Motion control method, system and device for a four-wheel differential mobile robot

Abstract

The present invention relates to a motion control method, system and device for a four-wheel differential mobile robot. The motion control method comprises the following steps: S1: collecting motion data, feature data and road condition information of the mobile robot; S2: establishing a plane coordinate system with the road surface on which the mobile robot is located as a reference plane; calculating the moving speed of the mobile robot; S3: using a model prediction method to calculate the next ideal state of the mobile robot based on the motion data and feature data; S4: constructing a control model based on an Actor-Critic framework, training the control model, and then outputting the optimal strategy and optimal control quantity. The present invention outputs the control quantity through the control model, and uses a reinforcement learning method to train and optimize the control model, thereby forming a continuous optimization control of the mobile robot, which not only improves the robustness of the decision-making process, but also improves the accuracy, stability and intelligence level of the mobile robot in trajectory tracking.

Description

Technical Field

[0001] The present invention relates to a mobile robot, in particular to a motion control method for a four-wheel differential mobile robot, a motion control system for a four-wheel differential mobile robot and a motion control device for a four-wheel differential mobile robot. Background Art

[0002] As the field of four-wheel differential mobile robots continues to develop, many challenges remain in their movement in complex environments, such as unstructured ones. Traditional PID (proportional, integral, differential) control and its derivative algorithms can achieve good control performance, but still rely on manually adjusted parameters to achieve relevant control indicators. Due to the complexity and variability of unstructured environments, it is difficult to estimate the most accurate obstacle information using a single lidar or depth camera. After identifying the obstacle information, how to control the four-wheel differential mobile robot to move smoothly according to the ideal control trajectory while maintaining the moving speed of the four-wheel differential mobile robot remains an important and difficult problem to overcome.

[0003] While traditional trajectory control methods can adjust their trajectory based on real-time road condition information, they rely on a single control strategy, making them incapable of handling complex road conditions. Using traditional trajectory control methods, the actual trajectory deviates from the intended trajectory while avoiding obstacles. While a four-wheel differential mobile robot can roughly follow the pre-set trajectory, the resulting error between the actual and ideal trajectory is difficult to eliminate, resulting in a slower movement speed. Traditional control strategies are also not robust, resulting in low control accuracy and stability for the mobile robot. Summary of the Invention

[0004] Based on this, it is necessary to provide a motion control method, system and device for a four-wheel differential mobile robot to address the problems of low robustness, poor accuracy and stability in trajectory tracking of existing four-wheel differential mobile robots.

[0005] The present invention is implemented by the following technical solution: A motion control method for a four-wheel differential mobile robot comprises the following steps:

[0006] S1: Collect motion data, feature data and road condition information of the mobile robot.

[0007] S2: Establish a plane coordinate system with the road surface on which the mobile robot is located as the reference plane. Map the corresponding pose coordinates in the plane coordinate system based on the position and direction of the mobile robot at each moment. Calculate the mobile robot's movement speed based on the change in the mobile robot's pose between two adjacent moments.

[0008] S3: Based on the motion data and feature data, the next ideal state of the mobile robot is calculated using a model prediction method. The model prediction method is as follows:

[0009] S31: Establish the state space equation based on the posture of the mobile robot.

[0010] S32: Derivative the state quantity, and then convert the state space equation into a state transfer equation.

[0011] S33: Input the state quantity and road condition information into a preset neural network, which then outputs a trajectory function. Randomly select two trajectory points in the trajectory function and perform Taylor expansion on the two trajectory points to obtain the state control equation.

[0012] S34: Optimally solve the state control equation to obtain the next ideal state.

[0013] S4: Build a control model based on the Actor-Critic framework. The Actor-Critic framework includes a policy network and a value network. The control model is trained using the next ideal state and road condition information as input, and then outputs the optimal policy and optimal control variable. The control model training method is as follows:

[0014] S41: Input the next ideal state and road condition information into the strategy network, and the strategy network uses the Markov process to make a decision and generate the next control variable.

[0015] S42: Input the next control amount into the value network, the value network outputs the corresponding value, and updates the strategy network according to the value.

