and flexibility of sub-action switching.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118114709B_ABST
    Figure CN118114709B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of quadruped robot control, and provides a method and system for designing quadruped robot actions and generating complex behaviors. According to the information of the quadruped robot, an animation model of the quadruped robot is established; according to the animation model, action design is performed; trajectory data of the designed actions are extracted, and an action library is established with the goal of maximizing the cumulative reward obtained by the quadruped robot; the actions in the action library are combined under the time axis to generate behavior trajectory data; according to the generated behavior trajectory data, the quadruped robot is controlled. Based on the visualized animation model, action design is performed in the form of action function, and the action is more flexible and editable. The action can be changed by easily modifying the function, and the efficiency of action design is improved under the premise of realizing precise control of the action; at the same time, the behavior is generated by combining under the time axis, the combination time process is short, and the behavior generation efficiency is high.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of quadruped robots, and in particular relates to a method and system for designing quadruped robot movements and generating complex behaviors. Background Art

[0002] Compared with other types of legged robots, quadruped robots have better static stability and dynamic performance, and are a more economical and practical configuration.

[0003] The inventors found that when controlling a quadruped robot in a complex environment, in the process of motion generation and behavior synthesis, the existing methods mainly have problems such as high professional requirements for motion design and behavior synthesis, as well as low efficiency of motion design and behavior synthesis. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a method and system for motion design and complex behavior generation of a quadruped robot. The present invention is based on a visualized animation model, which lowers the entry threshold of motion design and improves the efficiency of motion design. Action design is performed in the form of action functions, which makes the actions more flexible and editable. The actions can be changed by easily modifying the functions, which further improves the efficiency of motion design while achieving precise control of the actions. At the same time, behaviors are generated by combining on the timeline, the combination time process is short, and the behavior generation efficiency is high.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0006] In a first aspect, the present invention provides a method for designing quadruped robot motions and generating complex behaviors, comprising:

[0007] According to the information of the quadruped robot, an animation model of the quadruped robot is established;

[0008] According to the animation model, action design is performed; wherein, a linear function is selected for the trajectory parameter equation about time in the horizontal direction, and a power function is selected for the trajectory parameter equation about time in the vertical direction;

[0009] Extract trajectory data of the designed actions, optimize them with the goal of maximizing the cumulative reward obtained by the quadruped robot, and establish an action library;

[0010] The actions in the action library are combined under the time axis to generate behavior trajectory data; different time periods are created in the time axis; each time period includes a start time and an end time, which are used to specify which actions in the action library are executed within the time period;

[0011] Control the quadruped robot based on the generated behavior trajectory data.

[0012] Furthermore, when establishing the animation model of the quadruped robot, the control bone positions at the set joints are determined according to the joint properties, and based on the root bone, the remaining bones are added to the root bone according to the kinematic relationship and the coordinate axis positions; the link origin position is set, the origin position of each link is traversed, and the robot body coordinate system position is determined in the base coordinate system and the construction is completed; then, the relevant data of the link is obtained, and the relevant data is constructed into the quadruped robot model to form a complete quadruped robot animation model with the joints.

[0013] Furthermore, the trajectory equation of the quadruped robot jumping is:

[0014]

[0015] Among them, x(t) is the trajectory parameter equation about time in the horizontal direction, t is the movement time; S is the forward distance; T is the movement period; y(t) is the trajectory parameter equation about time in the vertical direction; H is the leg lifting height; n is the power function order.

[0016] Furthermore, different interpolation methods are selected to simulate the movements of the quadruped robot under physical constraints. Bezier curve interpolation is used when performing irregular curve motion; projectile interpolation is used in the process from leaving the ground to the vertex; and free fall interpolation is used in the process from the vertex to touching the ground.

[0017] Furthermore, when designing walking motion, choose the interpolation of compound cycloid form:

[0018]

[0019]

[0020]

[0021] Where S is the stride; T m is the movement cycle; t is the movement time; H is the height of the leg lift; x is the horizontal displacement; y is the vertical displacement.

[0022] Furthermore, the position of the center of mass of the quadruped robot, the speed of the center of mass, the acceleration of the center of mass and the angle of the center of mass, as well as the position, speed, acceleration and ground contact status of the foot end are extracted; optimization is performed with the goal of maximizing the cumulative reward obtained by the quadruped robot, and an action library is established.

[0023] Furthermore, the actions in the action library are combined to generate behaviors under the timeline.

