Biomimetic control method, system, device and medium for central pattern generator model

Abstract

This application discloses a biomimetic control method, system, device, and medium based on a central pattern generator model, relating to the field of biomimetic control technology. The method includes: acquiring motion performance data generated by a robot based on biomimetic control parameters; scoring the motion performance data to obtain a current score; when the current score is detected to be less than a score threshold, iteratively optimizing the initial parameters based on the motion performance data using a reinforcement learning agent to obtain optimized parameters; generating biomimetic control variations based on the optimized parameters using a normalized diversity central pattern generator model, and generating new biomimetic control parameters based on these variations; then, based on the new biomimetic control parameters, returning to the step of inputting the biomimetic control parameters into the robot, until the current score is detected to be greater than or equal to the score threshold. This application aims to solve the technical problem of how to improve the motion reliability of biomimetic robot systems.

Description

Technical Field

[0001] This application relates to the field of biomimetic control technology, and in particular to a biomimetic control method, system, device and medium for a central pattern generator model. Background Technology

[0002] CPG (Center Pattern Generator) mathematical models, as a key rhythm generation tool, are widely used in the field of robot motion control. Through their bio-inspired coordination mechanism, they provide core support for biomimetic motion control and have shown significant application value in complex motion scenarios such as legged robots and snake robots.

[0003] However, despite significant theoretical breakthroughs in the CPG model, technical bottlenecks remain in practical robot applications. Specifically, when implementing biomimetic gait control based on the CPG model, existing methods struggle to enable robots to quickly develop stable adaptive motion capabilities. Because gait generation heavily relies on human experience, robots often require repeated offline debugging to construct basic motion patterns. This not only prolongs the optimization cycle of robot motion performance but also limits its real-time response capabilities in dynamic environments. Particularly when facing complex terrain or sudden disturbances, existing solutions commonly exhibit problems such as sluggish motion mode switching and low energy efficiency, directly impacting the overall motion performance and application reliability of the robot system.

[0004] Therefore, how to improve the motion reliability of biomimetic robot systems is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The main objective of this application is to provide a biomimetic control method, system, device, and medium for a central pattern generator model, aiming to solve the technical problem of how to improve the motion reliability of biomimetic robot systems.

[0006] To achieve the above objectives, this application proposes a biomimetic control method for a central pattern generator model, the method comprising:

[0007] The biomimetic control parameters are input into the robot, and the motion performance data generated by the robot based on the biomimetic control parameters are obtained, wherein the biomimetic control parameters are obtained based on preset initial parameters;

[0008] The sports performance data is scored based on each preset performance index, and the scores corresponding to each performance index are added together to obtain the current score of the sports performance data.

[0009] The current score is compared with a preset score threshold. When the current score is detected to be less than the score threshold, the initial parameters are iteratively optimized based on the motion performance data by a preset reinforcement learning agent to obtain optimized parameters. The reinforcement learning agent is a model that has been pre-trained based on actual parameters, actual motion performance data, and the scoring results of each performance index and the corresponding optimization parameter target.

[0010] The normalized diversity central pattern generator model generates biomimetic control variation based on the optimized parameters, and generates new biomimetic control parameters based on the biomimetic control variation.

[0011] Based on the new bionic control parameters, return to the step of inputting the bionic control parameters into the robot until the current score is detected to be greater than or equal to the score threshold.

[0012] In one embodiment, prior to the step of generating biomimetic control variations based on the optimized parameters using a normalized diversity central pattern generator model, the method further includes:

[0013] Based on the difference between the unit amplitude and the current oscillator amplitude parameter and the preset waveform adjustment term, a first matrix is ​​constructed, and a second matrix is ​​constructed based on the initial parameters of the current oscillator and the preset proportional bias.

[0014] The product of the first matrix and the second matrix is ​​taken as the first target matrix;

[0015] The phase difference between the current oscillator and its adjacent oscillators is obtained, and a third matrix is ​​constructed based on the sine and cosine values ​​of the phase difference. The adjacent oscillators are the oscillators before and after the current oscillator in the chain structure.

[0016] Based on the initial parameters of the adjacent oscillators and the proportional bias, a fourth matrix is ​​constructed, and the product of the third matrix and the fourth matrix is ​​used as the second target matrix;

[0017] By superimposing the first target matrix and the second target matrix, a normalized diversity central pattern generator model is obtained.

[0018] In one embodiment, the waveform adjustment item includes a waveform type determination item. Before the step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and the preset waveform adjustment item, the method further includes:

[0019] Based on the operating frequency, linear coefficient, and nonlinear coefficient, a waveform type determination term is constructed. When the linear coefficient and the nonlinear coefficient are non-zero, the waveform of the biomimetic control variable approaches a square wave or a triangular wave. When both the linear coefficient and the nonlinear coefficient are zero, the waveform of the biomimetic control variable remains a sine wave. The operating frequency is the frequency used for biomimetic control of the robot.

[0020] The step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and a preset waveform adjustment term includes:

[0021] Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform type determination term.

[0022] In one embodiment, the step of constructing a first matrix comprising the difference between the unit amplitude and the amplitude parameter and the waveform type determination term includes:

[0023] Calculate the first cycle length of the waveform type determination item under a preset ratio, and calculate the second cycle length of the operating frequency under the preset ratio;

[0024] The scaling factor is obtained based on the ratio of the first cycle length to the second cycle length on the time scale;

[0025] The waveform type determination term is corrected by the scaling factor, and a first matrix is ​​constructed containing the difference between the unit amplitude and the amplitude parameter and the corrected waveform type determination term.

[0026] In one embodiment, the waveform adjustment term includes a waveform symmetry adjustment term. Before the step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and the preset waveform adjustment term, the method further includes:

[0027] Obtain the symmetry adjustment parameter, which is the ratio of the rise time to the total period;

[0028] The ratio of the symmetry adjustment parameter to the operating frequency is used as the rise time, and the fall time is obtained based on the difference between the total period and the rise time.

[0029] The reciprocal of the rising edge time is used as the rising edge frequency, and the reciprocal of the falling edge time is used as the falling edge frequency;

[0030] Based on the rising edge frequency, the falling edge frequency, and a preset exponential function, a waveform symmetry adjustment term is constructed so that the biomimetic control change has different frequency characteristics in the rising and falling phases of the waveform.

[0031] The step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and a preset waveform adjustment term includes:

[0032] Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform symmetry adjustment term.

[0033] In one embodiment, the method further includes:

[0034] In response to the frequency range selection command, multiple test frequencies are obtained, and the target change amount corresponding to each of the test frequencies is obtained through the normalized diversity central pattern generator model. The target change amount is the maximum change amount output by the normalized diversity central pattern generator model in a steady state.

[0035] The mapping relationship between each of the test frequencies and each of the target changes is fitted to obtain the constraint function;

[0036] The set operating frequency is input into the constraint function to obtain the target constraint value. The historical biomimetic control change is used as the interval center, and twice the value of the target constraint value is used as the interval width to construct the change constraint interval corresponding to the operating frequency. The operating frequency is the frequency used for biomimetic control of the robot.

[0037] If the biomimetic control change is detected to be within the change constraint range, the step of generating new biomimetic control parameters based on the biomimetic control change is executed.

[0038] If it is detected that the biomimetic control change amount is not within the change amount constraint range, the endpoint value with the smallest difference from the biomimetic control change amount in the change amount constraint range is taken as the new biomimetic control change amount, and based on the new biomimetic control change amount, the step of generating new biomimetic control parameters based on the biomimetic control change amount is executed.