[0016] S43: Update the value network according to the next control quantity and the actual next state quantity.

[0017] This motion control method dynamically interacts with the mobile robot's environment and automatically generates trajectories through model prediction, making it adaptable to diverse road environments. Furthermore, the method uses a control model to output a control variable and employs reinforcement learning to train and optimize the control model. This allows the control variable output from each iteration to serve as the input for the next iteration. The control strategy and control variable are then further optimized based on the actual trajectory of the mobile robot after each iteration, resulting in continuous optimal control of the mobile robot. This not only improves the robustness of the decision-making process but also enhances the accuracy, stability, and intelligence of the mobile robot's trajectory tracking.

[0018] In one embodiment, in step S2, the moving speed is expressed as:

[0019]

[0020] Where v is the linear velocity, ω is the angular velocity, and v r is the right wheel speed, v l is the left wheel speed, and l is the distance between the left and right driving wheels.

[0021] In one embodiment, the state space equation is expressed as:

[0022]

[0023] In the formula, X is the state quantity, x is the horizontal coordinate, y is the vertical coordinate, is the deflection angle, u is the control variable, τ f is the heading angle.

[0024] In one embodiment, the state transition equation is expressed as:

[0025]

[0026] Where, is the lateral velocity, is the longitudinal velocity, is the angular velocity.

[0027] In one embodiment, the state control equation is expressed as:

[0028] ξ k+1 =Aξ k +B△U k

[0029] η k =Cξk

[0030] Where, ξ k+1 is the next ideal state, ξ k is the current state, ΔU k is the variable of the control quantity, A is the state matrix, B is the control matrix, η k is the control quantity, and C is the output matrix.

[0031] In one embodiment, the next ideal state ξ k+1 Expressed as:

[0032]

[0033] Where, is the state quantity at the next moment, is the control quantity at the next moment.

[0034] In one embodiment, strategy a t and control quantity s t+1 Expressed as:

[0035] a t~π(a t-1 |s t ,x t ,y t ,Ψ t )

[0036] s t+1 ~p(s t+1 |s t ,a t ,x t ,y t ,Ψ t )

[0037] Where π is the value network and p is the policy network.

[0038] In one embodiment, the parameter update method of the policy network is as follows:

[0039]

[0040] The parameter update method of the value network is as follows:

[0041]

[0042] The present invention also provides a motion control system for a four-wheel differential mobile robot. The motion control system includes an acquisition module, a state generation module, a state prediction model, a state control model and a mobile control module.

[0043] The acquisition module is used to collect the motion data, feature data and road condition information of the mobile robot.

[0044] The state generation module is used to establish a plane coordinate system based on motion data and feature data, thereby obtaining the position and posture of the mobile robot, and calculating the current moving speed of the mobile robot based on the position and posture at two adjacent moments.

[0045] The state prediction model is used to output the next ideal state based on the posture, moving speed and road condition information.

[0046] The state control model consists of a policy network and a value network. The policy network calculates the next control variable based on the next ideal state and road condition information. The value network is used to: 1. Calculate a value based on the next control variable and update the policy network accordingly; 2. Self-update based on the next control variable and the next state variable.

[0047] The mobile control module is used to adjust the moving speed of the mobile robot according to the next control amount.

[0048] The present invention also provides a motion control device for a four-wheel differential mobile robot. The motion control device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The functional modules in the motion control device are deployed in the manner of the aforementioned motion control system for the four-wheel differential mobile robot. When the processor executes the computer program, it implements the steps of the aforementioned motion control method for the four-wheel differential mobile robot, thereby achieving optimized control of the mobile robot and improving the accuracy and smoothness of the mobile robot's trajectory tracking.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention can dynamically interact with the mobile robot's environment and automatically generate trajectories through model prediction, making it applicable to various road environments. The aforementioned motion control method also uses a control model to output a control variable and employs reinforcement learning to train and optimize the control model, so that the control variable outputted each time serves as the input for the next time. The control strategy and control variable are further optimized based on the actual trajectory of the mobile robot after each control, forming a continuous optimization control system for the mobile robot. This not only improves the robustness of the decision-making process, but also enhances the accuracy, stability, and intelligence of the mobile robot's trajectory tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a step diagram of the motion control method of the four-wheel differential mobile robot according to embodiment 1 of the present invention;

