A method and apparatus for accurate speed control of a wheeled robot

Abstract

The present disclosure relates to a wheeled robot speed accurate control method and device. The method comprises: collecting a deep deterministic policy gradient algorithm DDPG state quantity of a driving motor of a wheeled robot; based on a preset DDPG model, inputting the DDPG state quantity to generate a model-free adaptive control MFAC parameter; based on a preset MFAC model, inputting the MFAC parameter and a target speed of the wheeled robot to generate a PWM control signal of the driving motor of the wheeled robot; and based on the PWM control signal, driving the driving motor of the wheeled robot to complete accurate control of the speed of the wheeled robot. The present disclosure deploys the MFAC algorithm to the robot speed control, and realizes accurate control of the robot speed based on the model-free adaptive control of the robot speed control based on the DDPG.

Description

Technical Field

[0001] This disclosure relates to the field of robot control, and more specifically, to a method and apparatus for precise speed control of a wheeled robot. Background Technology

[0002] In robot movement, striking a balance between safety and efficiency is crucial. Rapid movement increases efficiency, but due to the unknown environment, safety decreases. Conversely, prioritizing safety reduces efficiency. This is especially true during navigation, where unexpected obstacles and sudden shrinking spaces can easily occur. Therefore, balancing speed control requires extremely high accuracy, speed, and stability in robot speed response. Wheeled robots, commonly used in robot navigation, offer advantages such as high energy efficiency, mechanical simplicity, and well-developed control systems. One of the most fundamental and important tasks of a robot is to navigate from a starting point to a destination using collision-free motion. This typically involves path planning and path tracking, necessitating precise control of the robot's speed.

[0003] Existing technologies include: linear controllers using PID controllers to track the heading of synchronous robots; proportional-integral (PI) controllers with feedforward compensators to track desired paths; adaptive sliding mode dynamic controllers; and PID-based adaptive tracking controllers, whose control laws are based on Lyapunov stability theory. However, all of the controllers mentioned above require prior implementation in simulation or with the aid of an online computer. Furthermore, these algorithms require significant computation time and large amounts of memory. Therefore, implementing these algorithms in microcontrollers with limited memory is impractical and difficult to implement in real-world robots. Moreover, for traditional PID controllers, firstly, for speed control, the speed response is prone to oscillations during the steady-state phase at low speeds. Secondly, for PID parameter tuning, conventional methods include empirical trial-and-error, critical proportional gain, decay curve methods, immune algorithms, and MEA parameter tuning methods, but all have significant limitations, are generally complex and cumbersome, and often rely heavily on human experience. Currently, the most commonly used parameter self-tuning algorithm with relatively good tuning results is the swarm intelligence optimization algorithm, but its optimization process is prone to getting trapped in local optima, failing to achieve good convergence.

[0004] Therefore, one or more methods are needed to solve the above problems.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this disclosure is to provide a method and apparatus for precise speed control of a wheeled robot, thereby overcoming, at least to some extent, one or more problems caused by the limitations and defects of related technologies.

[0007] According to one aspect of this disclosure, a method for precise speed control of a wheeled robot is provided, comprising:

[0008] The policy gradient algorithm DDPG is used to collect the depth deterministic state variables of the drive motors of a wheeled robot.

[0009] Based on the preset DDPG model, the DDPG state variables are used as input to generate model-free adaptive control (MFAC) parameters.

[0010] Based on the preset MFAC model, the PWM control signal for the drive motor of the wheeled robot is generated using the MFAC parameters and the target speed of the wheeled robot as input.

[0011] The PWM control signal is used to drive the drive motor of the wheeled robot, thereby achieving precise speed control of the wheeled robot.

[0012] In one exemplary embodiment of this disclosure, the method further includes:

[0013] Based on the encoder of the drive motor of the wheeled robot, the speed of the drive motor is measured, and the state variables of the deep deterministic policy gradient algorithm (DDPG) of the drive motor of the wheeled robot are collected. The DDPG state variables include speed response overshoot, speed response rise time, and speed response settling time.

[0014] In one exemplary embodiment of this disclosure, the method further includes establishing a Deep Deterministic Policy Gradient Algorithm (DDPG) model based on a state space module, an action space module, a reward function, and a loss function:

[0015] The state space module and action space module use the DDPG state variables as input to generate model-free adaptive control (MFAC) parameters.