[0039] The step of generating new biomimetic control parameters based on the biomimetic control variation includes:

[0040] The bionic control parameters of the robot at the current moment are adjusted and updated based on the bionic control change, resulting in new bionic control parameters.

[0041] In one embodiment, before the step of returning to perform the step of inputting the biomimetic control parameters to the robot, the method further includes:

[0042] The amplitude gain parameter is determined by the actual target amplitude, and the target biomimetic control parameter is generated by multiplying the biomimetic control parameter by the amplitude gain parameter.

[0043] The step of returning to execute the input of the bionic control parameters to the robot includes:

[0044] Return to the step of inputting the target biomimetic control parameters into the robot.

[0045] Furthermore, to achieve the above objectives, this application also proposes a biomimetic control system for a central pattern generator model, wherein the biomimetic control system for the central pattern generator model includes:

[0046] The performance data acquisition module is used to input biomimetic control parameters into the robot and acquire motion performance data generated by the robot based on the biomimetic control parameters, wherein the biomimetic control parameters are obtained based on preset initial parameters;

[0047] The performance data scoring module is used to score the sports performance data based on each preset performance index, and to add up the scores corresponding to each performance index to obtain the current score of the sports performance data.

[0048] The parameter self-optimization module is used to compare the current score with a preset score threshold, and when the current score is detected to be less than the score threshold, the initial parameters are iteratively optimized by a preset reinforcement learning agent based on the motion performance data to obtain optimized parameters. The reinforcement learning agent is a model obtained by pre-training based on actual parameters, actual motion performance data, and the scoring results of each performance index and the corresponding optimization parameter target.

[0049] The biomimetic control parameter generation module is used to generate biomimetic control variation based on the optimized parameters through a normalized diversity central pattern generator model, and to generate new biomimetic control parameters based on the biomimetic control variation.

[0050] The loop module is used to return to the step of inputting the bionic control parameters into the robot based on the new bionic control parameters, until the current score is detected to be greater than or equal to the score threshold.

[0051] In addition, to achieve the above objectives, this application also proposes an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the biomimetic control method of the central pattern generator model as described above.

[0052] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the bionic control method of the central pattern generator model as described above.

[0053] One or more technical solutions proposed in this application have at least the following technical effects:

[0054] In this application, for the biomimetic control parameters obtained from the initial parameters, the motion performance data corresponding to the biomimetic control parameters can be acquired. Then, the quality of the initial parameters for robot control can be determined by scoring the motion performance data through multiple performance indicators. Furthermore, when the current score of the motion performance parameters is lower than a preset score threshold, it can be determined that the current initial parameters have a poor effect on robot control. Therefore, the initial parameters can be iteratively optimized by a reinforcement learning agent to obtain optimized parameters, thereby realizing automatic updating or automatic optimization of the initial parameters. Furthermore, based on the optimized parameters obtained by the reinforcement learning agent, new biomimetic control parameters corresponding to the optimized parameters can be obtained through a normalized diversity central pattern generator model. Based on the new biomimetic control parameters, the steps of transmitting biomimetic control parameters to the robot, acquiring motion performance data, scoring motion performance data, and iteratively optimizing parameters can be repeatedly executed until the score obtained from the motion performance data evaluation is greater than or equal to the score threshold.

[0055] Compared with traditional methods that rely on human experience for debugging, this application achieves closed-loop adaptive adjustment of biomimetic control parameters through a collaborative optimization mechanism of reinforcement learning and Hopf oscillators. This enables the robot to quickly converge to a stable and efficient motion mode in a dynamic environment, thereby improving the motion reliability of the biomimetic robot system. Attached Figure Description

[0056] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating the first embodiment of the biomimetic control method for the central pattern generator model of this application.

[0059] Figure 2 This is a schematic diagram of the chain topology of a CPG, which is an embodiment of the biomimetic control method for the central pattern generator model of this application.

[0060] Figure 3This is a schematic diagram of various waveform outputs of an embodiment of the biomimetic control method for the central mode generator model of this application.

[0061] Figure 4 This is a schematic diagram illustrating the symmetry adjustment of an embodiment of the biomimetic control method for the central pattern generator model of this application.

[0062] Figure 5 This is a schematic diagram of the smooth rhythm output when the adjustable parameters are significantly changed, according to an embodiment of the biomimetic control method for the central pattern generator model of this application.

[0063] Figure 6 This is a schematic diagram of the integrated rhythm output with coupling relationship in a chain structure, representing an embodiment of the biomimetic control method for the central pattern generator model of this application.

[0064] Figure 7 This is a schematic diagram of parameter optimization for a specific embodiment of the biomimetic control method for the central pattern generator model of this application.

[0065] Figure 8 This is a schematic diagram of the module structure of the biomimetic control system of the central pattern generator model in the embodiments of this application;

[0066] Figure 9 This is a schematic diagram of the device structure of the hardware operating environment involved in the biomimetic control method of the central pattern generator model in the embodiments of this application.

[0067] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0068] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0069] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0070] Understandably, CPG (Center pattern generators) mathematical models, as a key rhythm generation tool, are widely used in the field of robot motion control. Through their bio-inspired coordination mechanisms, they provide core support for biomimetic motion control and demonstrate significant application value in complex motion scenarios such as legged robots and snake robots.

[0071] However, despite significant theoretical breakthroughs in the CPG model, technical bottlenecks remain in practical robot applications. Specifically, when implementing biomimetic gait control based on the CPG model, existing methods struggle to enable robots to quickly develop stable adaptive motion capabilities. Because gait generation heavily relies on human experience, robots often require repeated offline debugging to construct basic motion patterns. This not only prolongs the optimization cycle of robot motion performance but also limits its real-time response capabilities in dynamic environments. Particularly when facing complex terrain or sudden disturbances, existing solutions commonly exhibit problems such as sluggish motion mode switching and low energy efficiency, directly impacting the overall motion performance and application reliability of the robot system.

[0072] To address the aforementioned issues, this application provides a method for obtaining motion performance data corresponding to the biomimetic control parameters derived from initial parameters. This data can then be scored using multiple performance indicators to determine the effectiveness of the initial parameters in robot control. Furthermore, when the current score of the motion performance parameters falls below a preset score threshold, it indicates that the current initial parameters are ineffective for robot control. Therefore, the initial parameters can be iteratively optimized using a reinforcement learning agent to obtain optimized parameters, enabling automatic updating or optimization of the initial parameters. Moreover, based on the optimized parameters obtained through the reinforcement learning agent, a new biomimetic control parameter corresponding to the optimized parameter can be obtained using a normalized diversity central pattern generator model. Based on this new biomimetic control parameter, the steps of transmitting the biomimetic control parameter to the robot, acquiring motion performance data, scoring the motion performance data, and iteratively optimizing the parameters can be repeatedly executed until the score obtained from the motion performance data evaluation is greater than or equal to the score threshold.

[0073] Compared with traditional methods that rely on human experience for debugging, this application achieves closed-loop adaptive adjustment of biomimetic control parameters through a collaborative optimization mechanism of reinforcement learning and Hopf oscillators. This enables the robot to quickly converge to a stable and efficient motion mode in a dynamic environment, thereby improving the motion reliability of the biomimetic robot system.

[0074] It should be noted that the execution subject of the bionic control method of the central pattern generator model in this application can be an electronic device with data processing, network communication and program running functions, such as a tablet computer or personal computer, and this application does not limit it.

[0075] Based on the above, this application provides a biomimetic control method for a central pattern generator model, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the biomimetic control method for the central pattern generator model of this application.