[0052] Figure 2 for Figure 1 Flowchart of the motion control method of the four-wheel differential mobile robot;

[0053] Figure 3 for Figure 1 Simulation diagram of the moving speed of the four-wheel differential mobile robot;

[0054] Figure 4 for Figure 1 Schematic diagram of the posture modeling of the four-wheel differential mobile robot;

[0055] Figure 5 for Figure 1 Schematic diagram of the local network and the overall target network based on the AC framework;

[0056] Figure 6 Schematic diagram of trajectory tracking of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1;

[0057] Figure 7Schematic diagram of single-axis trajectory tracking of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1;

[0058] Figure 8 Schematic diagram of the trajectory tracking control strategy of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1;

[0059] Figure 9 Schematic diagram of the trajectory tracking error of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1;

[0060] Figure 10 Based on Figure 1 Schematic diagram of the motion control system of the four-wheel differential mobile robot motion control method. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component. When a component is considered to be "fixed to" another component, it may be directly fixed to the other component or there may be a central component.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0064] Example 1

[0065] See also Figure 1 and Figure 2 , Figure 1 This is a step diagram of the motion control method of the four-wheel differential mobile robot according to embodiment 1 of the present invention; Figure 2 for Figure 1 Flowchart of the motion control method of the four-wheel differential mobile robot. The motion control method of the four-wheel differential mobile robot includes:

[0066] S1: Collect the motion data, feature data and road condition information of the mobile robot. Among them, the feature data includes the wheelbase, front and rear wheel spacing of the mobile robot, etc., which can be obtained by direct measurement. Both motion data and road condition information can be obtained by installing corresponding sensors. For example, road condition information can be obtained by installing multiple cameras, radar detectors, etc. on the mobile robot. Specifically, at least one camera is installed on each side of the mobile robot, and at least one radar detector is installed on the front side of the mobile robot, and then the current road condition is analyzed by the radar detection distance and the video captured by the camera, and then the road trajectory and obstacle information are obtained. The motion data includes the position information, running direction, etc. of the mobile robot. Among them, the position information can be obtained by installing a GPS locator, and the running direction can be obtained by installing a direction recognition detector.

[0067] S2: Establish a plane coordinate system with the road surface where the mobile robot is located as the reference plane. According to the position and direction of the mobile robot at each moment, the corresponding posture coordinates are mapped in the plane coordinate system.

[0068] Specifically, we randomly select a point on the moving road as the origin, and randomly select two mutually perpendicular directions on the road as the X-axis and Y-axis to form a plane coordinate system based on the road. The position of the robot relative to the origin is also the position coordinate, which is recorded as (x i ,y i ).

[0069] The moving speed of the mobile robot is calculated based on the change of the position of the mobile robot at two adjacent moments. Assume that the speeds of the front and rear wheels on the left are V1 and V2 respectively, and the speeds of the front and rear wheels on the right are V3 and V4 respectively. Then the speed of the left wheel V l It can be expressed as:

[0070]

[0071] Right wheel speed V r It can be expressed as:

[0072]

[0073] Please combine Figure 3 , which is Figure 1 A simulation diagram of the moving speed of a four-wheel differential mobile robot. The distance between the left and right driving wheels of the mobile robot (wheelbase) is l, and the angle the mobile robot rotates at two adjacent moments is θ. During this process, the distance the right wheel travels more than the left wheel is d, and the radius of the circular movement curve formed by the mobile robot during this process is r. Let the linear velocity of the left wheel be v l , the linear velocity of the right wheel is vr , then the overall linear velocity v of the mobile robot can be expressed as:

[0074]

[0075] Since the sampling time between two adjacent moments is very small, it can be concluded from the geometric relationship:

[0076] θ≈tann(θ)

[0077] Expanding the above formula yields:

[0078]

[0079] According to the speed formula:

[0080]

[0081] Then the moving speed V of the mobile robot can be expressed as:

[0082]

[0083] By converting the wheel speed of the mobile robot into the linear velocity and angular velocity of the reference point of the entire mobile robot, the trajectory control of the robot can be achieved.