[0016] The reward function is a reward function generated based on the DDPG state quantity and the DDPG state quantity threshold.

[0017] The loss function is used to update the parameter values ​​of the target network and the target policy network in the DDPG algorithm.

[0018] In one exemplary embodiment of this disclosure, the method further includes:

[0019] The preset DDPG model is trained based on the preset training group data;

[0020] When the parameter values ​​output by the preset DDPG model converge, the training of the preset DDPG model is complete.

[0021] In one exemplary embodiment of this disclosure, the method further includes:

[0022] The preset reward function of the DDPG model is r = r1 + r2 + r3 + r4.

[0023] in,

[0024]

[0025]

[0026]

[0027] so represents the speed response overshoot magnitude, ot represents the speed response overshoot threshold, str represents the speed response rise time, trt represents the speed response rise time threshold, sts represents the speed response settling time, tst represents the speed response settling time threshold, ssp represents the preset given speed magnitude, and hf represents the maximum speed value of the speed response curve.

[0028] In one exemplary embodiment of this disclosure, the method further includes:

[0029] The preset MFAC model is

[0030]

[0031] Where η is the step size factor, μ is the weight factor, ρ is the adjustable step size sequence, and λ is the adjustable weight coefficient. Here, y is a pseudo-partial derivative, Δu is the difference in system input between two time points, and y is a pseudo-partial derivative. * Let y be the system's target output value, y be the system's input value at the current moment, and ε be a pre-defined sufficiently small positive number. The initial value of .

[0032] In one aspect of this disclosure, a precise speed control device for a wheeled robot is provided, comprising:

[0033] The state acquisition module is used to acquire the state variables of the depth deterministic policy gradient algorithm (DDPG) for the drive motors of the wheeled robot.

[0034] The parameter generation module is used to generate model-free adaptive control (MFAC) parameters based on a preset DDPG model and with the DDPG state variables as input.

[0035] The control signal generation module is used to generate PWM control signals for the drive motor of the wheeled robot based on a preset MFAC model, using the MFAC parameters and the target speed of the wheeled robot as inputs.

[0036] The drive control module is used to drive the drive motor of the wheeled robot based on the PWM control signal, thereby achieving precise control of the speed of the wheeled robot.

[0037] An exemplary embodiment of this disclosure provides a method for precise speed control of a wheeled robot. The method includes: acquiring depth-deterministic policy gradient algorithm (DDPG) state variables of the wheeled robot's drive motors; generating model-free adaptive control (MFAC) parameters based on a preset DDPG model, using the DDPG state variables as input; generating a PWM control signal for the wheeled robot's drive motors based on the preset MFAC model, using the MFAC parameters and the wheeled robot's target speed as input; and driving the wheeled robot's drive motors based on the PWM control signal to achieve precise speed control of the wheeled robot. This disclosure achieves precise speed control of the robot by deploying the MFAC algorithm into robot speed control and implementing model-free adaptive control based on DDPG.

[0038] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0039] The above and other features and advantages of this disclosure will become more apparent from the detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0040] Figure 1 A flowchart of a method for precise speed control of a wheeled robot according to an exemplary embodiment of the present disclosure is shown;

[0041] Figure 2 This illustration shows a technical roadmap of a method for precise speed control of a wheeled robot according to an exemplary embodiment of the present disclosure;

[0042] Figure 3 A brushless DC motor control model diagram of a precise speed control method for a wheeled robot according to an exemplary embodiment of the present disclosure is shown.

[0043] Figure 4 A flowchart illustrating the adaptive adjustment of MFAC parameters based on DDPG in a precise speed control method for a wheeled robot according to an exemplary embodiment of the present disclosure is shown.

[0044] Figure 5A schematic block diagram of a wheeled robot speed precision control device according to an exemplary embodiment of the present disclosure is shown. Detailed Implementation

[0045] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0046] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details described, or other methods, components, materials, apparatuses, steps, etc., can be employed. In other instances, well-known structures, methods, apparatuses, implementations, materials, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0047] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more software-hardened modules, or in different network and / or processor devices and / or microcontroller devices.