[0024] In a second aspect, the present invention further provides a quadruped robot motion design and complex behavior generation system, comprising:

[0025] The animation model building module is configured to: build an animation model of the quadruped robot according to the information of the quadruped robot;

[0026] The action design module is configured to: perform action design according to the animation model; wherein the trajectory parameter equation about time in the horizontal direction is in the form of a linear function, and the trajectory parameter equation about time in the vertical direction is in the form of a power function;

[0027] The action library building module is configured to: extract trajectory data of the designed action, optimize the action with the goal of maximizing the cumulative reward obtained by the quadruped robot, and build the action library;

[0028] The behavior generation module is configured to: generate behavior trajectory data after combining the actions in the action library under the time axis; create different time periods in the time axis; each time period includes a start time and an end time, which is used to specify which actions in the action library are executed within the time period;

[0029] The control module is configured to control the quadruped robot according to the generated behavior trajectory data.

[0030] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the quadruped robot motion design and complex behavior generation method described in the first aspect.

[0031] In a fourth aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the program, the steps of the quadruped robot motion design and complex behavior generation method described in the first aspect are implemented.

[0032] In a fifth aspect, the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the quadruped robot motion design and complex behavior generation method described in the first aspect are implemented.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. In the present invention, first, an animation model of a quadruped robot is established according to the information of the quadruped robot; then, action design is performed according to the animation model; wherein, a linear function is selected for the trajectory parameter equation about time in the horizontal direction, and a power function is selected for the trajectory parameter equation about time in the vertical direction; trajectory data of the designed actions are extracted, and an action library is established by optimizing the actions with the goal of maximizing the cumulative reward obtained by the quadruped robot; the actions in the action library are combined on the time axis to generate behavior trajectory data; finally, the quadruped robot is controlled according to the generated behavior trajectory data. Based on the visual animation model, the entry threshold of action design is lowered and the efficiency of action design is improved; the action design is carried out in the form of action function, which is more flexible and editable. The action can be changed by easily modifying the function. Under the premise of realizing precise control of the action, the efficiency of action design is further improved; at the same time, the combination of behaviors is generated under the timeline, the combination time process is short, and the behavior generation efficiency is high. In the timeline, users can create different time periods, each of which has a start time and an end time, which is used to specify which sub-actions should be executed during this time period. The introduction of time periods can improve the stability and flexibility of sub-action switching;

[0035] 2. Compared with the traditional quadruped robot motion design, the present invention adopts animation editing for motion design, visualizes the motion design process, lowers the entry threshold for motion design, and improves the efficiency of motion design; by writing a trajectory extraction script to extract the bone position, speed, acceleration, angle and other data corresponding to the corresponding planning point, the corresponding motion trajectory information is obtained, which is more flexible and convenient; the designed actions are combined on the timeline to generate behaviors. When combining, the combination time process is shorter, the behavior generation efficiency is higher, and the generated behavior can be visualized. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings in the specification that constitute a part of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments of this embodiment and their descriptions are used to explain this embodiment and do not constitute improper limitations on this embodiment.

[0037] Figure 1 This is a schematic diagram of the structure of a quadruped robot according to Example 1;

[0038] Figure 2 The overall flow chart of the quadruped robot motion design and complex behavior generation method of Example 1;

[0039] Figure 3 This is a parent-child relationship diagram of the skeleton of the quadruped robot animation model of Example 1;

[0040] Figure 4 This is a trajectory extraction flow chart of Example 1;

[0041] Figure 5 This is a flow chart of trajectory optimization using reinforcement learning for the quadruped robot of Example 1;

[0042] Figure 6 Generate a software page graph for the behavior of Example 1. DETAILED DESCRIPTION

[0043] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0044] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0045] Embodiment 1:

[0046] Compared with other types of legged robots, quadruped robots have better static stability and dynamic performance, and are a more economical and practical configuration. Quadruped robots are no longer simple industrial devices, but increasingly intelligent entities that can flexibly adapt to a variety of tasks. Their applications cover many fields, including industrial manufacturing, medical services, logistics distribution, and military applications. In order to complete complex tasks in complex scenarios, different actions need to be designed according to the specific tasks, and complex behaviors are formed in this way, which improves the ability of quadruped robots in collaboration, interaction, and execution of complex tasks.

