A parameter automatic adjustment method for bipedal robot walking controller

Abstract

The invention discloses a parameter automatic adjustment method applied to a walking controller of a biped robot, which decomposes walking controller parameters into first inverse kinematics controller parameters, first inverse dynamics controller parameters and first active impedance controller parameters; turns off the inverse dynamics controller and the active impedance controller of the walking controller, optimizes the first inverse kinematics controller parameters by using a Bayesian optimization method, and obtains second inverse kinematics controller parameters; keeps the second inverse kinematics controller parameters unchanged, and alternately optimizes the first inverse dynamics controller parameters and the first active impedance controller parameters by using a Bayesian optimization method, and obtains second inverse dynamics controller parameters and second active impedance controller parameters; combines the second inverse kinematics controller parameters, the second inverse dynamics controller parameters and the second active impedance controller parameters to obtain optimized walking controller parameters; the invention can improve the parameter accuracy of the walking controller.

Description

Technical Field

[0001] The invention belongs to the technical field of biped robots, and in particular relates to an automatic parameter adjustment method applied to a biped robot walking controller. Background Art

[0002] Biped robots are legged robots that use two legs to move. Biped robots are moving from laboratory environments to various complex working environments. Stable walking control is the basis for biped robots to complete various work tasks. In recent years, a variety of control methods for stable walking of biped robots have emerged around the world. These methods achieve stable walking control from the perspective of control law design.

[0003] The walking stability control of the biped robot is achieved through the walking controller, which consists of three controllers working simultaneously, namely the inverse kinematics controller, the inverse dynamics controller and the active impedance controller. Among them, the inverse kinematics controller and the inverse dynamics controller are used to convert the robot's task space trajectory (center of mass and two-foot trajectory) into the position and torque of the joint space, respectively, and the active impedance controller realizes joint flexibility through torque feedforward and position error control.

[0004] Sensitive walking controller parameters need to be adjusted correctly to ensure stability and good performance, which is very time-consuming. Manual adjustment of walking controller parameters is usually suboptimal. Therefore, it is necessary to conduct in-depth research on methods for automatic adjustment of controller parameters. At present, due to the large dimensions of the parameter space of the three controllers, which makes it difficult to find the optimal solution, the existing walking controller adjustment methods only automatically adjust the inverse dynamics controller, while ignoring the parameter adjustment of the inverse kinematics controller and the active impedance controller, which leads to a decrease in the parameter accuracy of the walking controller. Summary of the invention

[0005] The purpose of the present invention is to provide a parameter automatic adjustment method for a biped robot walking controller, which improves the parameter accuracy of the walking controller by adjusting an inverse kinematics controller, an inverse dynamics controller and an active impedance controller.

[0006] The present invention adopts the following technical solution: a parameter automatic adjustment method applied to a biped robot walking controller, comprising the following steps:

[0007] Decomposing walking controller parameters into first inverse kinematics controller parameters, first inverse dynamics controller parameters and first active impedance controller parameters;

[0008] Turn off the inverse kinematics controller and the active impedance controller of the walking controller, optimize the parameters of the first inverse kinematics controller by using the Bayesian optimization method, and obtain the parameters of the second inverse kinematics controller;

[0009] Keeping the parameters of the second inverse kinematics controller unchanged, alternately optimizing the parameters of the first inverse dynamics controller and the parameters of the first active impedance controller by using the Bayesian optimization method, so as to obtain the parameters of the second inverse dynamics controller and the second active impedance controller;

[0010] The optimized walking controller parameters are obtained by combining the second inverse kinematics controller parameters, the second inverse dynamics controller parameters and the second active impedance controller parameters.

[0011] Preferably, optimizing the first inverse kinematics controller parameters using the Bayesian optimization method includes:

[0012] Generate n sets of parameter sets K within the parameter value range of the inverse kinematics controller IK,j ; where j = 1, 2, ..., n;

[0013] Use each set of parameters to control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error;

[0014] Generate a data set based on the parameter set and its corresponding error;

[0015] Based on the Bayesian optimization method, the probability model is trained using the data set to obtain the Bayesian posterior probability corresponding to each set of parameters;

[0016] The second inverse kinematics controller parameters are generated according to the Bayesian posterior probability.

[0017] Preferably, after obtaining the Bayesian posterior probability corresponding to each parameter set, the following step is further included:

[0018] Expand the data set and use the expanded data set to train the learning probability model.