[0048] In this example embodiment, a method for precise speed control of a wheeled robot is first provided; see reference. Figure 1 As shown, the method for precise speed control of a wheeled robot may include the following steps:

[0049] Step S110: Collect the depth deterministic policy gradient algorithm (DDPG) state variables of the drive motors of the wheeled robot.

[0050] Step S120: Based on the preset DDPG model, using the DDPG state variables as input, generate model-free adaptive control (MFAC) parameters.

[0051] Step S130: Based on the preset MFAC model, using the MFAC parameters and the target speed of the wheeled robot as inputs, generate the PWM control signal for the drive motor of the wheeled robot.

[0052] Step S140: Drive the drive motor of the wheeled robot based on the PWM control signal to achieve precise control of the speed of the wheeled robot.

[0053] An exemplary embodiment of this disclosure provides a method for precise speed control of a wheeled robot. The method includes: acquiring depth-deterministic policy gradient algorithm (DDPG) state variables of the wheeled robot's drive motors; generating model-free adaptive control (MFAC) parameters based on a preset DDPG model, using the DDPG state variables as input; generating a PWM control signal for the wheeled robot's drive motors based on the preset MFAC model, using the MFAC parameters and the wheeled robot's target speed as input; and driving the wheeled robot's drive motors based on the PWM control signal to achieve precise speed control of the wheeled robot. This disclosure achieves precise speed control of the robot by deploying the MFAC algorithm into robot speed control and implementing model-free adaptive control based on DDPG.

[0054] like Figure 2 The diagram shown is a technical roadmap for precise speed control of a robot according to this disclosure. The following will further explain a method for precise speed control of a wheeled robot in this example embodiment.

[0055] Example 1:

[0056] In step S110, the depth deterministic policy gradient algorithm (DDPG) state variables of the drive motors of the wheeled robot can be collected.

[0057] In this example embodiment, the method further includes:

[0058] Based on the encoder of the drive motor of the wheeled robot, the speed of the drive motor is measured, and the state variables of the deep deterministic policy gradient algorithm (DDPG) of the drive motor of the wheeled robot are collected. The DDPG state variables include speed response overshoot, speed response rise time, and speed response settling time.

[0059] In step S120, model-free adaptive control (MFAC) parameters can be generated based on a preset DDPG model, using the DDPG state variables as input.

[0060] In this example embodiment, the method further includes establishing a Deep Deterministic Policy Gradient Algorithm (DDPG) model based on a state space module, an action space module, a reward function, and a loss function:

[0061] The state space module and action space module use the DDPG state variables as input to generate model-free adaptive control (MFAC) parameters.

[0062] The reward function is a reward function generated based on the DDPG state quantity and the DDPG state quantity threshold.

[0063] The loss function is used to update the parameter values ​​of the target network and the target policy network in the DDPG algorithm.

[0064] In this example embodiment, the method further includes:

[0065] The preset DDPG model is trained based on the preset training group data;

[0066] When the parameter values ​​output by the preset DDPG model converge, the training of the preset DDPG model is complete.

[0067] In this example embodiment, the method further includes:

[0068] The preset reward function of the DDPG model is r = r1 + r2 + r3 + r4.

[0069] in:

[0070]

[0071]

[0072]

[0073]

[0074] so represents the speed response overshoot magnitude, ot represents the speed response overshoot threshold, str represents the speed response rise time, trt represents the speed response rise time threshold, sts represents the speed response settling time, tst represents the speed response settling time threshold, ssp represents the preset given speed magnitude, and hf represents the maximum speed value of the speed response curve.

[0075] In step S130, a PWM control signal for the drive motor of the wheeled robot can be generated based on a preset MFAC model, using the MFAC parameters and the target speed of the wheeled robot as inputs.

[0076] In this example embodiment, the method further includes:

[0077] The preset MFAC model is

[0078]

[0079] Where η is the step size factor, μ is the weight factor, ρ is the adjustable step size sequence, and λ is the adjustable weight coefficient. Here, y is a pseudo-partial derivative, Δu is the difference in system input between two time points, and y is a pseudo-partial derivative. * Let y be the system's target output value, y be the system's input value at the current moment, and ε be a pre-defined sufficiently small positive number. for The initial value of .

[0080] In step S140, the drive motor of the wheeled robot can be driven based on the PWM control signal to achieve precise control of the wheeled robot's speed.