[0047] Aiming at the problem of acquiring quadruped robot operation / mobility skill behaviors in complex environments, the motion design and complex behavior generation methods of quadruped robots based on motion design are studied. To achieve the motion design and complex behavior generation of quadruped robots, multiple issues need to be considered:

[0048] The first is the design of quadruped robot movements. The sub-movements formed by motion design are the basis for generating complex behaviors. The trajectory planning-based method can use the objective function to design the movements of the quadruped robot, but this method has great difficulties in ensuring the flexibility of the quadruped robot's movements during execution and the overall coordination of the synthesized behavior in the later stage. In addition, the quadruped robot has diverse movements, and the required planning algorithms will also be different, which is difficult and inefficient. The motion data-based method is used to design the movements, and the target tracking algorithm is used to obtain the motion data in the video, which can achieve more realistic and natural movements. However, this method requires more animal motion data sets, and the target tracking data results are not accurate enough. The motion capture-based method can obtain high-precision motion details, but this method requires professional equipment and specific hardware and software, and the cost is high. The reinforcement learning-based method can achieve nonlinear and highly dynamic movements, but the training time of this method is long, the required computing resources are large, and in the absence of prior knowledge, this method is not suitable for situations where rules or constraints are determined in advance.

[0049] The second is the acquisition of motion trajectories. After the motion design of the quadruped robot is completed, the foundation is laid for the subsequent behavior combination and prototype verification. However, due to the different types of quadruped robots selected, the kinematic models used are also different, so the extraction of different motion trajectory data is different. Not only that, when using sensors to obtain motion trajectories, the accuracy and error of the sensors is a key issue. The sensors may be affected by noise, environmental conditions, and hardware limitations, resulting in inaccurate measurement results. Moreover, in the application scenario, the action state at the next moment may be affected by the previous moment, such as the ground contact state, etc. Therefore, in the extraction of motion trajectories, it is necessary to pay attention to the continuation of the previous and next states, which is more difficult in the implementation process.

[0050] The third is action synthesis behavior. After designing several actions, different action synthesis behaviors can be selected according to different tasks. However, in the process of behavior generation, in order to make the robot's movement look natural and smooth, the synthesized behavior needs to ensure that the movement trajectory is smooth and continuous in the joint space. Avoiding sudden changes in joint angles and speeds is very important for the stability and execution effect of the robot. When generating behaviors, it is necessary to consider the contact between the foot end and the ground. This part involves the contact control of the foot end, which requires data fusion of the foot end sensor, which is relatively complicated. When the quadruped robot moves, it is necessary to consider the coordinated movement of multiple foot ends to ensure the overall balance and stability of the robot. The synthetic behavior needs to consider the collaborative work between the foot ends to avoid conflicts and collisions, but this process needs to take into account the position changes of several feet, which can only be observed after all actions are combined through simulation experiments, and cannot be observed in advance.

[0051] In view of the high professional requirements for action design and behavior synthesis, as well as the low efficiency of action design and behavior synthesis, this embodiment provides a method for action design and complex behavior generation of a quadruped robot, including:

[0052] According to the information of the quadruped robot, an animation model of the quadruped robot is established;

[0053] According to the animation model, action design is performed; wherein, a linear function is selected for the trajectory parameter equation about time in the horizontal direction, and a power function is selected for the trajectory parameter equation about time in the vertical direction;

[0054] Extract trajectory data of the designed actions, optimize them with the goal of maximizing the cumulative reward obtained by the quadruped robot, and establish an action library;

[0055] The actions in the action library are combined under the time axis to generate behavior trajectory data; different time periods are created in the time axis; each time period includes a start time and an end time, which are used to specify which actions in the action library are executed within the time period;

[0056] Control the quadruped robot based on the generated behavior trajectory data.

[0057] Through the robot kinematic relationship, a robot three-dimensional model is constructed in the animation editing software, the coordinate systems of the robot's torso, legs and arms are defined, and the action design of the quadruped robot is performed. Compared with action design methods such as trajectory planning, target tracking, motion capture and reinforcement learning, the method in this embodiment can visualize and simplify the action design process, improve the efficiency of action design, and use a program script to write a coordinate point extraction plug-in to extract the trajectory coordinates of the corresponding planning points of the quadruped robot, and write a behavior generation system through the host computer. The system has the functions of storing a number of action trajectory data, generating behaviors by combining actions, and previewing the generated behaviors.

[0058] This embodiment uses an animation editing method to design quadruped robot motions, visualizes and facilitates the motion design process, extracts trajectory data of the quadruped robot, and encapsulates the motion trajectory data after trajectory optimization. Then, the quadruped robot motions are combined and generated under the timeline, which can realize the generation of complex behaviors.

[0059] like Figure 1 As shown, optionally, the quadruped robot in this embodiment has a front-elbow and rear-elbow leg structure, each leg has 3 degrees of freedom, and also includes at least one 6-degree-of-freedom robotic arm.