[0076] In this embodiment, the biomimetic control method of the central pattern generator model includes steps S10 to S50:

[0077] Step S10: Input the bionic control parameters into the robot and obtain the motion performance data generated by the robot based on the bionic control parameters, wherein the bionic control parameters are obtained based on preset initial parameters;

[0078] It is understandable that when using Hopf oscillators for biomimetic control of a robot, each robot joint corresponds to one Hopf oscillator; that is, one joint corresponds to one Hopf oscillator. In scenarios where multiple Hopf oscillators are chained together, the mathematical model of the Hopf oscillator will include self-excited oscillation terms and coupling terms, where the coupling terms represent the influence of the preceding and / or following Hopf oscillators. When applying the mathematical model of the Hopf oscillator, the preset initial parameters for each joint need to be input into the normalized diversity central pattern generator model corresponding to each joint, thereby obtaining the changes corresponding to each joint, and then obtaining the biomimetic control parameters based on these changes. Furthermore, it is understandable that the robot can be equipped with multiple sensors, so that when the robot moves based on the biomimetic control parameters, the robot's motion performance data can be obtained through the sensors pre-set on the robot.

[0079] Furthermore, it should be noted that when initially controlling the robot using multiple chained Hopf oscillators, technicians need to input parameters. Therefore, the aforementioned initial parameters were preset and input by the technicians. Additionally, the aforementioned biomimetic control parameters are obtained by integrating the output of the mathematical model after inputting the initial parameters into the mathematical model of the Hopf oscillator.

[0080] In one feasible implementation, motion performance data may include: position data (e.g., joint angles, end effector coordinates), velocity data (e.g., linear velocity, angular velocity), acceleration data (e.g., linear acceleration, angular acceleration), power consumption data, etc., and this application does not limit it.

[0081] Step S20: The motion performance data is scored based on each preset performance index, and the scores corresponding to each performance index are added together to obtain the current score of the motion performance data.

[0082] It is understood that performance metrics may include at least: positional error of a single joint, coordination between joints, and robustness of each joint. In other scenarios, performance metrics may be added or removed depending on the requirements of the evaluation; this application does not limit the specific content of the performance metrics.

[0083] For the scoring operation, in one feasible implementation, a pre-trained machine learning model can be used to score the motion performance data based on various performance indicators. This can be done by setting up a separate machine learning model for each performance indicator, or by using a single machine learning model to score multiple performance indicators simultaneously. This application does not limit the specific scoring method.

[0084] In one feasible implementation, after obtaining the scores corresponding to each performance index, the scores can be directly superimposed or weighted and superimposed, and the superimposed score can be used as the current score of the motion performance data so as to evaluate the biomimetic control effect of the initial parameters through the score.

[0085] Step S30: Compare the current score with a preset score threshold, and when the current score is detected to be less than the score threshold, use a preset reinforcement learning agent to iteratively optimize the initial parameters based on the motion performance data to obtain optimized parameters. The reinforcement learning agent is a model that has been pre-trained based on actual parameters, actual motion performance data, and the scoring results of each performance index and the corresponding optimization parameter target.

[0086] It should be noted that the score threshold is a pre-set score used to judge whether the current motion performance data is qualified, the actual parameters refer to the initial parameters used when training the reinforcement learning agent, the actual motion performance data is the motion performance data corresponding to the actual parameters, the optimization parameter target refers to the performance standard in the parameter optimization process, and the optimized parameters output by the reinforcement learning agent are the final solution that satisfies the optimization parameter target.

[0087] Understandably, when the score of the motion performance data is less than the score threshold, it indicates that the current motion performance data is unqualified and the parameters need to be optimized; when the score of the motion performance data is greater than or equal to the score threshold, it indicates that the current motion performance data is qualified, and thus, the robot can continue to be controlled by bionic control parameters.

[0088] For example, in a quadruped robot gait control scenario, motion performance data can be collected in real time using gyroscopes and foot sensors on the quadruped robot. This motion performance data can include 12 parameters such as trunk tilt angle, joint torque fluctuation coefficient, and gait period symmetry. Then, a machine learning model can score the motion performance data based on indicators such as coordination, robustness, and error, and the scores of each indicator are summed to obtain the current score. When the current score is detected to be below a threshold of 80 points, a reinforcement learning agent parameter optimization mechanism is triggered.

[0089] In one feasible implementation, the reinforcement learning agent is constructed using the PPO algorithm. Its state space contains actual motion performance data (28-dimensional features including joint angle time-series data, ground reaction force vector, and energy consumption rate), and its action space is defined as ±15% adjustments to the actual parameters (initial parameters of the Hopf oscillator: natural frequency ω = 2.1 Hz, coupling strength ε = 0.6, phase difference Φ = π / 3). During each iteration, the agent deploys eight sets of parameter adjustment schemes in parallel within a simulation environment. It collects actual motion performance data for each set through a physics engine and designs a reward function based on the scoring results of each performance indicator.

[0090] R = 0.4 × (1 - trunk oscillation amplitude) + 0.3 × gait cycle stability + 0.2 × (1 - energy consumption rate) + 0.1 × movement speed, while also introducing a fall penalty term. The scoring results are obtained through normalized calculations of various performance indicators (e.g., trunk oscillation amplitude is quantified using the ratio of amplitude to biological baseline). After 256 policy updates, the agent's output parameter adjustment strategy converges to satisfy the optimized parameter objectives: ω = 2.35 Hz, Φ = 5π / 12 (optimized parameters).

[0091] Step S40: Generate biomimetic control variation based on the optimized parameters using the normalized diversity central pattern generator model, and generate new biomimetic control parameters based on the biomimetic control variation.

[0092] It should be noted that inputting the optimized parameters into the normalized diversity central pattern generator model yields the biomimetic control variation, which characterizes the change in control parameters of the joint corresponding to the current oscillator. Therefore, after obtaining the biomimetic control variation, the biomimetic control parameters for the next time step can be obtained based on the current biomimetic control parameters and their variation. In other words, the new biomimetic control parameters are obtained.

[0093] Step S50: Based on the new bionic control parameters, return to the step of inputting the bionic control parameters into the robot until the current score is detected to be greater than or equal to the score threshold.

[0094] In this embodiment, new bionic control parameters need to be transmitted to the robot so that the robot can output new motion performance data based on these parameters. This allows for the evaluation of the control effect of the new bionic control parameters using the new motion performance data. Optimization can be stopped when the control effect of the new bionic control parameters meets expectations (the motion performance data score is greater than or equal to the score threshold).

[0095] In this application, for the biomimetic control parameters obtained from the initial parameters, the motion performance data corresponding to the biomimetic control parameters can be acquired. Then, the quality of the initial parameters for robot control can be determined by scoring the motion performance data through multiple performance indicators. Furthermore, when the current score of the motion performance parameters is lower than a preset score threshold, it can be determined that the current initial parameters have a poor effect on robot control. Therefore, the initial parameters can be iteratively optimized by a reinforcement learning agent to obtain optimized parameters, thereby realizing automatic updating or automatic optimization of the initial parameters. Furthermore, based on the optimized parameters obtained by the reinforcement learning agent, new biomimetic control parameters corresponding to the optimized parameters can be obtained through a normalized diversity central pattern generator model. Based on the new biomimetic control parameters, the steps of transmitting biomimetic control parameters to the robot, acquiring motion performance data, scoring motion performance data, and iteratively optimizing parameters can be repeatedly executed until the score obtained from the motion performance data evaluation is greater than or equal to the score threshold.