[0084] S3: Based on the motion data and feature data, the next ideal state of the mobile robot is calculated using a model prediction method. The model prediction method is as follows:

[0085] S31: Establish the state space equation based on the posture of the mobile robot.

[0086] Please combine Figure 4 , which is Figure 1 Schematic diagram of the pose modeling of the four-wheel differential mobile robot. The position of the mobile robot in the plane coordinate system is recorded as (x, y), and the deflection angle of the mobile robot relative to the X axis is recorded as Then the state space equation of the mobile robot can be expressed as:

[0087]

[0088] In the formula, X is the state quantity, which represents the three trajectory control targets of the four-wheel differential mobile robot, x is the horizontal coordinate, y is the vertical coordinate, is the deflection angle, u is the control variable, representing the two control quantities applied, τ f is the heading angle.

[0089] S32: Derivative the state quantity, and then transform the state space equation into the state transfer equation. The state transfer equation can be expressed as:

[0090]

[0091] Where, is the lateral velocity, is the longitudinal velocity, is the angular velocity.

[0092] S33: Input the state quantity and road condition information into a preset neural network, and then output a trajectory function. The neural network includes multiple activation functions to estimate the optimal next state quantity. The next state quantity is expressed by the trajectory function and is recorded as:

[0093]

[0094] Where, f(x t , μ t ) is the trajectory function at time t, x t 、μ t are the state quantity and control quantity at time t respectively.

[0095] Pick any two points in the trajectory function: and Among them, x r 、x k are the state quantities at any time r and k, μ r 、μ k are the control quantities at any time r and k respectively.

[0096] Place the two selected points at x k Taylor expansion is performed on , that is:

[0097]

[0098] Where x k-1 is the state quantity at time k-1.

[0099] After sorting, we can get:

[0100]

[0101] Let the estimated value of the state Control quantity estimate Among them, U k is the control quantity at time k, U r is the control quantity at time r. Introduce sampling time T, and in, is the estimated value of the previous control quantity.

[0102] We can get:

[0103]

[0104] Where, is the current state quantity, is the control quantity, A1 is the first state matrix, and B1 is the first control matrix.

[0105] In the process of making the optimal decision, we need to know all the prediction quantities and state quantities in the prediction time domain, and then find the optimal value of the cost matrix, so we need to construct a new state quantity The final state control equation is expressed as:

[0106] ξ k+1 =Aξ k +BΔU k

[0107] η k =Cξk

[0108] Where, ξ k+1 is the next ideal state, ξ k is the current state, ΔU k is the variable of the control amount, that is, the increase of each control amount compared with the previous control amount, A is the state matrix, B is the control matrix, η k is the control quantity, and C is the output matrix.

[0109] S34: Optimally solve the state control equation to obtain the next ideal state.

[0110] Define Y, Ψ, Θ as the intermediate transition matrix, and:

[0111]

[0112] Define ΔU(t)=[ΔU k (t), ΔU k+1 (t), ...ΔU k+Nc-1 (t),…ΔU k+Np-1 (t)] T

[0113] Then T=Ψ·ξ(k)+Θ·ΔU (t)

[0114] Set the optimization objective function to:

[0115]

[0116] Q=I(N p *N x ) R=I(N c *N u )

[0117] Where, X (k-1) is the state quantity of the previous moment, ρ is the weight coefficient, ε is the relaxation factor, I is the unit matrix, Np is the prediction time domain, N c To control the time domain, N x is the number of state variables, N u is the number of control quantities.

[0118] Q is the state weighting matrix, which is used to enable the state quantity to quickly reach the reference optimal value. R is the decision weighting matrix, which is used to be as stable as possible in the process of controlling the state quantity and ensure the continuity of decision implementation.

