Self-adaptive path tracking control method and system for robot in complex environment

Abstract

According to the robot adaptive path tracking control method and system in the complex environment, a robot adopts double differential driving wheels and a double universal wheel chassis, and the method comprises the following steps: 1, training the bandwidth of a controller by using an LADRC controller parameter optimization algorithm based on gradient descent to obtain an optimal bandwidth, and establishing an optimal controller by using the optimal bandwidth; 2, defining boundary conditions of robot path planning, including starting point and end point position coordinates and corresponding speed and / or acceleration parameters; 3, current posture estimation data, current target posture data and current surrounding environment information of the robot are continuously collected, data screening is carried out, and sorting is carried out according to a motion time sequence to serve as a known posture sequence; and 4, the cubic spline interpolation function generates a continuous smooth path curve by using boundary conditions and known attitude sequence interpolation, the next target attitude is extracted from the continuous smooth path curve and is sent to the optimal controller, the optimal controller controls the robot to track and reach the next target attitude, and the step 3 is repeated until the robot reaches an end point.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot technology, in particular to a robot adaptive path tracking control method and system in complex environment, which is suitable for robot chassis design and control field. BACKGROUND

[0002] With the wide application of robot technology in industrial production, logistics transportation, service and many other fields, higher requirements are put forward for the control accuracy, adaptability and stability of robot chassis. In the design of mobile chassis of robots, the traditional chassis structure and control method often have limitations in complex environments. For example, the chassis with single drive mode may not be sufficient in terms of maneuverability and adaptability to complex terrain. The emergence of the double-differential drive wheel and double universal wheel chassis structure in this patent is to better cope with these challenges. This chassis design can flexibly turn and move in complex terrain, but to fully utilize its advantages, accurate mileage calculation methods and matching control strategies are needed. At the same time, during the motion control of the robot, external disturbances and internal uncertainty factors have a significant impact on the control effect. Changes in motor speed, fluctuations in load, and various unknown environmental disturbances can all cause the robot to fail to accurately follow the predetermined trajectory. Traditional control algorithms may have problems such as decreased control accuracy and deteriorated stability when dealing with these complex disturbances. Therefore, an advanced control algorithm that can effectively suppress disturbances is needed, which is the reason for the introduction of the active disturbance rejection control algorithm. In addition, path planning for robots in complex environments is crucial. In scenarios with obstacles, multiple target points, and complex terrain, how to plan a path for the robot that can accurately reach the target position while ensuring smooth and efficient movement during the process becomes a key issue. The cubic spline interpolation planning algorithm shows unique advantages in this case, as it can generate a smooth path based on target position and surrounding environment information, overcoming the problems of path discontinuity and sudden turning that may be caused by traditional path planning methods, thereby improving the stability and accuracy of robot movement and further enhancing the robot's ability to perform tasks in complex environments. Based on these practical needs, this patent proposes a robot adaptive path tracking control method and system in complex environment.

[0003] When optimizing the parameters of the linear active disturbance rejection controller (LADRC controller), neural networks are generally used to optimize the parameters. Using neural networks to optimize parameters is a common technical means, as described in the invention patent with application number CN2017102060763 and the name of a design method for an active disturbance rejection controller based on RBF neural network, and the invention patent with application number CN2022109553137 and the name of a self-disturbance controller parameter optimization method based on deep reinforcement learning. SUMMARY

[0004] The present application provides a robot adaptive path tracking control method and system in complex environment, which can be applied to a high-stability chassis suitable for various complex terrains, uses a double-differential drive wheel and double universal wheel bottom structure to improve the load strength of the robot and the stability when running in complex environments, uses a self-disturbance control algorithm to further improve the stability of motor control, so that the robot can cope with various external disturbances, and uses a cubic spline interpolation planning to improve the smoothness of the planned path and improve the sensitivity and safety of the robot.