[0060] The method in this embodiment is optional. First, animation editing software is used for action design to visualize the action design process, thereby lowering the entry threshold for action design and improving the efficiency of action design. Secondly, by writing a trajectory extraction script, the bone position, speed, acceleration, angle and other data corresponding to the corresponding planning point are extracted to obtain the corresponding action trajectory information. The acquisition of trajectory data is flexible and convenient. Finally, the actions are combined on the timeline to generate behaviors. When combining, the combination time is short, the behavior generation efficiency is high, and the generated behaviors can be visualized. Figure 2 As shown, the specific steps of the method in this embodiment are as follows:

[0061] S1. Build animation model:

[0062] S1.1. Modeling the quadruped robot. Optionally, use 3D design software SolidWorks to design the components of the quadruped robot, create STL model files of the components, and assemble them into a complete quadruped robot.

[0063] S1.2. Convert the created STL model file into a URDF (Universal Robot Description Format) file. Optionally, after installing the corresponding conversion plug-in in SolidWorks, set the robot body coordinate system (base_link), modify the link in the design model to each separate link, modify the joint, and set the rotation range of the rotation joint to make it consistent with the actual lower rotation angle constraint.

[0064] S1.3. Display the exported URDF file in rviz. Optionally, in ROS-melodic under Ubuntu 18.04, write a launch file and start rviz to display the quadruped robot model. rviz is an open source tool in ROS, mainly used for robot visualization. It can display and debug the robot's sensor data, status information, and motion planning.

[0065] S1.4. Convert the generated URDF model into an fbx animation model. Optionally, write a conversion plug-in through a python script and import the plug-in into blender to complete the model conversion.

[0066] Optionally, the plug-in is written in Python and integrates Blender's Python API. It is mainly divided into three parts: robot model construction, importing urdf files, and exporting fbx files.

[0067] In summary, for the robot model construction part, the robot model is mainly built in the animation editing software. Specifically, first, the robot model main class is constructed, including functions such as cleaning, parsing and optimizing the URDF file; a series of model parameters are defined for building the robot animation model, including link attributes, joint attributes, coordinate axis attributes, material attributes and bone attributes. Then, the positions of the leg, torso and arm coordinate systems are determined, and the relative relationship of the coordinate system suitable for the action editing software is established. According to the URDF file, the root part of the model is found, the root bone root.bone bone is defined to control the overall model movement and rotation, and root.bone is added to the scene of the animation editing software blender. According to the joint attribute part in the URDF file, the control bone position at the set joint is determined, and based on root.bone, the remaining bone bones are added to root.bone according to the kinematic relationship and coordinate axis position. Set the link origin position, traverse the origin position of each link, and determine the base_link position in the base coordinate system and complete the construction. Then, obtain the relevant data of the link, such as mesh name, file path, position, rotation, scale and material, and build it into blender to form a complete link-joint model with the joint, and display the quadruped robot model in blender. Finally, through the position relationship of the joint, link the constructed skeleton model with the entity model, set the corresponding control bones in the specific position, and obtain the control link range of the joint in the URDF to let the skeleton control the corresponding link, realize the skeleton control entity link and skinning operation, and finally build the model.

[0068] In this embodiment, the main function of importing URDF files is to allow users to import URDF files through blender, and provide some options to control the merging, renaming, application of weld modifiers and whether to use unique texture names.

[0069] In this embodiment, the main function of exporting fbx files is to provide auxiliary tools for export operations, and to select whether to perform 3D modeling and rendering functions for the quadruped robot model, and to add the option of storing temporary textures, allowing users to select the exported fbx files in blender while retaining the texture information.

[0070] S2. Action design:

[0071] Compared with the action design based on trajectory planning, motion data, motion capture and reinforcement learning, in this embodiment, the action design based on animation editing software is more convenient and has a lower threshold, and is more convenient for the design of complex actions.

[0072] Blender uses animation to achieve the action design and motion control of the quadruped robot. Through skinning operations, the entity and the skeleton are bound one-to-one, and a complete skeleton is formed by setting the parent-child relationship to control the robot entity. The parent-child relationship of the skeleton is as follows: Figure 3 As shown. The skeleton that matches the physical quadruped robot model is mainly divided into three parts: trunk control bones, manipulator arm control bones and foot control bones. The trunk control bones can be rotated and moved to realize the rotation and movement of the robot as a whole, and the center of mass motion trajectory data of the quadruped robot can be obtained; the manipulator arm control bones include base bones, six control bones and end bones. The base bones cannot be moved or rotated. The six control bones represent that the manipulator has six degrees of freedom and can be rotated to realize the movement of the manipulator. The end bones of the manipulator are mainly used to obtain the motion trajectory and control the whole manipulator after the introduction of inverse kinematics; the foot control bones include shoulder joint lateral swing bones, thigh joint pitch bones, calf joint pitch bones and foot end bones, etc. Among them, the movement of the foot can be controlled by rotating the shoulder joint bones, thigh joint bones and calf joint bones. The foot end bones are mainly used to obtain the foot motion trajectory and control the whole leg after the introduction of inverse kinematics control. The successfully constructed animation model is imported into the animation editing software Blender, and the movement of the associated entity, i.e. the quadruped robot model, is controlled by moving and rotating the control bones in the set skeleton, thus finally realizing the action design.