[0096] Compared with traditional methods that rely on human experience for debugging, this application achieves closed-loop adaptive adjustment of biomimetic control parameters through a collaborative optimization mechanism of reinforcement learning and Hopf oscillators. This enables the robot to quickly converge to a stable and efficient motion mode in a dynamic environment, thereby improving the motion reliability of the biomimetic robot system.

[0097] Furthermore, based on the first embodiment of the biomimetic control method for the central pattern generator model of this application described above, a second embodiment of the biomimetic control method for the central pattern generator model of this application is proposed.

[0098] In this embodiment, before step S40 above, the method further includes:

[0099] Step S60: Based on the difference between the unit amplitude and the current oscillator amplitude parameter and the preset waveform adjustment term, construct the first matrix, and construct the second matrix according to the initial parameters of the current oscillator and the preset proportional bias.

[0100] Understandably, the optimal learning rate differs for different amplitudes. Therefore, in this application, the amplitude is normalized to a unit value of one to select a learning rate that provides stable output and rapid convergence based on a fixed unit amplitude. Furthermore, a waveform adjustment term, related to the oscillator operating frequency, is used to adjust the waveform type and its symmetry.

[0101] In this embodiment, the first matrix contains the difference between the unit amplitude and the current oscillator amplitude parameter, as well as a waveform adjustment term; the second matrix contains the initial parameters of the current oscillator, as well as the difference between the initial parameters and the proportional bias.

[0102] Step S70: The product of the first matrix and the second matrix is ​​taken as the first target matrix;

[0103] It is understandable that the first objective matrix is ​​actually the self-excited oscillation term mentioned above.

[0104] Step S80: Obtain the phase difference between the current oscillator and its adjacent oscillators, and construct a third matrix based on the sine and cosine values ​​of the phase difference. The adjacent oscillators are the oscillators preceding and following the current oscillator in the chain structure.

[0105] It is understandable that the sine and cosine values ​​of the aforementioned phase difference are elements of the third matrix.

[0106] Step S90: Based on the initial parameters of the adjacent oscillators and the proportional bias, construct a fourth matrix, and use the product of the third matrix and the fourth matrix as the second target matrix;

[0107] Understandably, the fourth matrix includes the initial parameters of adjacent oscillators, as well as the difference between the initial parameters of adjacent oscillators and the proportional bias. The second objective matrix is ​​actually the coupling term mentioned above.

[0108] Step S100: Superimpose the first target matrix and the second target matrix to obtain the normalized diversity central pattern generator model.

[0109] It is understood that, in this embodiment, the normalized diversity central pattern generator model is represented as the sum of multiple matrices. When using the normalized diversity central pattern generator model, it is only necessary to input the specific parameter values ​​of each variable in the normalized diversity central pattern generator model, and the sum of the matrices can be used as the biomimetic control change quantity.

[0110] In this embodiment, the self-excited oscillation term construction method and coupling term construction method proposed in this application can quantify and normalize the expression form of the diverse central pattern generator model. Thus, under the premise of chain structure, we can simultaneously pay attention to the influence of the current oscillator itself on the change, the influence of the previous oscillator on the current oscillator, and the influence of the next oscillator on the current oscillator, which is convenient for coordinating each joint and thus improving the biomimetic control effect of each joint.

[0111] Furthermore, it should be noted that when the amplitude is a unit amplitude, in order to avoid the divergence of the normalized diversity central pattern generator model, a learning rate that is as large as possible can be selected. This not only allows the normalized diversity central pattern generator model to converge quickly, but also reduces the error between the peak value of the curve and the unit amplitude.

[0112] In one feasible implementation, the waveform adjustment item includes a waveform type determination item, and prior to step S60 above, the method further includes:

[0113] Step S110: Based on the operating frequency, linear coefficient, and nonlinear coefficient, construct a waveform type determination term, wherein when the linear coefficient and the nonlinear coefficient are non-zero, the waveform of the biomimetic control change quantity approaches a square wave or a triangular wave, and when the linear coefficient and the nonlinear coefficient are both zero, the waveform of the biomimetic control change quantity remains a sine wave, and the operating frequency is the frequency used for biomimetic control of the robot.

[0114] For example, the output of the CPG oscillator is obtained. and Two state variables are used, and the waveform type determination item can be determined based on either of these two state variables to select... For example (and similarly, subsequent related content will also be based on this). (Using an example for illustration), the waveform type determination item can be:

[0115] ;

[0116] in, The operating frequency is an adjustable frequency parameter used to determine the periodic characteristics of the output rhythm. The coefficients are linear. These are nonlinear coefficients.

[0117] In another feasible implementation, the waveform type determination item can also be:

[0118] ;

[0119] Based on this, step S60 above also includes:

[0120] Step S601: Construct a first matrix containing the difference between the unit amplitude and the amplitude parameter, as well as the waveform type determination term.

[0121] Therefore, the waveform of the biomimetic control parameters can be adjusted by changing the values ​​of the linear and nonlinear coefficients.

[0122] In one feasible implementation, step S601 above includes:

[0123] Step S6011: Calculate the first cycle length of the waveform type determination item under the preset ratio, and calculate the second cycle length of the operating frequency under the preset ratio;

[0124] Step S6012: Based on the ratio of the first cycle length and the second cycle length on the time scale, obtain the scaling factor;

[0125] Step S6013: Correct the waveform type determination term using the scaling factor, and construct a first matrix containing the difference between the unit amplitude and the amplitude parameter and the corrected waveform type determination term.

[0126] For example, taking a preset ratio of one-quarter as an example, the process of obtaining the scaling factor k is as follows:

[0127] For operating frequency Standard cycle for: Therefore, the length of the second period is: For the waveform type determination item, the instantaneous period is: ,because Will with output The length of the first cycle changes, therefore, it needs to be calculated through integration; hence, the length of the first cycle... for:

[0128] ,

[0129] To ensure that the actual cycle matches the set cycle, the following must be met: Substitute and Then, the scaling factor k can be obtained as:

[0130] ;

[0131] Therefore, the waveform adjustment term can be divided by k to correct the waveform adjustment term through the scaling factor. Then, a first matrix containing the corrected waveform adjustment term can be constructed so that the actual output period of the bionic control parameters is consistent with the operating frequency.

[0132] Furthermore, the waveform adjustment item includes a waveform symmetry adjustment item, and prior to step S60 above, the method further includes:

[0133] Step S120: Obtain the symmetry adjustment parameter, which is the ratio of the rise time to the total period;

[0134] Step S130: The ratio of the symmetry adjustment parameter to the operating frequency is used as the rise time, and the fall time is obtained based on the difference between the total period and the rise time.

[0135] It is understandable that the falling edge time is the difference between the total period and the rising edge time.

[0136] Step S140: Take the reciprocal of the rising edge time as the rising edge frequency, and take the reciprocal of the falling edge time as the falling edge frequency;

[0137] Step S150: Based on the rising edge frequency, the falling edge frequency, and a preset exponential function, construct a waveform symmetry adjustment term so that the biomimetic control change amount has different frequency characteristics in the rising and falling phases of the waveform.

[0138] For example, rising edge frequency With falling edge frequency It can be represented as:

[0139] ;

[0140] .

[0141] In addition, the waveform symmetry adjustment item can be:

[0142] ;

[0143] Therefore, this application can adjust the waveform symmetry of the bionic control parameters through the waveform symmetry adjustment term, thereby outputting diverse waveforms.