[0005] The present application solves the above technical problems by the following technical solutions:

[0006] The present application provides a robot adaptive path tracking control method in complex environment, characterized in that the robot uses a double-differential drive wheel and double universal wheel chassis, two universal wheels are added to the front side of the robot, and the control method comprises:

[0007] Step one: use the LADRC controller parameter optimization algorithm based on gradient descent to train the bandwidth of the LADRC controller used by the robot, obtain the optimal bandwidth, and establish the optimal LADRC controller using the optimal bandwidth;

[0008] Step two: define the boundary conditions of the robot path planning, including the position coordinates of the starting point and the ending point, and the corresponding speed and / or acceleration parameters;

[0009] Step three: continuously collect the current attitude estimation data, current target attitude data, and current surrounding environment information of the robot by laser radar and camera, and perform data screening to eliminate obvious errors or redundant information, sort them according to the motion sequence as the known attitude sequence, and the current surrounding environment information includes whether there are obstacles and the collected obstacle position data when there are obstacles;

[0010] Step four: construct a cubic spline interpolation function, which uses the boundary conditions and the known attitude sequence for interpolation processing to generate a continuous and smooth path curve, extracts the next target attitude including the next position coordinate and motor speed from the path curve, and inputs the next target attitude into the optimal LADRC controller, so that the robot can track and reach the next target attitude, and repeat step three until the robot reaches the end point.

[0011] In step one, the LADRC controller parameter optimization algorithm based on gradient descent:

[0012] S1, initialize the bandwidth of the controller Set the learning rate η, the maximum number of iterations max, the convergence threshold th, the performance evaluation weight coefficients w1, w2, and w3;

[0013] S2. Under the control of the robot using the latest LADRC controller, the robot's operating data is collected in real time, including the robot's linear velocity v(t), current position, and motor power P(t) during the current acquisition cycle. The current velocity change Δv(t) = v(t) - v(t-1) and the position error e(t) = target position - current position are calculated.

[0014] S3. Construct a performance evaluation function for the automatic parameter adjustment mechanism and use the performance evaluation function to quantify the bandwidth ω of the current controller. c Control performance under:

[0015]

[0016] Where T represents the number of collection cycles from the start to the current collection cycle, and t represents the tth collection cycle;

[0017] S4, the current bandwidth ω c Add a small perturbation Δω c , the bandwidth after generating the perturbation

[0018] S5, based on the bandwidth after perturbation calculate

[0019] S6, use the central difference method to estimate the performance evaluation function for the current bandwidth ω c Gradient

[0020]

[0021] S7. Update the bandwidth of the LADRC controller using gradient descent

[0022] Repeat the above iterative process S2-S7 until the number of iterations reaches the maximum number of iterations max or the performance evaluation index function J(ω c ) is less than the convergence threshold th, and the optimal bandwidth ω is obtained c =ω′ c Applied to LADRC controller to establish the optimal LADRC controller.

[0023] The present invention also provides an adaptive path tracking control system for a robot in a complex environment. The robot adopts a dual differential drive wheel and a dual universal wheel chassis, with two universal wheels additionally provided on the front side of the robot. The control system includes:

[0024] The training optimization module is used for training the bandwidth of the LADRC controller used by the robot by using a gradient descent-based LADRC controller parameter optimization algorithm, obtaining an optimal bandwidth, and establishing an optimal LADRC controller by using the optimal bandwidth.

[0025] The boundary setting module is used for explicitly defining the boundary conditions of the robot path planning, including the position coordinates of the starting point and the ending point, and corresponding speed and / or acceleration parameters.

[0026] The data acquisition module is used for continuously acquiring the current pose estimation data, the current target pose data and the current surrounding environment information of the robot by using the laser radar and the camera, performing data screening to eliminate obvious errors or redundant information, and sorting according to the motion time sequence as a known pose sequence.

[0027] The path tracking module is used for constructing a cubic spline interpolation function, performing interpolation processing on the boundary conditions and the known pose sequence by using the cubic spline interpolation function to generate a continuous and smooth path curve, extracting a next target pose including a next position coordinate and a motor speed from the path curve, and feeding the next target pose to the optimal LADRC controller.

[0028] The positive progress effect of the application is that:

[0029] In the application, the double-differential driving wheel and the double universal wheel design make the chassis have good maneuverability and stability, the wheel mileage is calculated through chassis kinematics and dynamics modeling, data for robot positioning and navigation are provided, complex terrains can be adapted to, terrain errors can be reduced, and the mileage calculation precision is guaranteed.

[0030] In the application, the cubic spline interpolation path planning generates a smooth path between discrete points (such as targets, obstacle avoidance points, etc.) for the differential chassis in complex environments, and is crucial for stable movement and precise control of the chassis.