[0019] Preferably, the expanded data set includes:

[0020] Continue to generate new parameter sets using the acquisition function;

[0021] Augment the dataset with the new parameter set.

[0022] Preferably, alternately optimizing the first inverse dynamics controller parameters and the first active impedance controller parameters using the Bayesian optimization method comprises:

[0023] Turn on the inverse kinematics controller, inverse dynamics controller, and active impedance controller at the same time;

[0024] Initialize the first inverse dynamics controller parameters and the first active impedance controller parameters, control the bipedal robot to complete a fixed motion trajectory, and calculate the corresponding error y min ;

[0025] Keeping the parameters of the first active impedance controller unchanged, optimizing the parameters of the first inverse dynamics controller based on the Bayesian optimization method;

[0026] Based on the first active impedance controller parameters and the optimized first inverse dynamics controller parameters, the biped robot is controlled to complete a fixed motion trajectory and the corresponding error is calculated. in, is the optimized first inverse dynamics controller parameter, K IM is the first active impedance controller parameter;

[0027] when When As the new first inverse dynamics controller parameters, As a new min ;

[0028] Keeping the new first inverse dynamics controller parameters unchanged, optimizing the first active impedance controller parameters based on the Bayesian optimization method;

[0029] Based on the new first inverse dynamics controller parameters and the optimized first active impedance controller parameters Control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error

[0030] when Less than the new y min When As the new first active impedance controller parameter, As a new min ;

[0031] According to the above method, the parameters of the first active impedance controller and the parameters of the first inverse dynamics controller are alternately optimized until the optimization termination condition is met, and the parameters of the second inverse dynamics controller are obtained. and the second active impedance controller parameters

[0032] Preferably, the error is calculated using the following objective function:

[0033] J(K)=r c +r lf +r rf +L,

[0034] Where J(K) is the error based on the controller parameters K, which includes the inverse kinematics controller parameters, inverse dynamics controller parameters and active impedance controller parameters, r c is the tracking error of the center of mass of the biped robot, r lf is the tracking error of the left foot position of the biped robot, r rfis the right foot position tracking error of the biped robot, and L is the fall penalty term.

[0035] Preferably, the center of mass position tracking error is calculated by the following method:

[0036]

[0037] in, represents the root mean square error of the center of mass tracking along the i-axis, i = x, y or z, is the maximum tracking error along the i-axis.

[0038] Preferably, the value of L is as follows:

[0039]

[0040] Among them, the evaluation is not interrupted, which means that the bipedal robot completes the fixed motion trajectory, and the evaluation is interrupted, which means that the bipedal robot does not complete the fixed motion trajectory. To evaluate the interruption time, T is the time that the biped robot plans to complete the fixed motion trajectory.

[0041] Preferably, the conditions for evaluating the interruption are:

[0042] The rotation of the torso exceeds the threshold or the center of mass height is below the threshold.

[0043] Another technical solution of the present invention: a parameter automatic adjustment device applied to a bipedal robot walking controller, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the above-mentioned parameter automatic adjustment method applied to a bipedal robot walking controller when executing the computer program.

[0044] The beneficial effects of the present invention are as follows: the present invention realizes the joint automatic adjustment of parameters of the three controllers, namely the inverse kinematics controller, the inverse dynamics controller and the active impedance controller, divides the high-dimensional parameter space into low-dimensional subspaces to solve the problem of difficult optimization solutions in high-dimensional conditions, the parameters of the inverse kinematics controller are automatically adjusted using Bayesian optimization, and the parameters of the inverse dynamics controller and the active impedance controller are automatically adjusted using alternating Bayesian optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic diagram of a parameter automatic adjustment method applied to a biped robot walking controller according to an embodiment of the present invention;

[0046] Figure 2 is the center of mass trajectory in the motion trajectory in the embodiment of the present invention;

[0047] Figure 3 is the left foot trajectory in the motion trajectory in the embodiment of the present invention;

[0048] Figure 4 is the right foot trajectory in the motion trajectory in the embodiment of the present invention;

[0049] Figure 5 Graph showing the relationship between the objective function value and the number of evaluations in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] At present, some controller parameter automatic adjustment methods have been proposed, but these methods do not consider the problem of automatic adjustment of parameters of inverse kinematics controller and active impedance controller. This is because when the three controllers are automatically adjusted and optimized at the same time, the dimension of the parameter space will be greatly expanded, which will make it difficult to find the optimal solution. Therefore, how to automatically adjust the control parameters of the three controllers of inverse kinematics, inverse dynamics and active impedance is a problem that technicians in this field need to solve.