[0119] Let error = Ψ·ξ(k),

[0120]

[0121] The above problem can be modeled as a constrained quadratic programming problem:

[0122]

[0123] Where J is the optimization function, H is the quadratic term matrix, f1 is the linear term matrix, and a1, b1, l1, and u1 are all constraint matrices.

[0124] make

[0125] but

[0126]

[0127] Optimize and solve the state quantity The state at the next moment can be solved.

[0128] S4: Construct a control model based on the Actor-Critic framework. The Actor-Critic framework includes a policy network and a value network. The next ideal state and road condition information are used as input to train the control model, and then the optimal policy a is output. t .

[0129] Please combine Figure 5 , which is Figure 1 Schematic diagram of the local network and overall target network based on the AC framework. First, the control model based on the Actor-Critic framework is initialized. The initialization content includes:

[0130] Initialize the local network:

[0131] Initialize the overall target network:

[0132] Initialize dynamic obstacle parameters:

[0133] Initialize network parameters:

[0134] Initialize the replay experience pool data:

[0135] Initial hyperparameters: step size α, β, update gradient λ.

[0136] Subsequently, the control model is trained and the corresponding next state quantity and next control quantity are output. The training method of the control model is as follows:

[0137] S41: Input the next ideal state and road condition information into the strategy network, which uses the Markov process to make decisions and generate the next control variable. Each decision updates the state and action, and outputs the next state and the next control variable:

[0138] a t ~π(a t-1 |s t ,x t ,y t ,Ψ t )

[0139] s t+1 ~p(s t+1 |s t ,a t ,x t ,y t ,Ψ t )

[0140] S42: Input the next control quantity into the value network, the value network outputs the corresponding value, and updates the policy network according to the value. The parameter update of the policy network is expressed as:

[0141]

[0142] S43: Update the value network according to the next control quantity and the actual next state quantity. The parameter update of the value network is expressed as:

[0143]

[0144] In addition, in order to achieve cyclic training of the policy network and the value network, the control model should also be updated as follows:

[0145] Update replay experience pool data:

[0146] Perform gradient update loop: Estimate advantage function: A π (s t ,a t )=Q(s t ,a t)-V(s t )

[0147] Update the overall target network: θ←θ-λJ(θ)

[0148] Experience replay training: φ←φ-λ π J π (φ)

[0149] By training the control model and optimizing the control of the mobile robot according to the next control quantity output by the strategy network, the mobile robot can gradually approach the predicted operation trajectory during operation. At the same time, the value output by the corresponding value network can better reflect whether the next control quantity currently output is close to the ideal optimal control quantity, thereby realizing continuous optimization control of the mobile robot.

[0150] While traditional trajectory control methods require manual pre-setting of the trajectory, the motion control method of this embodiment dynamically interacts with the mobile robot's environment and automatically generates a trajectory using a model prediction method, making it adaptable to diverse road environments. This motion control method also uses a control model to output a control variable and employs reinforcement learning to train and optimize the control model. This allows the control variable output each time to serve as the input for the next control step. The control strategy and control variable are then further optimized based on the actual trajectory of the mobile robot after each control step, resulting in continuous optimal control of the mobile robot. This not only improves the robustness of the decision-making process, but also enhances the accuracy, stability, and intelligence of the mobile robot's trajectory tracking.

[0151] Experimental verification

[0152] First, a model prediction method is used to generate a reference trajectory based on the actual environment the mobile robot is in. The position, speed, and heading angle in the reference trajectory are recorded separately, and the corresponding trajectory curve is generated from them.

[0153] Subsequently, the mobile robot is controlled using the above control method, the actual movement trajectory of the mobile robot is recorded, and the corresponding position, speed, and heading angle are used as references to generate the corresponding actual trajectory curve.

[0154] See also Figure 6 and Figure 7 , Figure 6 Schematic diagram of trajectory tracking of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1; Figure 7 Schematic diagram of single-axis trajectory tracking of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1. Figure 6It can be seen that in the initial stage of trajectory tracking, the error between the actual trajectory and the reference trajectory is large. However, with the continuous training of the control model, the control strategy is optimized, and the actual trajectory of the mobile robot gradually approaches the reference trajectory.