[0031] In the present application, the bandwidth of the LADRC controller used by the robot is trained by using the LADRC controller parameter optimization algorithm of gradient descent, the optimal bandwidth is obtained, and the optimal LADRC controller is established by using the optimal bandwidth. Compared with the existing controller parameter optimization technology using neural network, the performance evaluation function of the parameter automatic adjustment mechanism is designed and the LADRC controller is optimized by using gradient descent. The algorithm has short training period and fast convergence speed, can quickly complete model iteration and deployment, has high optimization precision, and the performance evaluation function is deeply integrated with speed and power perception, so that the control efficiency is greatly improved, the running stability is significantly enhanced, and the energy consumption is effectively optimized. The optimized linear active disturbance rejection controller is used to track the robot path, and the control precision and stability are increased. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 The flowchart of the robot adaptive path tracking control method of the preferred embodiment of the present application.

[0033] Figure 2 The design diagram of the double-differential drive wheel and double universal wheel chassis of the preferred embodiment of the present application.

[0034] Figure 3 The flowchart of the linear active disturbance rejection control of the preferred embodiment of the present application.

[0035] Figure 4 The control effect diagram of the linear active disturbance rejection control of the preferred embodiment of the present application.

[0036] Figure 5 The structure block diagram of the robot adaptive path tracking control system of the preferred embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0038] As shown in Figures 1-5 , the embodiments of the present application provide a robot adaptive path tracking control method for complex environment, so that the mobile robot can flexibly, stably and safely run on various complex terrains.

[0039] The robot adopts a double-differential drive wheel and double-universal wheel chassis, and two universal wheels are additionally arranged on the front side of the robot. Figure 2As shown, compared with the traditional differential chassis, two universal wheels are added to the front side of the robot to improve the overall stability of the robot, and this does not introduce more complex analysis variables for the robot system.

[0040] Firstly, the chassis is built based on the mechanical design shown in Figure 2 , and then kinematics is solved, which provides a theoretical basis for subsequent motion trajectory calculation. Both of them serve the dual-differential drive wheel and dual-universal wheel chassis design, so as to accurately grasp the motion state of the chassis and facilitate the calculation of the robot motion posture. At the same time, the tracking differentiator in the LADRC controller (linear active disturbance rejection controller) optimizes the command, combined with the current state of the robot observed by the extended state observer, and after accurately estimating the system disturbance by the extended state observer, the linear state error feedback control law generates accurate control signals according to these information, ensuring the stability of the chassis motion (see Figure 3 ). The data points obtained in the cubic spline interpolation planning are the basis, and the boundary conditions are set according to these data, and then the cubic spline interpolation function is constructed, and finally the path generation and optimization are realized. This planning cooperates with the chassis design and active disturbance rejection control algorithm, so that the chassis can run smoothly according to the optimized path in complex environment, and realize the overall efficient and accurate control (see Figure 4 ).

[0041] Let the left wheel speed in the dual-differential drive wheel be v l , the right wheel speed be v R , the distance between the two wheels be d, the linear speed of the robot in the plane Cartesian coordinate system be v = 0.5(v l +v R ), and the angular speed be , which are important variables to describe its motion state.

[0042] Based on the above speed relationship, the position (x, y, θ) of the robot in the plane Cartesian coordinate system is:

[0043] x = v cos θ

[0044] y = v sin θ

[0045] θ = ω

[0046] The heading angle θ is the angle between the robot and the x-axis in the plane Cartesian coordinate system. The x-coordinate of the robot is updated according to the x-direction component of the linear speed v, the y-coordinate is updated according to the y-component of the linear speed in the y-direction, and the heading angle θ is updated according to the angular speed ω.

[0047] As shown in Figure 1 , the control method comprises:

[0048] Step one: train the bandwidth of the LADRC controller used by the robot using the LADRC controller parameter optimization algorithm based on gradient descent, obtain the optimal bandwidth, and establish the optimal LADRC controller using the optimal bandwidth.

[0049] In step one, the LADRC controller parameter optimization algorithm based on gradient descent:

[0050] S1, initialize the bandwidth of the controller Set the learning rate η, the maximum number of iterations max, the convergence threshold th, the performance evaluation weight coefficients w1, w2, and w3;

[0051] S2, under the control of the latest LADRC controller of the robot, real-time acquisition of robot operation data, including the current acquisition cycle of robot linear speed v(t), current position and motor power P(t), calculation of current speed change Δv(t) = v(t)-v(t-1), position error e(t) = target position-current position;

[0052] S3, construct the performance evaluation function of the parameter automatic adjustment mechanism, and use the performance evaluation function to quantify the bandwidth ω c of the current controller:

[0053]