[0073] S2.1. When designing actions, use Python to write action functions in the script in blender to design actions.

[0074] Alternatively, take the bound gait of a quadruped robot as an example. The bound gait is mainly divided into two cycles. The first cycle is the ground-pushing cycle, during which the quadruped robot is in the support phase. The second cycle is the jumping cycle, during which the quadruped robot is in the air phase. In the support phase, a larger ground-pushing force should be provided to achieve the switch to the air phase, i.e. The value of should be as large as possible. In addition, the trajectory equation needs to satisfy the following conditions: when t = 0, x(0) = 0, y(0) = 0, and it should be ensured that at t = 0, The value of is as large as possible, that is, y′(0)=k, k≠0; when In the horizontal direction, the foot should reach half of the forward distance, that is In the vertical direction, the foot should reach the height of the leg lift, that is, And at this time, the vertical velocity and acceleration should be 0, that is, When t=T, in the horizontal direction, the foot end should reach the forward distance, that is, x(T)=S, and in the vertical direction, the foot end should return to the initial height, that is, y(T)=0, and to ensure symmetry, the speed at this time should be the same as the initial speed, that is, y′(T)=-k, k≠0.

[0075] In summary, for the horizontal trajectory parameter equation x(t) about time, a linear function is selected, and according to the task requirements, the following can be obtained:

[0076]

[0077] For the vertical trajectory parameter equation y(t) about time, according to the task requirements, the power function is selected as the function form of y(t):

[0078]

[0079] Substitute the task requirements to be met and find:

[0080]

[0081] The function form of y(t) is:

[0082]

[0083] Finally, the trajectory equation of the quadruped robot bound is as follows:

[0084]

[0085] The above formula is the input control motion period T, forward distance S, leg lifting height H and power function order n, and outputs the bound trajectory parameters of the horizontal and vertical direction parameter equations x(t) and y(t).

[0086] The above process takes bound gait as an example. By importing the derived trajectory parameter equation into the quadruped robot controller written in Python, the bound gait movement of the quadruped robot under the blender animation editing software is realized, and the movement process is natural and smooth. In addition, the controller written can realize the motion design of the quadruped robot under multiple control functions. Compared with the motion design in animation software in general situations, the motion design in the form of action function in the controller will make the action more flexible and editable. The action can be changed by easily modifying the function, and after using the function, the action can be precisely controlled. Moreover, the use of action functions can introduce logic and conditions into the action, which means that the robot can perform different actions based on certain conditions, thereby simulating more intelligent and flexible behaviors.

[0087] S2.2. When designing actions, since no physical constraints are added, it is impossible to ensure that the quadruped robot conforms to the action form in the physical environment. Since animation is used in action design, different interpolation methods can be selected to simulate the actions under physical constraints. By selecting different interpolation methods to achieve transitions between key frames, the actions performed can be made more natural.

[0088] The Bezier curve interpolation method is used mainly for the quadruped robot in irregular curve motion. The cubic Bezier curve is selected as the interpolation method between key frames. This interpolation method allows the animation transition to be smoother and more natural, while giving the user greater control to adjust the shape of the curve to affect the speed and easing effect of the action. By creating a cubic Bezier curve between key frames, the user can achieve more complex movements and changes in the animation. Any coordinate on the cubic Bezier curve is:

[0089] B(t)=(1-t) 3 P0+3t(1-t) 2 P1+3t 2 (1-t)P2+t 3 P3

[0090] Among them, P0, P1, P2, and P3 are control points. For the selection of control points for interpolation between two keyframes, first select the values ​​at these two keyframes to ensure that the first and last points of the control points are the endpoints of the final curve (which means that the first and last control points will be on the final curve), and the lines connecting each of them with the adjacent points are tangent to the final curve. Then select two control points that affect the shape of the curve. The selection of these two points determines the curvature and direction of the curve between the keyframes, that is, it affects the path and curvature of the curve.