[0052] The present invention discloses a parameter automatic adjustment method for a biped robot walking controller, comprising the following steps: decomposing a walking controller parameter K into a first inverse kinematics controller parameter K IK , the first inverse dynamics controller parameter K ID and the first active impedance controller parameter K IM ; Turn off the inverse kinematics controller and active impedance controller of the walking controller, and use the Bayesian optimization method to optimize the first inverse kinematics controller parameter K IK , get the second inverse kinematics controller parameters Maintain the second inverse kinematics controller parameters The first inverse dynamics controller parameters and the first active impedance controller parameters are alternately optimized by using the Bayesian optimization method to obtain the second inverse dynamics controller parameters and the second active impedance controller parameters; the second inverse kinematics controller parameters, the second inverse dynamics controller parameters and the second active impedance controller parameters are combined to obtain the optimized walking controller parameters.

[0053] The present invention realizes the joint automatic adjustment of parameters of three controllers: an inverse kinematics controller, an inverse dynamics controller and an active impedance controller, divides the high-dimensional parameter space into low-dimensional subspaces to solve the problem of difficult optimization solutions in high-dimensional conditions, the parameters of the inverse kinematics controller are automatically adjusted using Bayesian optimization, and the parameters of the inverse dynamics controller and the active impedance controller are automatically adjusted using alternating Bayesian optimization.

[0054] In addition, the present invention decomposes the high-dimensional parameter space and proposes an automatic adjustment method for the parameters of a bipedal robot walking controller that is easy to optimize and solve while taking into account three controllers, freeing researchers from tedious parameter adjustment and achieving better control performance than manual adjustment.

[0055] In one embodiment, optimizing the first inverse kinematics controller parameters using a Bayesian optimization method includes:

[0056] Turn off the inverse kinematics controller and active impedance controller, put the biped robot into position control state, and use the inverse kinematics controller alone to control the robot's walking. Generate n sets of parameter sets K within the parameter value range of the inverse kinematics controller IK,j ; Wherein, j = 1, 2, ..., n; each set of parameter sets is used to control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error.

[0057] In one embodiment, the fixed motion trajectory refers to the planned center of mass and two-foot trajectory. This embodiment adopts a method based on a linear inverted pendulum for planning. The planned center of mass trajectory, left foot trajectory, and right foot trajectory are respectively as follows: Figure 2 , 3 and 4.

[0058] As a specific implementation method, the error is calculated using the following objective function:

[0059] J(K)=r c +r lf +r rf +L,

[0060] Where J(K) is the error based on the controller parameters K, which includes the inverse kinematics controller parameters, inverse dynamics controller parameters and active impedance controller parameters, r c is the tracking error of the center of mass of the biped robot, r lf is the tracking error of the left foot position of the biped robot, r rf is the right foot position tracking error of the biped robot, and L is the fall penalty term.

[0061] Specifically, the center of mass position tracking error is calculated by the following method:

[0062]

[0063] in, represents the root mean square error of the center of mass tracking along the i-axis, i = x, y or z, is the maximum tracking error along the i-axis. The coordinate system can be the robot's foot coordinate system or the world coordinate system. The error is obtained by data collection during the execution of the actual robot in a simulation environment or a real environment. In this embodiment, Simscape Multibody is selected as the virtual environment.

[0064] More specifically, the value rule of L is:

[0065]

[0066] Among them, the evaluation is not interrupted, which means that the bipedal robot completes the fixed motion trajectory, and the evaluation is interrupted, which means that the bipedal robot does not complete the fixed motion trajectory. is the time for evaluating the interruption, and T is the time for the biped robot to complete a fixed motion trajectory. The conditions for evaluating the interruption are: the rotation of the trunk exceeds the threshold or the height of the center of mass is lower than the threshold.

[0067] More specifically, an inverse kinematics controller is used to control the center of mass and the swinging foot to track the planned position:

[0068]

[0069]

[0070] Where c represents the center of mass position, x represents the swing foot position, the superscript plan represents the value of the planned tracking, the superscript des represents the value input to the inverse kinematics solver, and the superscript real represents the real value of the current robot. c,IK (Inverse Kinematics Controller Center of Mass Scale Factor) and K f,IK (Inverse kinematics controller two-foot proportional coefficient) is the parameter that needs to be automatically adjusted, and their collection is the inverse kinematics controller parameter K IK , whose dimension is 6.