[0155] Figure 7 Based on the sampling time Figure 6 Further decomposition of the trajectory curve. Figure 7 The graph above shows the trajectory of the mobile robot on the X-axis. Between 0 and 2 seconds, there is a significant deviation between the actual X-axis position of the mobile robot and the reference position. However, as the sampling time progresses, the actual X-axis position of the mobile robot remains almost consistent with the reference position. Figure 7 The chart below shows the mobile robot's trajectory on the Y axis. From 0 to 5 seconds, the mobile robot's actual position on the Y axis gradually approaches the reference position. After 6 seconds, the mobile robot's actual position is almost identical to the reference position.

[0156] Please combine Figure 8 , which is a schematic diagram of the trajectory tracking control strategy of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1. Figure 8 The upper middle chart shows the change of moving speed with sampling time. Figure 8 The lower middle graph shows the change in heading angle over sampling time. It can be seen that in the initial stage, the mobile robot's speed and heading angle remain stable, then gradually change, and then stabilize again in the final stage. This shows that the motion control method of this embodiment has the effect of gradually optimizing control. This is because the strategy in the initial stage is randomly selected, and the control of the mobile robot's state is not optimal. In subsequent control, through parameter updates in the policy network and value network, the output strategy and control variables are gradually optimized, bringing the mobile robot's trajectory closer to the ideal trajectory.

[0157] Please combine Figure 9 , which is a schematic diagram of the trajectory tracking error of the four-wheel differential mobile robot based on the Actor-Critic framework in Example 1. Figure 9 The upper graph is the longitudinal error graph. Figure 9 The middle graph is the lateral error graph. It can be seen that the mobile robot's lateral and longitudinal errors gradually decrease in the initial stage, indicating that the trajectory is gradually approaching the ideal trajectory. In the subsequent period, the lateral error approaches zero, while the longitudinal error, while relatively close to zero, fluctuates more. The longitudinal error is primarily determined by the mobile robot's speed control, while the lateral error is more dependent on obstacle avoidance. The motion control method of this embodiment is more advantageous for obstacle avoidance.

[0158] Figure 9 The lower graph is the heading angle error graph. It can be seen that in the initial stage, the heading angle error gradually decreases and approaches zero in the middle and late stages, reaching a relatively stable state, indicating that the stability of the mobile robot gradually improves during movement.

[0159] See also Figure 10 , which is based on Figure 1 A schematic diagram of the motion control system structure of a four-wheel differential mobile robot motion control method is provided. Based on the above motion control method, this embodiment also provides a motion control system for a four-wheel differential mobile robot to facilitate precise control of the mobile robot. See FIG. The motion control system includes an acquisition module, a state generation module, a state prediction model, a state control model, and a mobile control module.

[0160] The acquisition module is used to collect the mobile robot's motion data, feature data, and road condition information. This acquisition module can be a variety of sensors installed on the mobile robot, or a measurement device separate from the mobile robot. Motion data includes the robot's direction of travel, linear velocity, and angular velocity. Feature data includes the distance between the robot's left and right wheels, and the distance between its front and rear wheels. Road condition information includes road trajectory and obstacle information.

[0161] The state generation module is used to establish a plane coordinate system based on motion data and feature data, thereby obtaining the mobile robot's position and pose. It also calculates the mobile robot's current velocity based on the position and pose at two adjacent moments. The mobile robot's position and pose include its coordinates relative to the origin and its angle relative to the X-axis. The velocity includes both linear velocity and angular velocity.

[0162] The state prediction model outputs the next ideal state based on the vehicle's position, speed, and road condition information. This model includes activation functions in multiple hidden layers. These activation functions convert the input position, speed, and road condition information into corresponding signals. These signals are then filtered and classified to generate a trajectory based on the current road conditions. The next point in the trajectory is then output as the next ideal state.

[0163] The state control model includes a policy network and a value network. The policy network is used to calculate the next control variable based on the next ideal state and road condition information. In this embodiment, the policy network can use the state prediction model as the initial network model.