[0054] Where T represents the number of acquisition cycles from the start to the current acquisition cycle, and t represents the tthacquisition cycle;

[0055] S4, add a small disturbance Δω c to the current bandwidth ω c to generate the disturbed bandwidth

[0056] S5, calculate based on the disturbed bandwidth

[0057] S6, estimate the gradient of the performance evaluation function with respect to the current bandwidth ω c using the central difference method

[0058]

[0059] S7, update the bandwidth of the LADRC controller using gradient descent

[0060] Repeat the above S2-S7 iteration process until the number of iterations reaches the maximum number of iterations max or the performance evaluation index function J(ω c ) is less than the convergence threshold th after continuous multiple iterations, and the optimal bandwidth ω c= ω' c The optimal LADRC controller is established by applying to the LADRC controller.

[0061] Step two: The boundary conditions of the robot path planning are determined, including the position coordinates of the starting point and the ending point, and the corresponding speed and / or acceleration parameters. These parameters directly determine the initial and final dynamic characteristics of the target pose, and are the premise for the controller to achieve smooth tracking.

[0062] Step three: The current pose estimation data, current target pose data, and current surrounding environment information (including whether there are obstacles and the obstacle position data collected at times) of the robot are continuously collected by the laser radar and camera, and data screening is performed to eliminate obvious errors or redundant information, laying a reliable foundation for subsequent path generation. The data are sorted according to the motion sequence as the known pose sequence, thereby ensuring the time continuity of the path points.

[0063] Step four: A cubic spline interpolation function is constructed, which uses the boundary conditions and the known pose sequence for interpolation processing to generate a continuous and smooth path curve. The next target pose including the next position coordinate and motor speed is extracted from the path curve and given to the optimal LADRC controller. The optimal LADRC controller controls the robot to track and reach the next target pose. Step three is repeated until the robot reaches the ending point.

[0064] In step four, it is verified whether the generated path curve meets the kinematic and dynamic constraints of the robot, to ensure that the output target pose is within the controllable range of the controller.

[0065] The core of the cubic spline interpolation planning in step four is to provide a continuous and smooth target pose for the optimal LADRC controller, to ensure that there is no abrupt transition between adjacent target poses, and to meet the smoothness requirements of the robot motion.

[0066] In this embodiment, only the controller bandwidth ω c The observer convergence speed and the controller tracking performance can be simultaneously regulated, which greatly reduces the engineering debugging threshold, and is especially suitable for control scenarios with time-varying and complex disturbances, such as motor speed control.

[0067] In this embodiment, whether the robot can accurately and stably reach these target poses depends on the regulation performance of the LADRC controller after parameter optimization, forming a closed-loop cooperation of "path planning-controller tracking".

[0068] For example, Figure 5As shown, the embodiment of the present application also provides a robot adaptive path tracking control system in a complex environment, the robot adopts a double differential drive wheel and a double universal wheel chassis, two universal wheels are additionally arranged on the front side of the robot, and the control system comprises a training optimization module 1, a boundary setting module 2, a data acquisition module 3 and a path tracking module 4.

[0069] The training optimization module 1 is used for training the bandwidth of the LADRC controller used by the robot by using a gradient descent-based LADRC controller parameter optimization algorithm, obtaining an optimal bandwidth, and establishing an optimal LADRC controller by using the optimal bandwidth.

[0070] The boundary setting module 2 is used for explicitly defining the boundary conditions of the robot path planning, including the position coordinates of the starting point and the ending point, and the corresponding speed and / or acceleration parameters.

[0071] The data acquisition module 3 is used for continuously acquiring the current attitude estimation data, the current target attitude data and the current surrounding environment information of the robot by using a laser radar and a camera, performing data screening to eliminate obvious errors or redundant information, sorting according to the motion time sequence as a known attitude sequence, and the current surrounding environment information including whether there is an obstacle and the obstacle position data collected at times.

[0072] The path tracking module 4 is used for constructing a cubic spline interpolation function, the cubic spline interpolation function is used for interpolation processing to generate a continuous and smooth path curve by using the boundary conditions and the known attitude sequence, a next target attitude including a next position coordinate and a motor speed is extracted from the path curve and given to the optimal LADRC controller, the optimal LADRC controller controls the robot to track and reach the next target attitude, and the data acquisition module is repeatedly called until the robot reaches the ending point.

[0073] The embodiment of the present application also provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to call the instructions stored in the memory to execute the foregoing control method.