[0091] The projectile interpolation and free-fall interpolation methods are mainly used for the action interpolation of the quadruped robot during the bouncing process. In the bouncing action, there are mainly two processes. The first process is the process of leaving the ground and reaching the vertex. In this process, projectile interpolation is used. The height and speed at the time of leaving the ground are known. The speed is zero when reaching the vertex. Ignoring the influence of air resistance, the time and height of reaching the vertex can be calculated by the following formula:

[0092]

[0093] Among them, v0 is the speed at the moment of leaving the ground; g is the acceleration due to gravity; t is the time difference from leaving the ground to the vertex; h0 is the height at the moment of leaving the ground; h1 is the height at the vertex.

[0094] Once the height and speed at the time of leaving the ground and reaching the top are determined, the projectile interpolation method can be used for interpolation. The calculation formula used is:

[0095]

[0096] Among them, h represents the height; t represents the time; this equation can be used to give the height value at each moment from the ground to the vertex to achieve interpolation.

[0097] The second process is the vertex-touchdown process. In this process, free fall interpolation is used, and the calculation formula used is:

[0098]

[0099] This equation can be used to give the height value at each moment from the vertex to the ground, and interpolation can be achieved. Using these two interpolation methods, interpolation that meets physical constraints can be achieved for the process of lifting off the ground, vertex, and landing.

[0100] In addition, different interpolation methods can be selected for other different actions. For example, circular interpolation or sinusoidal interpolation can be used for circular motion, and cycloid interpolation can be selected for the designed walking action to realize interpolation between walking key frames. This interpolation method can be used to simulate natural walking or moving actions. In the process of selecting cycloid interpolation, it should be fully considered that the impact force should be as small as possible when the foot starts to lift and fall, that is, x″(t=0)=0, y″(t=0)=0, x″(t=T m )=0,y″(t=T m )=0, where T m is the swing phase period, and when reaching the highest point, the speed and acceleration of the foot end in the y direction should be 0, that is, According to the walking action, the interpolation equation of the selected compound cycloid form is as follows:

[0101]

[0102]

[0103]

[0104] Among them, S is the stride; H is the leg lift height; x is the horizontal displacement; y is the vertical displacement.

[0105] The composite cycloid is selected as the interpolation method for walking motion, and the motion after interpolation is more natural.

[0106] S2.3. When designing actions, introducing the inverse kinematics model of the robot arm and foot in the animation editing software can simplify the action design process and make the designed actions appear more natural. By writing relevant python programs in the script program in blender, the inverse kinematics model of the robot arm is introduced into the action design process. By adjusting the position of the end of the robot arm, the corresponding joint angles can be calculated and assigned to the corresponding bones to control the actions of the quadruped robot.

[0107] S3, motion trajectory extraction and optimization:

[0108] After the action design is completed, it is necessary to extract and optimize the trajectory data of the designed action, and finally encapsulate the optimized trajectory data to provide an action library for subsequent behavior combination.

[0109] S3.1. Extract the action data trajectory through the trajectory extraction plug-in. The trajectory extraction process is as follows: Figure 4 As shown. For the determined trajectory extraction planning points, the trajectory data of the quadruped robot action is extracted. The extracted data should include the position, velocity, acceleration, angle and angular velocity of the center of mass and the position, velocity, acceleration and ground contact state of the foot end.

[0110] S3.2. After extracting the motion trajectory, since the blender animation editing software does not introduce physical constraints, the obtained motion trajectory may not conform to the actual situation under physical conditions, so the motion trajectory needs to be optimized.

[0111] The reinforcement learning method is used to optimize the motion trajectory of the quadruped robot. The flow chart of the reinforcement learning optimization trajectory is as follows: Figure 5 As shown:

[0112] The goal of optimization is to select a strategy π that maximizes the cumulative reward obtained by the quadruped robot during this optimization process:

[0113]

[0114] Among them, γ∈(0,1) is the discount function, and τ(π) represents the robot's motion trajectory, which is affected by the quadruped robot's strategy π and the environment.

[0115] For the selection of reward functions, the following three reward functions are mainly selected for model training:

[0116] Body Position Imitation Reward

[0117]

[0118] in, is the body position at time t, is the expected body position at time t.

[0119] Body Posture Imitation Reward Body Posture Imitation Reward

[0120]

[0121] in, is the body posture at time t, is the expected body posture at time t.

[0122] Foot end imitation reward

[0123]

[0124] in, is the position of the foot end of the single leg relative to the center of mass at time t, is the expected position of the foot end of the single leg relative to the center of mass at time t, and e is the foot number.