[0144] It is understandable that the rising edge frequency and the falling edge frequency conform to... relation:

[0145] ,

[0146] Prove that the waveform symmetry adjustment term has the same effect as the set frequency. Because they share the same periodicity, no additional scaling factor is needed when adjusting waveform symmetry.

[0147] Based on this, step S60 above includes:

[0148] Step S602: Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform symmetry adjustment term.

[0149] It is understandable that by introducing a waveform symmetry adjustment term into the first matrix in this application, the output waveform of the biomimetic control variable can be flexibly changed based on the symmetry adjustment parameters input by the user, thereby improving the diversity of waveforms.

[0150] Furthermore, in one feasible implementation, the method further includes:

[0151] Step S160: In response to the frequency range selection command, multiple test frequencies are obtained, and the target change amount corresponding to each of the test frequencies is obtained through the normalized diversity central mode generator model. The target change amount is the maximum change amount output by the normalized diversity central mode generator model in a stable state.

[0152] It is understandable that the frequency range selection instruction includes the frequency range selected by the user, and the test frequency refers to the frequency within the frequency range selected by the user.

[0153] Step S170: Fit the mapping relationship between each of the test frequencies and each of the target changes to obtain a constraint function;

[0154] For example, the test frequency can be used as the independent variable and the target change as the dependent variable to input a pre-selected function (linear function, quadratic function, etc.). Then, the coefficients of the function can be determined by minimizing the sum of squared residuals between the predicted value and the actual observed value. The calculated function coefficients can then be combined with the set function (linear function, quadratic function, etc.) to obtain the constraint function.

[0155] Step S180: Input the set working frequency into the constraint function to obtain the target constraint value, and construct the change constraint interval corresponding to the working frequency with the historical bionic control change amount as the interval center and twice the target constraint value as the interval width. The working frequency is the frequency used to perform bionic control on the robot.

[0156] In one feasible implementation, the historical biomimetic control variation can be used as the center of the variation constraint interval. That is, in a constraint function CF(w), the historical biomimetic control variation is... At that time, the range of variation constraints can be .

[0157] Step S190: If the biomimetic control change is detected to be within the change constraint range, execute the step of generating new biomimetic control parameters based on the biomimetic control change.

[0158] Step S200: If it is detected that the bionic control change amount is not within the change amount constraint range, take the endpoint value with the smallest difference between the change amount constraint range and the bionic control change amount as the new bionic control change amount, and perform the step of generating new bionic control parameters based on the new bionic control change amount based on the bionic control change amount.

[0159] Understandably, the variation constraint interval is used to prevent peaks in the waveform of the biomimetic control parameters when the operating frequency changes significantly. Therefore, after obtaining the biomimetic control variation, it is necessary to confirm whether the variation falls within the variation constraint interval corresponding to the operating frequency. If it does, it indicates that the biomimetic control variation will not cause peaks, and thus biomimetic control parameters can be generated from the variation. If it does not fall within the variation constraint interval, the biomimetic control variation needs to be truncated using the variation constraint interval. Specifically, the endpoint with the smallest difference from the variation of the two endpoints of the variation constraint interval can be used as the new biomimetic control variation to avoid peaks.

[0160] In this embodiment, the present application avoids the occurrence of spikes by constraining the range of changes, thereby improving the stability of the CPG model.

[0161] Furthermore, in one feasible implementation, the above step of generating new biomimetic control parameters based on the biomimetic control variation includes:

[0162] The bionic control parameters of the robot at the current moment are adjusted and updated based on the bionic control change, resulting in new bionic control parameters.

[0163] For example, the bionic control parameters at the current moment are: The time step is The amount of change in biomimetic control is At that time, the new biomimetic control parameters can be expressed as:

[0164] .

[0165] Furthermore, in one feasible implementation, prior to step S50 above, the method further includes:

[0166] Step S210: Determine the amplitude gain parameter through the actual target amplitude, and multiply the biomimetic control parameter with the amplitude gain parameter to generate the target biomimetic control parameter;

[0167] It should be noted that when performing biomimetic control on robotic fish, the amplitude is usually set to a non-unit amplitude. Therefore, for the normalized diversity central pattern generator model with the unit amplitude mentioned above, compensation is still required at the output of the normalized diversity central pattern generator model in the form of amplitude gain.

[0168] It is understandable that the aforementioned actual target amplitude is the amplitude set for the robotic fish, and the amplitude gain parameter is a parameter determined based on the actual target amplitude. For example, if the actual target amplitude set for the robotic fish is 5, then the amplitude gain coefficient is also 5. Then, the biomimetic control parameters obtained through the unit amplitude can be multiplied by the amplitude gain parameter to obtain the target biomimetic control parameters for the robotic fish.

[0169] Based on this, step S50 above includes:

[0170] Step S501: Return to the step of inputting the target bionic control parameters into the robot.

[0171] In this embodiment, the amplitude gain parameter can be used to adapt the present application to scenarios that require amplitude compensation, thereby improving the scenario adaptability of the present application.

[0172] Furthermore, in one feasible implementation, the method further includes:

[0173] Step S220: When the change in the amplitude gain parameter is detected to be greater than a set value, the amplitude gain parameter is gradually updated through an amplitude smoothing transition mechanism;

[0174] In this embodiment, due to the output of the oscillator It has been normalized, therefore in amplitude Significant changes may cause changes in the joint angle of the CPG output. The amplitude parameter modification rule is designed to reduce the scaling effect of the old amplitude in the oscillator output before setting the new amplitude, so that the oscillator can autonomously smooth large amplitude changes.

[0175] For example, in one feasible implementation, the amplitude can be modified by the following mathematical formula:

[0176] ,

[0177] ,

[0178] in, The original amplitude parameter value. The proposed new amplitude parameter value, It is an amplitude variable.

[0179] Furthermore, in one feasible implementation, the aforementioned unit amplitude, waveform type determination term, scaling factor, and waveform symmetry adjustment term can also be used simultaneously to limit the output of the normalized diversity central pattern generator model. In this case, the expression for the normalized diversity central pattern generator model can be:

[0180] ;

[0181] Where i represents the i-th hopf oscillator. , for and The derivative, that is, the amount of change in biomimetic control. and Here are the intrinsic parameters, representing the learning rate and coupling coefficient, respectively. , The 1 in the figure represents the unit amplitude. For proportional bias, This characterizes the first hopf oscillator before and after the i-th hopf oscillator in the chain structure. Let be the phase difference between the i-th hopf oscillator and the j-th hopf oscillator. These are waveform adjustment items used to control the waveform and waveform symmetry (equivalent to a set of waveform type determination items, scaling factors, and waveform symmetry adjustment items).

[0182] In scenarios involving the combination of waveform type determination items, scaling factors, and waveform symmetry adjustment items, It consists of two parts representing waveform diversity and periodic asymmetry, and can be specifically represented as:

[0183] ,

[0184] in, Indicating waveform diversity, This indicates periodic asymmetry.

[0185] Furthermore, based on the above formula, constraint functions and Euler integrals can be introduced to obtain biomimetic control parameters.

[0186] Specifically, the constraint interval for the change obtained through the constraint function CF(w) can be:

[0187] ,

[0188] in, and It is the change in biomimetic control in the previous oscillation cycle, that is, the historical change in biomimetic control.

[0189] When the derivative of the oscillator output is detected to be within the variation constraint range, the biomimetic control parameters can be obtained using the following formula:

[0190] .