[0081] In this example embodiment, the present disclosure discloses a robot speed control system based on DDPG and using MFAC as the controller, with the controlled object being the robot's brushless DC motor. The purpose is to adjust the DDPG network parameters based on the quality of the motor's speed response during control. Through continuous feedback self-learning, the neural network can be trained to develop a set of parameter values ​​for the MFAC controller that optimizes the motor's speed response, ultimately improving the stability, accuracy, and speed of the robot's speed response.

[0082] Example 2:

[0083] In the embodiments of this example, as Figure 3 The diagram shown is a control model diagram of a brushless DC motor. This disclosure is mainly based on the chassis controller designed by EPRobot for its intelligent vehicle, which is based on an ARM Cortex-M4 microcontroller. The A4950 chip is used as the motor driver chip. The motor control algorithm is written in the microcontroller to control the speed of the brushless DC motor. The program is written using the RT-Thread micro real-time operating system. The control algorithm uses the MFAC compact format model-free adaptive control algorithm to accurately control the speed of the vehicle motor.

[0084] The state-space equation for the brushless DC motor of a smart car is:

[0085]

[0086]

[0087] Where E a armature voltage, I a For armature current, R a For armature resistance, L a For armature inductance, J m The equivalent rotational inertia of the motor and the load reflected onto the motor shaft, K b The back electromotive force constant, ω m Where K is the motor shaft angle, B is the motor torque constant, and K is the motor torque constant. m It is the equivalent adhesive friction coefficient of the motor and the load reflected onto the motor shaft.

[0088] The motor's state-space equations reveal it to be a single-input single-output (SISO) discrete-time nonlinear system. When applying the MFAC algorithm as the controller, the system must satisfy three corresponding assumptions.

[0089] After deriving the state space equations of the motor, it was proven that it satisfies Assumption 1 and Assumption 2. Assumption 3 is a constraint on the output change of the motor speed control system, i.e., the output speed. The magnitude of the input control signal is finite, and the magnitude of the output speed value is also finite. Therefore, the system is a generalized Lipschitz system. Thus, the motor speed controller system satisfies all three assumptions.

[0090] Theorem: For the above motor speed controller system, if the three assumptions are satisfied, then there must exist a vector called the pseudo-gradient vector. So that when Δu(k)≠0, we have

[0091]

[0092] Identify The goal is to make To approximate y(k+1)-y(k) infinitely, while ensuring that the value of the pseudo-partial derivative does not change too rapidly, the following estimation criterion function is designed:

[0093]

[0094] in The introduction of the term penalizes the parameter. The parameter estimation algorithm derived from the criterion function should be able to track time-varying parameters, considering only the k-th sampling time. The derivative yields the following equation:

[0095]

[0096] Where μ>0 is the weight factor. η∈(0,1] is the added step size factor, which aims to make the algorithm more flexible and general.

[0097] In this example embodiment, the control objective of this disclosure is to apply a control action u(k) to the motor speed control system at time k, so that the system output is the desired value y. * (k+1), therefore the control input criterion function is listed as follows:

[0098] J(u(k))=|y * (k+1)-y(k+1)| 2 +λ|u(k)-u(k-1)| 2 In this criterion function, due to the term λ|u(k)-u(k-1)| 2 The introduction of this control variable restricts the change of the control quantity and can overcome steady-state error.

[0099] Will Substituting into the above equation, taking the derivative of u(k) and setting it equal to zero, we get:

[0100]

[0101] Where ρ is an adjustable step size sequence and λ is an adjustable weight coefficient.

[0102] Based on the above, the complete model-free adaptive controller designed according to the object in this paper is shown below:

[0103]

[0104] Parameter tuning for MFAC controllers is also a crucial issue. Manual tuning is time-consuming and inefficient, as it's difficult to grasp the underlying patterns of the parameters. When using intelligent swarm optimization algorithms, such as Particle Swarm Optimization (PSO), the tuning process is prone to getting stuck in local optima, failing to find optimal parameter values. Using offline robot motion data as input for parameter identification, such as PID parameter identification, requires manual adjustment of system-related parameters, introducing human experience and significant uncertainty. Subsequent input of the identified parameters into the algorithm reveals poor speed response. Therefore, using system-based parameter tuning suffers from instability and limited applicability. Deep reinforcement learning agents interact with the environment, generating new states, while the environment provides rewards. This cycle continues, with the agent constantly interacting and generating more new data. This new data is then used to modify the agent's action strategy. After several iterations, the agent learns the necessary action strategy to complete the task, and with sufficient training, the reward value improves. When the system stabilizes, the corresponding actions also tend to stabilize, and the corresponding parameters converge to their optimal values. Therefore, deep reinforcement learning eliminates the limitations of human experience, and because it is a mechanism of continuous training and exploration, it avoids the problem of getting stuck in local optima during training.