[0125] After optimization is completed through reinforcement learning, the resulting motion trajectory will be more in line with physical constraints, and then the optimized trajectory data will be encapsulated as sub-actions that make up the behavior.

[0126] S4, action combination and behavior generation:

[0127] After designing and optimizing, a single action is packaged into multiple action files, and the action combination generation behavior can be realized through the behavior generation software. The behavior generation software page is as follows: Figure 6 The behavior generation software mainly includes three parts: sub-action library, behavior generator and display.

[0128] The sub-action library is one of the core features of the quadruped robot behavior generation software. Here users can create, manage and edit various action sequences, including each sub-action that the robot should take. The software provides an intuitive interface where users can easily drag, drop, copy and paste sub-actions in order to create complex behaviors. Each sub-action includes the following information:

[0129] Sub-action name, used to identify the name of the action so that users can identify and manage the action sequence. Motion data, including the position, angle, speed and acceleration of the robot joints, can be edited through the blender animation software. Timing information, determines the execution order of each sub-action in the action sequence.

[0130] All encapsulated action files are stored in the sub-action library. When the sub-action data file is selected, the sub-action name will be automatically added to the sub-action library, indicating that the sub-action has been stored from the action library to the sub-action library. The sub-action library can store multiple sub-actions, and several action names can be visualized in the sub-action library. In this process, the sub-action trajectory data can be transferred to achieve the subsequent behavior generation.

[0131] The behavior generator is another important component of the quadruped robot behavior generation software. The behavior generator allows users to combine sub-actions in the sub-action library into complex behaviors and generate behavior data and execution scripts. The behavior generator provides the following functions:

[0132] Action combination: users can select and combine several sub-actions by dragging and dropping sub-actions in the sub-action library to create complex robot behaviors. Data export: the generated behavior data and execution scripts can be exported to the blender action editor and the robot control system for display action preview and actual robot motion control. Timing control: the introduction of the timeline in the behavior generator can realize the execution of different sub-actions under the timeline, and finally realize the complex behavior under the entire timeline. Scalability: the behavior generator can also realize the behavior construction of different parts of a robot such as the arm layer, torso layer and leg layer, or the behavior construction of multiple robots.

[0133] In the behavior generator, the concept of timeline is introduced, which represents the passage of time and allows users to create, combine and edit sub-actions in different time periods. Different time units can be selected in the action axis to determine the granularity of the timeline, such as seconds or milliseconds. In the timeline, users can create different time periods, each of which has a start time and an end time, which is used to specify which sub-actions should be executed during this time period. The introduction of time periods can improve the stability and flexibility of sub-action switching.

[0134] The display is the third key component of the quadruped robot behavior generation software. It allows the user to display the robot performing actions in a virtual environment in order to preview and debug action sequences. The display provides the following functions:

[0135] 3D model, displays the precise 3D model of the quadruped robot, including its structure and joints. Action preview, the display can show the effect of the robot performing actions in real time, and the user can observe the robot's motion trajectory, posture and range of motion.

[0136] Through the display, users can quickly check and modify the action sequence to ensure that the robot can achieve the desired effect when executed. This helps to improve development efficiency and reduce the cost of manufacturing and testing robots.

[0137] S5. Quadruped robot control:

[0138] In this embodiment, the control of the quadruped robot can be achieved through conventional technology; in some other embodiments, optionally, the extracted action / behavior trajectory data is imported into the controller, combined with the whole body control, to obtain the joint torque under the corresponding trajectory data, implement simulation experiments in a simulation environment, and judge the realization of the designed action in the physical environment.

[0139] Embodiment 2:

[0140] This embodiment provides a quadruped robot motion design and complex behavior generation system, including:

[0141] The animation model building module is configured to: build an animation model of the quadruped robot according to the information of the quadruped robot;

[0142] The action design module is configured to: perform action design according to the animation model; wherein the trajectory parameter equation about time in the horizontal direction is in the form of a linear function, and the trajectory parameter equation about time in the vertical direction is in the form of a power function;

[0143] The action library building module is configured to: extract trajectory data of the designed action, optimize the action with the goal of maximizing the cumulative reward obtained by the quadruped robot, and build the action library;

[0144] The behavior generation module is configured to: generate behavior trajectory data after combining the actions in the action library under the time axis; create different time periods in the time axis; each time period includes a start time and an end time, which is used to specify which actions in the action library are executed within the time period;

[0145] The control module is configured to control the quadruped robot according to the generated behavior trajectory data.