[0191] Please refer to Figure 2 , Figure 2This is a schematic diagram of the chain topology of a CPG, representing an embodiment of the biomimetic control method for the central pattern generator model of this application. Let N represent the i-th CPG model (i.e., the i-th normalized diversity central pattern generator model) in the chain topology, where N is the total number of CPG models. The joint angle output by the i-th CPG model. Let be the phase parameters of the i-th CPG model. Let be the amplitude parameter of the i-th CPG model. The operating frequency of the i-th CPG model. Let be the connection weight parameters of the i-th CPG model. Figure 2 The adjustable parameters and inherent parameters in the model are all initial parameters. This is achieved through the above-described normalized diversity central pattern generator model expression, variation constraint interval, and... The expression in this application can improve the stability and diversity of the output waveform of the CPG model.

[0192] Please refer to further details. Figure 3 , Figure 3 This is a schematic diagram illustrating various waveform outputs of an embodiment of the biomimetic control method for the central pattern generator model of this application. Figure 3 This includes sine waves, square waves, triangular waves, and asymmetrical square wave waveforms. It is understandable that by modifying the above... By adjusting the values ​​of linear, nonlinear, and symmetry parameters, different waveforms can be obtained, leading to different trends in the biomimetic control parameters. This facilitates the output of different biomimetic control parameters in different scenarios, allowing the robot to smoothly switch between walking, running, and other movement modes.

[0193] Specifically, an example of adjusting the waveform symmetry by adjusting the symmetry adjustment parameter can be found in [reference needed]. Figure 4 , Figure 4 This is a schematic diagram illustrating the symmetry adjustment of an embodiment of the biomimetic control method for the central pattern generator model of this application. Figure 4 It can be seen that when the symmetry adjustment parameter is adjusted to 0.5, the time ratio of the rising edge to the falling edge is 1:1, thus presenting a symmetrical waveform. When the symmetry adjustment parameter is adjusted to 0.75, the time ratio of the rising edge to the falling edge is 3:1, thus presenting an asymmetrical waveform.

[0194] Further, please refer to Figure 5 , Figure 5This diagram illustrates the smooth rhythmic output of a biomimetic control method for the central pattern generator model of this application when the adjustable parameters are significantly changed. As can be seen from the diagram, when the adjustable parameters change significantly, the output angle curve changes smoothly without abrupt jumps or interruptions. Taking the key time points in the diagram as examples: from 0 to 1.5 seconds, parameters ω are 4π, A is 20, and b is 0, resulting in stable waveform fluctuations. From 1.5 to 2.5 seconds, ω changes to 2π, A and b remain unchanged, and the waveform fluctuations slow down but remain stable. From 2.5 to 4.5 seconds, A changes to 5, and b changes to -1, the waveform amplitude decreases and the center position shifts downward, resulting in a smooth transition. From 4.5 to 6.5 seconds, A changes to 30, and b changes to 2, the waveform fluctuations accelerate, the amplitude increases, and the center position shifts upward, resulting in a smooth change. From 6.5 to 9 seconds, the parameters remain ω at 6π, A at 10, and b at 0, and the waveform fluctuates smoothly. Throughout the entire parameter change process, the waveform transitions smoothly without abrupt changes. It is evident that this application can achieve stable and diverse biomimetic control effects through a normalized diversity central pattern generator model that possesses stability and diversity.

[0195] Furthermore, please continue to refer to Figure 6 , Figure 6 This is a schematic diagram of the integrated rhythm output with coupling relationship in a chain structure, representing an embodiment of the biomimetic control method for the central pattern generator model of this application. Figure 6 The diagram includes the output angle curves of three joints (joint units J1, J2, and J3) and the changes in their adjustable parameters. The frequency ω of joint unit J1 changes from 4π to 2π, causing its output angle fluctuation to slow down. The amplitude A of joint unit J2 changes from 20 to 30 and then back to 20, and the connection weight β changes from 0.5 to 0.3 and then back to 0.5, correspondingly changing the fluctuation amplitude of its output angle and the coupling strength with J3. The bias b of joint unit J3 changes from 0 to -1 and then back to 0, and the phase difference Δφ23 changes from π to 0, changing the position of the fluctuation center of the output angle and its phase relationship with J2.

[0196] Throughout the process, the output angle curves of the three joints maintain a smooth transition and the desired phase relationship. When the frequency of J1 decreases, the movements of J2 and J3 adjust accordingly without abrupt changes. When the amplitude of J2 changes, its own movement and its coupled movement with J3 also gradually adapt. As the phase difference decreases, the movements of J2 and J3 gradually change from out of phase to synchronized, resulting in a coherent overall motion. It is evident that, with changes in adjustable parameters, this application can enable multiple joints to coordinate and generate a smooth rhythmic output through coupling relationships.

[0197] Furthermore, please refer to Figure 7 , Figure 7This diagram illustrates parameter optimization in a specific embodiment of the biomimetic control method for the central pattern generator model of this application. The ND-CPG parameters are manually set as initial parameters; the ND-CPG mathematical model is a normalized diversity central pattern generator model; the biomimetic robotic fish system is a robot; the kinematic / dynamic performance data is motion performance data; and the ND-CPG mathematical model possesses the functions of amplitude normalization (i.e., unit amplitude), waveform diversification (switchable waveforms, waveform symmetry), and constraint of changes through change constraint terms. Figure 7 In the process, for manually set parameters, one can first use the aforementioned parameters that have unit amplitude, constraint functions, and... The ND-CPG mathematical model is processed, and the resulting bionic control parameters are transmitted to the bionic robotic fish system. Then, the ND-CPG parameters can be iteratively optimized by the enhanced agent based on the motion performance data of the bionic robotic fish system, thereby improving the efficiency of bionic control.

[0198] Therefore, this application can improve the stability of the CPG model through unit amplitude and constraint functions. Furthermore, this application can also achieve waveform diversity through waveform adjustment terms, scaling factors, and waveform symmetry adjustment terms.

[0199] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the biomimetic control method of the central pattern generator model of this application. Any simple transformations based on this technical concept are within the protection scope of this application.

[0200] This application also provides a biomimetic control system based on a central pattern generator model; please refer to [reference needed]. Figure 8 The biomimetic control system of the central pattern generator model includes:

[0201] The performance data acquisition module 10 is used to input biomimetic control parameters into the robot and acquire motion performance data generated by the robot based on the biomimetic control parameters, wherein the biomimetic control parameters are obtained based on preset initial parameters;

[0202] The performance data scoring module 20 is used to score the sports performance data based on each preset performance index, and to add up the scores corresponding to each performance index to obtain the current score of the sports performance data.

[0203] The parameter self-optimization module 30 is used to compare the current score with a preset score threshold, and when the current score is detected to be less than the score threshold, iteratively optimizes the initial parameters based on the motion performance data through a preset reinforcement learning agent to obtain optimized parameters. The reinforcement learning agent is a model obtained by pre-training based on actual parameters, actual motion performance data, and the scoring results of each performance index and the corresponding optimization parameter target.

[0204] The biomimetic control parameter generation module 40 is used to generate biomimetic control variation based on the optimized parameters through a normalized diversity central pattern generator model, and to generate new biomimetic control parameters based on the biomimetic control variation.

[0205] The loop module 50 is used to return to the step of inputting the bionic control parameters into the robot based on the new bionic control parameters, until the current score is detected to be greater than or equal to the score threshold.

[0206] In one embodiment, the biomimetic control system of the central pattern generator model further includes:

[0207] The first construction module is used to construct a first matrix based on the difference between the unit amplitude and the current oscillator amplitude parameter and the preset waveform adjustment term, and to construct a second matrix based on the initial parameters of the current oscillator and the preset proportional bias.