[0071] In the embodiment of the present invention, the inverse dynamics controller used controls the position of the center of mass and the two-foot tracking plan:

[0072]

[0073]

[0074] Where c represents the center of mass, P represents the position of the foot, the superscript plan represents the value of the planned tracking, the superscript des represents the value input to the inverse dynamics solver, and the superscript real represents the real value of the current robot. cp,ID (Inverse Dynamics Controller Center of Mass Proportional Coefficient), K cd,ID (inverse dynamics controller center of mass differential coefficient), K fp,ID (Inverse Dynamics Controller Foot Proportional Coefficient) and Kfd,ID (The differential coefficients of the two feet of the inverse dynamics controller) are the parameters that need to be automatically adjusted, and their collection is the inverse dynamics controller parameter K ID , whose dimension is 12.

[0075] In the embodiment of the present invention, the active impedance controller used is:

[0076]

[0077] Among them, τ represents the torque applied to the joint, τ FF is the feedforward term calculated by inverse dynamics, q IK is the angular position calculated by inverse kinematics. The superscript real indicates the real value from the robot. K p,IM (active impedance controller spring constant) and K d,IM (Active impedance controller damping coefficient) is the parameter that needs to be automatically adjusted, and their collection is the active impedance controller parameter K IM , is the time derivative of the angular position calculated by inverse kinematics, It is the differential of the actual joint angle position with respect to time during the robot evaluation process, and its dimension is 12.

[0078] After the above process, a data set is generated based on the parameter set and its corresponding error

[0079] Next, based on the Bayesian optimization method, the data set is used to train the learning probability model to obtain the Bayesian posterior probability corresponding to each set of parameter sets.

[0080] In this embodiment, a Gaussian process is used to perform regression learning on the objective function J(K):

[0081]

[0082] The prior mean m(K) is defined as zero. The kernel function k(K,K′) is Matern5 / 2:

[0083]

[0084]

[0085] Among them, K i and K j Two different sets of parameter values ​​are taken respectively, σ l and σ f are the definition parameters of the kernel function respectively.

[0086] Then, the second inverse kinematics controller parameters are generated according to the Bayesian posterior probability.

[0087] In another embodiment, after obtaining the Bayesian posterior probability corresponding to each parameter set, the method further includes: expanding the data set, and using the expanded data set to train the learning probability model. Expanding the data set includes: using the acquisition function to continue to generate new parameter sets; and expanding the data set with the new parameter sets.

[0088] Specifically, the acquisition function To determine K IK,i+1 ,Right now In this embodiment, the acquisition function of expected improvement is adopted. Then K IK,i+1 Complete the fixed motion trajectory and calculate the corresponding error J(K IK,i+1 ), and then add the new data to the original data set, that is

[0089] Next, repeat the above steps until the termination condition is met, and then calculate the second inverse kinematics controller parameters obtained by automatic adjustment according to the probability model at the termination time. In this embodiment, the termination condition is that the number of repetitions reaches 150 times.

[0090] like Figure 5 The figure shows the relationship between the objective function value and the number of evaluations. As shown in the figure, within 10 evaluations, the robot can achieve stable walking through position control. Within 50 evaluations, the objective value will be lower than the target value obtained in the manual tuning parameters. Finally, the objective function value reaches its minimum value within 100 evaluations.

[0091] In one embodiment, alternately optimizing the first inverse kinematics controller parameters and the first active impedance controller parameters using the Bayesian optimization method includes: simultaneously turning on the inverse kinematics controller, the inverse kinematics controller, and the active impedance controller, and placing the biped robot in a torque control state. The inverse kinematics controller parameters are fixed to

[0092] Initialize the first inverse dynamics controller parameters and the first active impedance controller parameters, that is, assign K ID , K IM Initial value, K IM ←K IM,0 , K IM ←K IM,0 , control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error y min , that is, y min ←J(K ID,0 ,K IM,0 ).

[0093] Next, keep the first active impedance controller parameter K IM The first inverse dynamics controller parameters are optimized based on the Bayesian optimization method. Based on the first active impedance controller parameters and the optimized first inverse dynamics controller parameters, the biped robot is controlled to complete a fixed motion trajectory and the corresponding error is calculated. in, is the optimized first inverse dynamics controller parameter, K IM is the first active impedance controller parameter.