[0146] The working method of the system is the same as the quadruped robot motion design and complex behavior generation method in Example 1, and will not be repeated here.

[0147] Embodiment 3:

[0148] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the quadruped robot motion design and complex behavior generation method described in Example 1 are implemented.

[0149] Embodiment 4:

[0150] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, the steps of the quadruped robot motion design and complex behavior generation method described in Example 1 are implemented.

[0151] Embodiment 5:

[0152] This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the quadruped robot motion design and complex behavior generation method described in Example 1 are implemented.

[0153] The above description is only a preferred embodiment of the present embodiment and is not intended to limit the present embodiment. For those skilled in the art, the present embodiment may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiment shall be included in the protection scope of the present embodiment.

Claims

1. A method for designing quadruped robot motions and generating complex behaviors, characterized in that: include: According to the information of the quadruped robot, an animation model of the quadruped robot is established; According to the animation model, action design is performed; wherein, a linear function is selected for the trajectory parameter equation about time in the horizontal direction, and a power function is selected for the trajectory parameter equation about time in the vertical direction; Extract trajectory data of the designed actions, optimize them with the goal of maximizing the cumulative reward obtained by the quadruped robot, and establish an action library; The actions in the action library are combined under the time axis to generate behavior trajectory data; different time periods are created in the time axis; each time period includes a start time and an end time, which are used to specify which actions in the action library are executed within the time period; Control the quadruped robot based on the generated behavior trajectory data.

2. A method for designing motion and generating complex behaviors of a quadruped robot as claimed in claim 1, characterized in that: When building an animation model for a quadruped robot, determine the positions of the control bones at the set joints based on the joint properties, and add the remaining bones to the root bone based on the kinematic relationship and coordinate axis positions; set the link origin position, traverse the origin position of each link, and determine the robot body coordinate system position in the base coordinate system and complete the construction; then, obtain the relevant data of the link, and build the relevant data into the quadruped robot model to form a complete quadruped robot animation model with the joints.

3. A method for designing quadruped robot motion and generating complex behaviors as claimed in claim 1, characterized in that: The trajectory equation of the quadruped robot's jumping gait is: Among them, x(t) is the trajectory parameter equation about time in the horizontal direction, t is the movement time; S is the forward distance; T is the movement period; y(t) is the trajectory parameter equation about time in the vertical direction; H is the leg lifting height; n is the power function order.

4. A method for designing quadruped robot motion and generating complex behaviors as claimed in claim 1, characterized in that: Different interpolation methods are selected to simulate the movements of the quadruped robot under physical constraints. Bezier curve interpolation is used when performing irregular curve motion; projectile interpolation is used in the process from leaving the ground to the vertex; and free fall interpolation is used in the process from the vertex to touching the ground.

5. A method for designing quadruped robot motion and generating complex behaviors as claimed in claim 4, characterized in that: Choose a compound cycloid form of interpolation: Where S is the stride; T m is the movement cycle; t is the movement time; H is the height of the leg lift; x is the horizontal displacement; y is the vertical displacement.

6. A method for designing quadruped robot motions and generating complex behaviors as claimed in claim 1, characterized in that: The position, velocity, acceleration and angle of the center of mass of the quadruped robot, as well as the position, velocity, acceleration and ground contact status of the foot end are extracted.

7. A quadruped robot motion design and complex behavior generation system, characterized in that: include: The animation model building module is configured to: build an animation model of the quadruped robot according to the information of the quadruped robot; The action design module is configured to: perform action design according to the animation model; wherein the trajectory parameter equation about time in the horizontal direction is in the form of a linear function, and the trajectory parameter equation about time in the vertical direction is in the form of a power function; The action library building module is configured to: extract trajectory data of the designed action, optimize the action with the goal of maximizing the cumulative reward obtained by the quadruped robot, and build the action library; The behavior generation module is configured to: generate behavior trajectory data after combining the actions in the action library under the time axis; create different time periods in the time axis; each time period includes a start time and an end time, which is used to specify which actions in the action library are executed within the time period; The control module is configured to control the quadruped robot according to the generated behavior trajectory data.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the quadruped robot motion design and complex behavior generation method as described in any one of claims 1 to 6 are implemented.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the program, the steps of the quadruped robot motion design and complex behavior generation method as described in any one of claims 1-6 are implemented.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the quadruped robot motion design and complex behavior generation method as described in any one of claims 1 to 6 are implemented.

Citation Information

Patent Citations

  • Quadruped robot motion control method based on reinforcement learning and position increment

    CN114563954A

  • Custom Motion Trajectories for Robot Animation

    US20190111563A1