[0208] The second construction module is used to take the product of the first matrix and the second matrix as the first target matrix;

[0209] The third construction module is used to obtain the phase difference between the current oscillator and its adjacent oscillators, and to construct a third matrix based on the sine and cosine values ​​of the phase difference. The adjacent oscillators are the oscillators before and after the current oscillator in the chain structure.

[0210] The fourth construction module is used to construct a fourth matrix based on the initial parameters of the adjacent oscillators and the proportional bias, and to use the product of the third matrix and the fourth matrix as the second target matrix;

[0211] The fifth construction module is used to superimpose the first target matrix and the second target matrix to obtain the normalized diversity central pattern generator model.

[0212] In one embodiment, the waveform adjustment item includes a waveform type determination item, and the biomimetic control system of the central pattern generator model further includes:

[0213] The waveform type determination module is used to construct a waveform type determination item based on the operating frequency, linear coefficient, and nonlinear coefficient. When the linear coefficient and the nonlinear coefficient are non-zero, the waveform of the biomimetic control variable approaches a square wave or a triangular wave. When both the linear coefficient and the nonlinear coefficient are zero, the waveform of the biomimetic control variable remains a sine wave. The operating frequency is the frequency used for biomimetic control of the robot.

[0214] Based on this, the first construction module mentioned above is also used for:

[0215] Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform type determination term.

[0216] In one embodiment, the first construction module described above is further configured to:

[0217] Calculate the first cycle length of the waveform type determination item under a preset ratio, and calculate the second cycle length of the operating frequency under the preset ratio;

[0218] The scaling factor is obtained based on the ratio of the first cycle length to the second cycle length on the time scale;

[0219] The waveform type determination term is corrected by the scaling factor, and a first matrix is ​​constructed containing the difference between the unit amplitude and the amplitude parameter and the corrected waveform type determination term.

[0220] In one embodiment, the waveform adjustment term includes a waveform symmetry adjustment term, and the biomimetic control system of the central pattern generator model includes:

[0221] The first parameter acquisition module is used to acquire the symmetry adjustment parameter, which is the ratio of the rise time to the total period;

[0222] The time calculation module is used to take the ratio of the symmetry adjustment parameter to the operating frequency as the rising edge time, and to obtain the falling edge time based on the difference between the total period and the rising edge time.

[0223] The frequency calculation module is used to take the reciprocal of the rising edge time as the rising edge frequency and the reciprocal of the falling edge time as the falling edge frequency.

[0224] The sixth construction module is used to construct a waveform symmetry adjustment term based on the rising edge frequency, the falling edge frequency, and a preset exponential function, so that the biomimetic control change quantity has different frequency characteristics in the rising and falling phases of the waveform.

[0225] Based on this, the first construction module mentioned above is also used for:

[0226] Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform symmetry adjustment term.

[0227] In one embodiment, the biomimetic control system of the central pattern generator model further includes:

[0228] The second parameter acquisition module is used to respond to the frequency range selection command, obtain multiple test frequencies, and obtain the target change amount corresponding to each of the test frequencies through the normalized diversity central mode generator model. The target change amount is the maximum change amount output by the normalized diversity central mode generator model in a stable state.

[0229] The function fitting module is used to fit the mapping relationship between each of the test frequencies and each of the target changes to obtain the constraint function;

[0230] The constraint interval construction module is used to input the set working frequency into the constraint function to obtain the target constraint value, and construct the change constraint interval corresponding to the working frequency with the historical bionic control change amount as the interval center and twice the target constraint value as the interval width. The working frequency is the frequency used for bionic control of the robot.

[0231] The first switching module is used to execute the step of generating new biomimetic control parameters based on the biomimetic control change when the biomimetic control change is detected to be within the change constraint range.

[0232] The second transfer module is used to, when it is detected that the bionic control change amount is not within the change amount constraint range, take the endpoint value with the smallest difference between the change amount and the bionic control change amount in the change amount constraint range as the new bionic control change amount, and perform the step of generating new bionic control parameters based on the new bionic control change amount based on the new bionic control change amount.

[0233] Based on this, the aforementioned biomimetic control parameter generation module 40 is also used for:

[0234] The bionic control parameters of the robot at the current moment are adjusted and updated based on the bionic control change, resulting in new bionic control parameters.

[0235] In one embodiment, the biomimetic control system of the central pattern generator model further includes:

[0236] An amplitude adjustment module is used to determine the amplitude gain parameter based on the actual target amplitude, and multiply the biomimetic control parameter by the amplitude gain parameter to generate the target biomimetic control parameter;

[0237] Based on this, the aforementioned loop module 50 is also used for:

[0238] Return to the step of inputting the target biomimetic control parameters into the robot.

[0239] The biomimetic control system for the central pattern generator model provided in this application, employing the biomimetic control method for the central pattern generator model in the above embodiments, can solve the technical problem of how to improve the motion reliability of biomimetic robot systems. Compared with the prior art, the beneficial effects of the biomimetic control system for the central pattern generator model provided in this application are the same as the beneficial effects of the biomimetic control method for the central pattern generator model provided in the above embodiments, and other technical features in the biomimetic control system for the central pattern generator model are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0240] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the biomimetic control method of the central pattern generator model in Embodiment 1 above.

[0241] The following is for reference. Figure 9 It shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of this application. Figure 9 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0242] like Figure 9As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although the diagrams show electronic devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0243] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0244] The electronic device provided in this application employs the biomimetic control method of the central pattern generator model in the above embodiments, which can solve the technical problem of how to improve the motion reliability of biomimetic robot systems. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the biomimetic control method of the central pattern generator model provided in the above embodiments, and other technical features of this electronic device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0245] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0246] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0247] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the bionic control method of the central pattern generator model in the above embodiments.

[0248] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0249] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0250] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device causes the following: It inputs initial parameters into a preset normalized diversity central pattern generator model to obtain biomimetic control parameters output by the normalized diversity central pattern generator model; it inputs the biomimetic control parameters into a robot and acquires motion performance data generated by the robot based on the biomimetic control parameters; it scores the motion performance data to obtain a current score and compares the current score with a preset score threshold; when the current score is detected to be less than the score threshold, it iteratively optimizes the initial parameters based on the motion performance data using a preset reinforcement learning agent to obtain optimized parameters, and generates new biomimetic control parameters based on the optimized parameters using the normalized diversity central pattern generator model; it returns to the step of inputting the biomimetic control parameters into the robot until the current score is detected to be greater than or equal to the score threshold.

[0251] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0252] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0253] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0254] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the biomimetic control method of the above-described central pattern generator model, which can solve the technical problem of how to improve the motion reliability of a biomimetic robot system. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the biomimetic control method of the central pattern generator model provided in the above embodiments, and will not be repeated here.

[0255] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the bionic control method for the central pattern generator model as described above.

[0256] The computer program product provided in this application can solve the technical problem of how to improve the motion reliability of a biomimetic robot system. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the biomimetic control method of the central pattern generator model provided in the above embodiments, and will not be repeated here.