[0105] In this example embodiment, considering the characteristics of intelligent vehicles, the DDPG-related metrics are designed as follows:

[0106] Controller parameter tuning environment design: This patent aims to achieve automatic parameter adjustment. The intelligent agent corresponds to the parameter change process of the controller. Its target is a brushless DC motor. Therefore, the environment is set to the entire speed response process of the motor under the control of the controller.

[0107] State Space: During algorithm execution, the completion of an action directly affects the parameter values, which in turn influence the three speed response metrics. Therefore, the parameter values ​​after action completion are used as state values. The state space module serves two purposes: firstly, it directly reflects the control algorithm's response to the given target; secondly, it calculates the reward value in the DDPG algorithm, using this reward value to guide the DDPG algorithm's self-learning optimization to achieve better control performance.

[0108] Action space: The action space of the entire system enables automated parameter adjustment. By continuously optimizing the action values ​​generated by DDPG, the speed response of the system is controlled, thereby continuously optimizing the control results. Therefore, the action is defined as the dynamic change of the controller's parameter values. For a model-free adaptive controller, the action is the dynamic change of the controller's four parameter values ​​η, μ, ρ, and λ.

[0109] Reward Function: The convergence of deep reinforcement learning depends on a reasonable reward setting. This paper combines the three state values ​​of the velocity response to set the reward function, making the velocity response more obvious after each parameter adjustment. Considering the three velocity response values—overshoot, rise time (t_r), and settling time (t_s)—a weight function is set for these three values ​​to serve as the reward function, aiming to maximize the reward function value when the overshoot is small, the rise time is short, and the settling time is short. Therefore, the reward function is set as follows, where so is the overshoot magnitude, ot is the overshoot threshold, str is the rise time, trt is the rise time threshold, sts is the settling time, tst is the settling time threshold, ssp is the set velocity magnitude, and hf is the maximum velocity value of the velocity response curve.

[0110]

[0111]

[0112]

[0113]

[0114] r = r1 + r2 + r3 + r4

[0115] In this example embodiment, for the speed control optimization problem and the MFAC controller parameter adjustment problem, the corresponding controller input / output values ​​and system parameters such as the state, action, and reward corresponding to the DDPG algorithm on this intelligent vehicle platform are defined as shown in Table 1:

[0116]

[0117]

[0118] Table 1: Definition of Overall Structural Parameters of Robot Speed ​​Control System. In this example embodiment, the loss function for training the DDPG network is as follows:

[0119]

[0120] Loss=-Q θ (s, a)

[0121] The first formula is the loss function of the Q-target network, where w is the parameter value to be trained and updated by the Q-network, s is the current state, a is the current action, s′ is the previous state, a′ is the previous action, and Q... w (s, a) represents the Q-value generated by the Q-network at the current moment under the current state and current action, and r represents the reward value under the previous action. The first formula represents the Q-value generated by the network at the previous time step. The second formula is the loss function of the policy network, where θ represents the parameter values ​​to be trained and updated by the policy network, s represents the current state, a represents the current action, and Q represents the loss function of the policy network. θ (s, a) represents the Q-value under the current θ parameters. Since the policy network's role is to determine the superiority of the Q-value, -Q is directly used. θ (s, a) is used as the loss function. The network parameters are updated based on the value of the loss function, and ultimately the network parameters are continuously optimized to achieve our predetermined target value.

[0122] In the embodiments of this example, as Figure 4 The diagram shows the flowchart of adaptive adjustment of MFAC parameters based on DDPG. After the action values ​​generated during training are converted into specific MFAC parameter values, they are sent to the controller via serial port. After receiving the data, the microcontroller writes the parameter values ​​into the MFAC algorithm. Given a target speed, the motor is driven. During the response process, the response speed is obtained through the encoder and then sent to the host computer via serial port. After obtaining the data, the host computer obtains the three corresponding indicators of the speed through parsing and calculation, and calculates the reward value through these three indicators. This is then used for the next training, and the process is repeated until the parameter values ​​converge to the optimal value.