[0094] In this embodiment, Bayesian optimization uses Gaussian process to perform regression learning on the objective function, and the prior mean m(K) is defined as zero. The kernel function k(K, K′) is Matern5 / 2, and the termination condition is that the number of evaluations reaches 100.

[0095] when When As the new first inverse dynamics controller parameters, As a new min ,Right now

[0096] Next, keep the new first inverse dynamics controller parameters unchanged, optimizing the first active impedance controller parameters based on the Bayesian optimization method; based on the new first inverse dynamics controller parameters and the optimized first active impedance controller parameters Control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error

[0097] when Less than the new y min When As the new first active impedance controller parameter, As a new min .

[0098] Similarly, in this embodiment, Bayesian optimization uses Gaussian process to perform regression learning on the objective function, the prior mean m(K) is defined as zero, the kernel function k(K, K′) is Matern5 / 2, and the termination condition is that the number of evaluations reaches 100.

[0099] Then, according to the above method, the first active impedance controller parameters and the first inverse dynamics controller parameters are alternately optimized until the optimization termination condition is met, and the second inverse dynamics controller parameters are obtained. and the second active impedance controller parameters

[0100] In this embodiment, the termination condition is that the number of repetitions reaches 4. Maintain the first active impedance controller parameter KIM The value of the objective function remains unchanged and is 0.2196.

[0101] Another technical solution of the present invention: a parameter automatic adjustment device applied to a bipedal robot walking controller, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the above-mentioned parameter automatic adjustment method applied to a bipedal robot walking controller when executing the computer program.

[0102] The above-mentioned device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The device may include but is not limited to a processor and a memory. Those skilled in the art will appreciate that the device may include more or fewer components, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0103] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0104] In some embodiments, the memory may be an internal storage unit of the device, such as a hard disk or memory of the device. In other embodiments, the memory may also be an external storage device of the device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the device. Further, the memory may also include both an internal storage unit of the device and an external storage device. The memory is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory may also be used to temporarily store data that has been output or is to be output.

[0105] It should be noted that the specific content of the above-mentioned device is based on the same concept as the embodiment of the method of the present invention. Its specific functions and technical effects can be found in the method embodiment part, and will not be repeated here.

[0106] In the above embodiments, the description of each embodiment has its own emphasis. For the parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. Those of ordinary skill in the art can appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented with electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0107] In summary, inverse kinematics controller, inverse dynamics controller and active impedance controller are widely used in the motion control of bipedal robots. There are a large number of control parameters that need to be adjusted in these three controllers. Manual adjustment of them requires a lot of time and rich experience, so it is necessary to design a method for automatic adjustment of controller parameters. The parameters of these three controllers are automatically adjusted jointly. However, the total dimension of the parameters of these three controllers is very high, which affects the convergence and computational efficiency of the optimization algorithm. The existing methods are limited to the automatic adjustment of one of the controller parameters.

[0108] To address this problem, the present invention utilizes the prior knowledge of the three controllers and invents a method for joint automatic adjustment of parameters that can reduce the number of dimensions optimized at the same time. First, the inverse kinematics controller is decoupled from the joint tuning. Taking advantage of the fact that the inverse kinematics controller alone can drive the robot to walk, the inverse kinematics controller parameters are adjusted individually to obtain the optimal inverse kinematics controller parameters. Afterwards, the optimal inverse kinematics controller parameters are substituted into the joint control, and the inverse dynamics controller and active impedance controller parameters are jointly adjusted in an alternating optimization manner, and finally the joint optimal parameters of the three controllers are obtained. The method of the present invention solves the problem of joint adjustment of the parameters of the three controllers in the case of low-dimensional optimization, and improves the convergence and computational efficiency of the optimization algorithm.