[0164] The value network is used to: 1. Calculate a value based on the next control variable and update the policy network accordingly. After the next control variable output by the policy network is fed into the value network, the value network analyzes the next state variable based on the next control variable and road condition information. This value is then compared with the next state variable of the ideal trajectory to generate an evaluation of the current next control variable, or value. The policy network is then updated based on the magnitude of this value. In other words, the policy network always outputs a higher value than the value network outputs when outputting the next control variable.

[0165] Second, self-update based on the next control variable and the next state variable. Due to changes in road conditions and in actual trajectory control, the impact of the next control variable on the actual trajectory gradually changes. Therefore, based on the relationship between the next control variable and the next state variable, the value network is also updated accordingly, so that the output value better represents the preference of the next control variable.

[0166] The mobile control module is used to adjust the mobile robot's speed based on the next control variable, which includes the next angular velocity and next linear velocity. It also calculates the control variable based on the current angular velocity and linear velocity, and then converts it into the mobile robot's acceleration, rotation angle, and other parameters, achieving precise control of the mobile robot.

[0167] To implement the aforementioned motion control method for a four-wheel differential mobile robot and apply it to existing mobile robots, thereby simplifying the trajectory control process in practical applications and improving the control accuracy and stability of the mobile robot, this embodiment also provides a motion control device for a four-wheel differential mobile robot. The motion control device includes a memory, a processor, and a computer program stored in the memory and executable by the processor. When the processor executes the computer program, it implements the steps of the aforementioned motion control method for a four-wheel differential mobile robot, thereby achieving optimized control of the existing mobile robot and improving the accuracy and stability of the mobile robot's trajectory tracking.

[0168] The computer device may be a smartphone, tablet computer, laptop computer, desktop computer, rack server, blade server, tower server, or cabinet server (including a standalone server or a server cluster consisting of multiple servers) that can execute programs. The computer device of this embodiment includes at least, but is not limited to, a memory and a processor that can be interconnected via a system bus.

[0169] In this embodiment, the memory (i.e., readable storage medium) includes flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of a computer device, such as the hard disk or internal memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk equipped with the computer device, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Of course, the memory may also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory is generally used to store the operating system and various application software installed on the computer device. In addition, the memory may also be used to temporarily store various types of data that have been output or are about to be output.

[0170] In some embodiments, the processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor is typically used to control the overall operation of a computer device. In this embodiment, the processor is used to execute program code stored in a memory or process data, thereby implementing the steps of the aforementioned motion control method for a four-wheel differential mobile robot, thereby achieving optimized control of an existing mobile robot and improving the accuracy and stability of the mobile robot's trajectory tracking.

[0171] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A motion control method for a four-wheel differential mobile robot, which is used to generate a predicted path based on the environment in which the mobile robot is located, and then generate a corresponding control strategy based on the predicted path and the current motion state of the mobile robot; characterized in that: The motion control method comprises the following steps: S1: Collecting motion data, feature data and road condition information of the mobile robot; S2: establishing a plane coordinate system with the road surface on which the mobile robot is located as a reference plane; mapping corresponding posture coordinates in the plane coordinate system according to the position and travel direction of the mobile robot at each moment, and calculating the moving speed of the mobile robot according to the posture change of the mobile robot at two adjacent moments; S3: Calculate the next ideal state of the mobile robot using a model prediction method based on the motion data and the feature data; the model prediction method is specifically as follows: S31: Establishing a state space equation according to the posture of the mobile robot; S32: Derivative the state quantity, thereby converting the state space equation into a state transfer equation; S33: Inputting the state quantity and the road condition information into a preset neural network, thereby outputting a trajectory function; arbitrarily selecting two trajectory points in the trajectory function, and performing Taylor expansion on the two trajectory points to obtain a state control equation; S34: Optimally solving the state control equation to obtain the next ideal state; S4: Constructing a control model based on an Actor-Critic framework; the Actor-Critic framework includes a policy network and a value network; using the next ideal state and the road condition information as input to train the control model, thereby outputting an optimal policy and an optimal control variable; the training method of the control model is as follows: S41: Inputting the next ideal state and the road condition information into the strategy network, and the strategy network adopts a Markov process to make a decision and generate the next control variable; S42: Inputting the next control amount into the value network, the value network outputting a corresponding value, and updating the parameters of the strategy network according to the value; S43: Using the next control quantity to control the trajectory of the mobile robot, and then updating the parameters of the value network according to the actual next state quantity fed back by the mobile robot, thereby forming continuous optimization control of the mobile robot.