[0123] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0124] Furthermore, in this example embodiment, a precise speed control device for a wheeled robot is also provided. (See reference...) Figure 5 As shown, the wheeled robot speed precision control device 500 may include: a state quantity acquisition module 510, a parameter generation module 520, a control signal generation module 530, and a drive control module 540. Wherein:

[0125] The state quantity acquisition module 510 is used to acquire the state quantities of the depth deterministic policy gradient algorithm (DDPG) of the drive motor of the wheeled robot.

[0126] The parameter generation module 520 is used to generate model-free adaptive control (MFAC) parameters based on a preset DDPG model and with the DDPG state variables as input.

[0127] The control signal generation module 530 is used to generate a PWM control signal for the drive motor of the wheeled robot based on a preset MFAC model, with the MFAC parameters and the target speed of the wheeled robot as inputs.

[0128] The drive control module 540 is used to drive the drive motor of the wheeled robot based on the PWM control signal, so as to achieve precise control of the speed of the wheeled robot.

[0129] The specific details of each of the wheeled robot speed precision control device modules mentioned above have been described in detail in the corresponding wheeled robot speed precision control method, so they will not be repeated here.

[0130] It should be noted that although several modules or units of a wheeled robot speed precision control device 500 have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0131] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0132] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0133] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for accurate speed control of a wheeled robot, characterized by, The method comprises: Collecting the deep deterministic policy gradient algorithm DDPG state quantity of the driving motor of the wheeled robot; The preset DDPG model takes the DDPG state quantity as input to generate a model-free adaptive control (MFAC) parameter. , wherein so a size of a speed response overshoot, ot a speed response overshoot threshold value, str a speed response rise time, trt a speed response rise time threshold value, sts a speed response settling time, tst a speed response settling time threshold value, ssp a preset given speed size, hf a maximum speed value of a speed response curve; Based on the preset MFAC model, the MFAC parameters and the target speed of the wheeled robot are input to generate the PWM control signal of the driving motor of the wheeled robot; the preset MFAC model is wherein, is a step factor, is a weight factor, is an adjustable step sequence, is an adjustable weight coefficient, is a pseudo partial derivative, is an input difference between the previous and the current time, is a target output value of the system, y is an input value of the system at the current time, is a pre-set positive number sufficiently small, is an initial value of Based on the PWM control signal, the driving motor of the wheeled robot is driven to complete the accurate control of the speed of the wheeled robot.

2. The method of claim 1, wherein, The method further comprises: Based on the encoder of the driving motor of the wheeled robot, the speed of the driving motor is measured to complete the collection of the deep deterministic policy gradient algorithm DDPG state quantity of the driving motor of the wheeled robot, and the DDPG state quantity includes the speed response overshoot, the speed response rise time and the speed response stable time.

3. The method of claim 1, wherein, The method further comprises establishing a deep deterministic policy gradient algorithm DDPG model based on a state space module, an action space module, a reward function and a loss function: The state space module and the action space module input the DDPG state quantity to generate the model-free adaptive control MFAC parameters; The reward function is a reward function generated based on the DDPG state quantity and the DDPG state quantity threshold; The loss function is used to update the parameter values of the target network and the target policy network in the DDPG algorithm.

4. The method of claim 3, wherein, The method further comprises: Based on the preset training group data, the preset DDPG model is trained; When the parameter values output by the preset DDPG model converge, the training of the preset DDPG model is completed.

5. A wheeled robot velocity precision control device based on the method of claim 1, characterized by, The device comprises: A state quantity collection module for collecting the deep deterministic policy gradient algorithm DDPG state quantity of the driving motor of the wheeled robot; A parameter generation module for inputting the DDPG state quantity into the preset DDPG model to generate the model-free adaptive control MFAC parameters; A control signal generation module for inputting the MFAC parameters and the target speed of the wheeled robot into the preset MFAC model to generate the PWM control signal of the driving motor of the wheeled robot; A driving control module for driving the driving motor of the wheeled robot based on the PWM control signal to complete the accurate control of the speed of the wheeled robot.

Images