Claims

1. A method for automatically adjusting parameters of a biped robot walking controller, characterized in that: The following steps are involved: Decomposing walking controller parameters into first inverse kinematics controller parameters, first inverse dynamics controller parameters and first active impedance controller parameters; Turning off the inverse kinematics controller and the active impedance controller of the walking controller, optimizing the first inverse kinematics controller parameters by using the Bayesian optimization method, and obtaining the second inverse kinematics controller parameters; Keeping the parameters of the second inverse kinematics controller unchanged, alternately optimizing the parameters of the first inverse dynamics controller and the parameters of the first active impedance controller using the Bayesian optimization method to obtain the parameters of the second inverse dynamics controller and the second active impedance controller; Combining the second inverse kinematics controller parameters, the second inverse dynamics controller parameters, and the second active impedance controller parameters to obtain optimized walking controller parameters; Optimizing the first inverse kinematics controller parameters using the Bayesian optimization method includes: Generate n sets of parameter sets K within the parameter value range of the inverse kinematics controller IK,j ; where j = 1, 2, ..., n; Use each set of parameters to control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error; Generate a data set according to the parameter set and its corresponding error; Based on the Bayesian optimization method, the data set is used to train the learning probability model to obtain the Bayesian posterior probability corresponding to each parameter set; generating a second inverse kinematics controller parameter according to the Bayesian posterior probability; Using the Bayesian optimization method to alternately optimize the first inverse dynamics controller parameters and the first active impedance controller parameters includes: Turn on the inverse kinematics controller, inverse dynamics controller, and active impedance controller at the same time; Initialize the first inverse dynamics controller parameters and the first active impedance controller parameters, control the bipedal robot to complete a fixed motion trajectory, and calculate the corresponding error y min ; Keeping the parameters of the first active impedance controller unchanged, optimizing the parameters of the first inverse dynamics controller based on the Bayesian optimization method; Based on the first active impedance controller parameters and the optimized first inverse dynamics controller parameters, the biped robot is controlled to complete a fixed motion trajectory and the corresponding error is calculated. in, is the optimized first inverse dynamics controller parameter, K IM is the first active impedance controller parameter; when When As the new first inverse dynamics controller parameters, As a new min ; Keeping the new first inverse dynamics controller parameters unchanged, optimizing the first active impedance controller parameters based on the Bayesian optimization method; Based on the new first inverse dynamics controller parameters and the optimized first active impedance controller parameters Control the bipedal robot to complete a fixed motion trajectory and calculate the corresponding error when Less than the new y min When As the new first active impedance controller parameter, As a new min ; According to the above method, the parameters of the first active impedance controller and the parameters of the first inverse dynamics controller are alternately optimized until the optimization termination condition is met, and the parameters of the second inverse dynamics controller are obtained. and the second active impedance controller parameters 2. The method for automatically adjusting parameters of a biped robot walking controller according to claim 1, characterized in that: After obtaining the Bayesian posterior probability corresponding to each set of parameters, it also includes: The data set is expanded, and the expanded data set is used to train a learning probability model.

3. The method for automatically adjusting parameters of a biped robot walking controller according to claim 2, characterized in that: The dataset is expanded to include: Continue to generate new parameter sets using the acquisition function; The data set is augmented with a new set of parameters.

4. The method for automatically adjusting parameters of a biped robot walking controller according to claim 3, characterized in that: The error is calculated using the following objective function: J(K)=r c +r lf +r rf +L, Where J(K) is the error based on the controller parameters K, which includes the inverse kinematics controller parameters, inverse dynamics controller parameters and active impedance controller parameters, r c is the tracking error of the center of mass of the biped robot, r lf is the tracking error of the left foot position of the biped robot, r rf is the right foot position tracking error of the biped robot, and L is the fall penalty term.

5. The method for automatically adjusting parameters of a biped robot walking controller according to claim 4, characterized in that: The center of mass position tracking error is calculated by the following method: in, represents the root mean square error of the center of mass tracking along the i-axis, i = x, y or z, is the maximum tracking error along the i-axis.

6. The method for automatically adjusting parameters of a biped robot walking controller according to claim 5, characterized in that: The value of L is as follows: Among them, the evaluation is not interrupted, which means that the bipedal robot completes the fixed motion trajectory, and the evaluation is interrupted, which means that the bipedal robot does not complete the fixed motion trajectory. To evaluate the interruption time, T is the time that the biped robot plans to complete the fixed motion trajectory.

7. The method for automatically adjusting parameters of a biped robot walking controller according to claim 6, characterized in that: The conditions for interrupting the evaluation are: The rotation of the torso exceeds the threshold or the center of mass height is below the threshold.

8. A parameter automatic adjustment device for a biped robot walking controller, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for automatic parameter adjustment applied to a bipedal robot walking controller as described in any one of claims 1 to 7 is implemented.

Images