2. The motion control method of the four-wheel differential mobile robot according to claim 1, characterized in that: In step S2, the movement speed includes linear speed and angular speed. The linear speed is the average of the left wheel speed and the right wheel speed. The angular speed is obtained according to the ratio of the angle turned by the mobile robot at two adjacent moments to the time. The movement speed is expressed as: Where v is the linear velocity, ω is the angular velocity, and v r is the right wheel speed, v l is the left wheel speed, and l is the distance between the left and right driving wheels.

3. The motion control method of the four-wheel differential mobile robot according to claim 1, characterized in that: In step S31, the state space equation includes a position equation and a control equation; the position equation is used to characterize the position coordinates and current deflection angle of the mobile robot; the control equation is used to characterize the moving speed and rotation angle of the mobile robot; the state space equation is expressed as: In the formula, X is the state quantity, x is the horizontal coordinate, y is the vertical coordinate, is the deflection angle, u is the control variable, τ f is the heading angle, and v is the linear velocity.

4. The motion control method of the four-wheel differential mobile robot according to claim 3, characterized in that: In step S32, the state transition equation is expressed as: Where, is the lateral velocity, is the longitudinal velocity, is the angular velocity.

5. The motion control method of the four-wheel differential mobile robot according to claim 1, characterized in that: In step S33, the state control equation includes an ideal state equation and an ideal control equation; the ideal state equation is used to characterize the ideal state of the mobile robot at the next moment based on the current state and control quantity; the ideal control equation is used to characterize the ideal control quantity at the next moment when the mobile robot moves on the ideal motion trajectory.

6. The motion control method of the four-wheel differential mobile robot according to claim 5, characterized in that: In step S34, the next ideal state includes the ideal state quantity and ideal control quantity of the mobile robot at the next moment; the next ideal state ξ k+1 Expressed as: Where, is the state quantity at the next moment, is the control quantity at the next moment.

7. The motion control method of the four-wheel differential mobile robot according to claim 6, characterized in that: In step S41, the control model interacts with the dynamic road condition information and estimates the best next ideal state through a multi-layer neural network. Each interaction updates the state and action.

8. The motion control method of the four-wheel differential mobile robot according to claim 1, characterized in that: In step S42 and step S43, each time the state and control amount at the next moment are output, the parameters of the strategy network and the value network are updated.

9. A motion control system for a four-wheel differential mobile robot, which adopts the motion control method for a four-wheel differential mobile robot according to any one of claims 1 to 8, characterized in that: The motion control system comprises: A collection module, which is used to collect motion data, feature data and road condition information of the mobile robot; a state generation module, configured to establish a plane coordinate system based on the motion data and the feature data, thereby obtaining the position and posture of the mobile robot, and calculating the current moving speed of the mobile robot based on the position and posture at two adjacent moments; A state prediction model, configured to output a next ideal state based on the posture, the moving speed, and the road condition information; A state control model comprising a policy network and a value network; the policy network is configured to calculate a next control variable based on the next ideal state and the road condition information; the value network is configured to: first, calculate a value based on the next control variable and update the policy network based on the value; and second, perform self-update based on the next control variable and the next state variable. A movement control module is used to adjust the movement speed of the mobile robot according to the next control amount.

10. A motion control device for a four-wheel differential mobile robot, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The functional modules in the motion control device are deployed in the form of a motion control system for a four-wheel differential mobile robot as described in claim 9. When the processor executes the computer program, the steps of the motion control method for a four-wheel differential mobile robot as described in any one of claims 1 to 8 are implemented to achieve optimized control of the mobile robot and improve the accuracy and smoothness of the trajectory tracking of the mobile robot.

Images