[0257] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A biomimetic control method for a central pattern generator model, characterized in that, The biomimetic control method of the central pattern generator model includes: The biomimetic control parameters are input into the robot, and the motion performance data generated by the robot based on the biomimetic control parameters are obtained, wherein the biomimetic control parameters are obtained based on preset initial parameters; The sports performance data is scored based on each preset performance index, and the scores corresponding to each performance index are added together to obtain the current score of the sports performance data. The current score is compared with a preset score threshold. When the current score is detected to be less than the score threshold, the initial parameters are iteratively optimized based on the motion performance data by a preset reinforcement learning agent to obtain optimized parameters. The reinforcement learning agent is a model that has been pre-trained based on actual parameters, actual motion performance data, and the scoring results of each performance index and the corresponding optimization parameter target. Based on the difference between the unit amplitude and the current oscillator amplitude parameter and the preset waveform adjustment term, a first matrix is ​​constructed, and a second matrix is ​​constructed according to the initial parameters of the current oscillator and the preset proportional bias. The product of the first matrix and the second matrix is ​​taken as the first target matrix; The phase difference between the current oscillator and its adjacent oscillators is obtained, and a third matrix is ​​constructed based on the sine and cosine values ​​of the phase difference. The adjacent oscillators are the oscillators before and after the current oscillator in the chain structure. Based on the initial parameters of the adjacent oscillators and the proportional bias, a fourth matrix is ​​constructed, and the product of the third matrix and the fourth matrix is ​​used as the second target matrix; By superimposing the first target matrix and the second target matrix, a normalized diversity central pattern generator model is obtained; The normalized diversity central pattern generator model generates biomimetic control variation based on the optimized parameters, and generates new biomimetic control parameters based on the biomimetic control variation. Based on the new bionic control parameters, return to the step of inputting the bionic control parameters into the robot until the current score is detected to be greater than or equal to the score threshold.

2. The biomimetic control method for the central pattern generator model as described in claim 1, characterized in that, The waveform adjustment terms include a waveform type determination term. Before the step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and the preset waveform adjustment terms, the method further includes: Based on the operating frequency, linear coefficient, and nonlinear coefficient, a waveform type determination term is constructed. When the linear coefficient and the nonlinear coefficient are non-zero, the waveform of the biomimetic control variable approaches a square wave or a triangular wave. When both the linear coefficient and the nonlinear coefficient are zero, the waveform of the biomimetic control variable remains a sine wave. The operating frequency is the frequency used for biomimetic control of the robot. The step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and a preset waveform adjustment term includes: Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform type determination term.

3. The biomimetic control method for the central pattern generator model as described in claim 2, characterized in that, The step of constructing a first matrix including the difference between the unit amplitude and the amplitude parameter and the waveform type determination term includes: Calculate the first cycle length of the waveform type determination item under a preset ratio, and calculate the second cycle length of the operating frequency under the preset ratio; The scaling factor is obtained based on the ratio of the first cycle length to the second cycle length on the time scale; The waveform type determination term is corrected by the scaling factor, and a first matrix is ​​constructed containing the difference between the unit amplitude and the amplitude parameter and the corrected waveform type determination term.

4. The biomimetic control method for the central pattern generator model as described in claim 1, characterized in that, The waveform adjustment terms include waveform symmetry adjustment terms. Before the step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and the preset waveform adjustment terms, the method further includes: Obtain the symmetry adjustment parameter, which is the ratio of the rise time to the total period; The ratio of the symmetry adjustment parameter to the operating frequency is used as the rise time, and the fall time is obtained based on the difference between the total period and the rise time. The reciprocal of the rising edge time is used as the rising edge frequency, and the reciprocal of the falling edge time is used as the falling edge frequency; Based on the rising edge frequency, the falling edge frequency, and a preset exponential function, a waveform symmetry adjustment term is constructed so that the biomimetic control change has different frequency characteristics in the rising and falling phases of the waveform. The step of constructing the first matrix based on the difference between the unit amplitude and the amplitude parameter and a preset waveform adjustment term includes: Construct a first matrix that includes the difference between the unit amplitude and the amplitude parameter, as well as the waveform symmetry adjustment term.

5. The biomimetic control method for the central pattern generator model as described in claim 1, characterized in that, The method further includes: In response to the frequency range selection command, multiple test frequencies are obtained, and the target change amount corresponding to each of the test frequencies is obtained through the normalized diversity central pattern generator model. The target change amount is the maximum change amount output by the normalized diversity central pattern generator model in a steady state. The mapping relationship between each of the test frequencies and each of the target changes is fitted to obtain the constraint function; The set operating frequency is input into the constraint function to obtain the target constraint value. The change constraint interval corresponding to the operating frequency is constructed with the historical bionic control change amount as the interval center and twice the target constraint value as the interval width. The operating frequency is the frequency used for bionic control of the robot. If the biomimetic control change is detected to be within the change constraint range, the step of generating new biomimetic control parameters based on the biomimetic control change is executed. If it is detected that the biomimetic control change amount is not within the change amount constraint range, the endpoint value with the smallest difference from the biomimetic control change amount in the change amount constraint range is taken as the new biomimetic control change amount, and the step of generating new biomimetic control parameters based on the new biomimetic control change amount is executed based on the new biomimetic control change amount. The step of generating new biomimetic control parameters based on the biomimetic control variation includes: The bionic control parameters of the robot at the current moment are adjusted and updated based on the bionic control change, resulting in new bionic control parameters.

6. The biomimetic control method for the central pattern generator model as described in claim 1, characterized in that, Before the step of returning to perform the step of inputting the biomimetic control parameters into the robot, the method further includes: The amplitude gain parameter is determined by the actual target amplitude, and the target biomimetic control parameter is generated by multiplying the biomimetic control parameter by the amplitude gain parameter. The step of returning to execute the input of the bionic control parameters to the robot includes: Return to the step of inputting the target biomimetic control parameters into the robot.

7. A biomimetic control system based on a central pattern generator model, characterized in that, The biomimetic control system of the central pattern generator model includes: The performance data acquisition module is used to input biomimetic control parameters into the robot and acquire motion performance data generated by the robot based on the biomimetic control parameters, wherein the biomimetic control parameters are obtained based on preset initial parameters; The performance data scoring module is used to score the sports performance data based on each preset performance index, and to add up the scores corresponding to each performance index to obtain the current score of the sports performance data. The parameter self-optimization module is used to compare the current score with a preset score threshold, and when the current score is detected to be less than the score threshold, the initial parameters are iteratively optimized by a preset reinforcement learning agent based on the motion performance data to obtain optimized parameters. The reinforcement learning agent is a model obtained by pre-training based on actual parameters, actual motion performance data, and the scoring results of each performance index and the corresponding optimization parameter target. The first construction module is used to base the value on the difference between the unit amplitude and the current oscillator amplitude parameter. The first matrix is ​​constructed based on the preset waveform adjustment terms, and the second matrix is ​​constructed based on the initial parameters of the current oscillator and the preset proportional bias. The second construction module is used to take the product of the first matrix and the second matrix as the first target matrix; The third construction module is used to obtain the phase difference between the current oscillator and its adjacent oscillators, and to construct a third matrix based on the sine and cosine values ​​of the phase difference. The adjacent oscillators are the oscillators before and after the current oscillator in the chain structure. The fourth construction module is used to construct a fourth matrix based on the initial parameters of the adjacent oscillators and the proportional bias, and to use the product of the third matrix and the fourth matrix as the second target matrix; The fifth construction module is used to superimpose the first target matrix and the second target matrix to obtain the normalized diversity central pattern generator model; The biomimetic control parameter generation module is used to generate biomimetic control variation based on the optimized parameters through the normalized diversity central pattern generator model, and to generate new biomimetic control parameters based on the biomimetic control variation. The loop module is used to return to the step of inputting the bionic control parameters into the robot based on the new bionic control parameters, until the current score is detected to be greater than or equal to the score threshold.

8. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the biomimetic control method for the central pattern generator model as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the bionic control method of the central pattern generator model as described in any one of claims 1 to 6.

Images