[0074] The embodiment of the present application also provides a computer-readable storage medium having computer program instructions stored thereon, and the computer program instructions are executed by a processor to implement the foregoing control method.

[0075] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for executing various aspects of the present application.

[0076] Although the specific embodiments of the present application have been described above, it is understood by those skilled in the art that these are merely illustrative, and the scope of protection of the present application is defined by the appended claims. Those skilled in the art can make various changes or modifications to the embodiments without departing from the principles and the essence of the present application, and such changes and modifications fall within the scope of protection of the present application.

Claims

1. A robot adaptive path tracking control method for complex environments, characterized in that: The robot uses a dual differential drive wheel and dual universal wheel chassis. The two universal wheels are added to the front of the robot. The control method includes: Step 1: Use the gradient descent-based LADRC controller parameter optimization algorithm to train the bandwidth of the LADRC controller used by the robot to obtain the optimal bandwidth, and use the optimal bandwidth to establish the optimal LADRC controller; Step 2: Define the boundary conditions for robot path planning, including the position coordinates of the starting and ending points, and the corresponding speed and / or acceleration parameters; Step 3: The robot continuously collects the current posture estimation data, the current target posture data, and the current surrounding environment information through the lidar and camera, and filters the data to eliminate obvious errors or redundant information. The data is sorted according to the motion time sequence as a known posture sequence. The current surrounding environment information includes whether there are obstacles and the obstacle location data collected at times. Step 4: Construct a cubic spline interpolation function. The cubic spline interpolation function uses boundary conditions and known posture sequences to perform interpolation processing to generate a continuous and smooth path curve. The next target posture including the next position coordinates and motor speed is extracted from it and given to the optimal LADRC controller. The optimal LADRC controller controls the robot to track and reach the next target posture. Repeat step 3 until the robot reaches the end point.

2. The method for adaptive path tracking control of a robot in a complex environment according to claim 1, wherein: In step 1, the LADRC controller parameter optimization algorithm based on gradient descent is: S1, initialization controller bandwidth Set the learning rate η, the maximum number of iterations max, the convergence threshold th, and the performance evaluation weight coefficients w1, w2, and w3; S2. Under the control of the robot using the latest LADRC controller, the robot's operating data is collected in real time, including the robot's linear velocity v(t), current position, and motor power P(t) during the current acquisition cycle. The current velocity change Δv(t) = v(t) - v(t-1) and the position error e(t) = target position - current position are calculated. S3. Construct a performance evaluation function for the automatic parameter adjustment mechanism and use the performance evaluation function to quantify the bandwidth ω of the current controller. c Control performance under: Where T represents the number of collection cycles from the start to the current collection cycle, and t represents the tth collection cycle; S4, the current bandwidth ω c Add a small perturbation Δω c , the bandwidth after generating the perturbation S5, based on the bandwidth after perturbation calculate S6, use the central difference method to estimate the performance evaluation function for the current bandwidth ω c Gradient S7. Update the bandwidth of the LADRC controller using gradient descent Repeat the above iterative process S2-S7 until the number of iterations reaches the maximum number of iterations max or the performance evaluation index function J(ω c ) is less than the convergence threshold th, and the optimal bandwidth ω is obtained c =ω′ c Applied to LADRC controller to establish the optimal LADRC controller.

3. The adaptive path tracking control method for a robot in a complex environment according to claim 1, characterized in that: In step 4, the generated path curve is verified to see if it complies with the robot's kinematic and dynamic constraints, ensuring that the output target posture is within the controllable range of the controller.

4. The method for controlling a robot's adaptive path tracking in a complex environment according to claim 1, wherein The speed of the left wheel in the dual differential drive is v l , right wheel speed is v R The distance between the two wheels is d, and the linear velocity of the robot in the plane Cartesian coordinate system is v = 0.5 (v l +v R ) and angular velocity The position of the robot in the Cartesian coordinate system (x, y, θ): x=vcosθ y=vsinθ θ=ω The heading angle θ is the angle between the robot and the x-axis in the plane Cartesian coordinate system. The robot's x-coordinate is updated according to the component of the linear velocity v in the x-direction, the y-coordinate is updated according to the y-component of the linear velocity in the direction, and the heading angle θ is updated according to the angular velocity ω.

5. A robot adaptive path tracking control system for complex environments, characterized by: The robot uses a dual differential drive wheel and dual universal wheel chassis. The two universal wheels are added to the front of the robot. The control system includes: A training optimization module is used to train the bandwidth of the LADRC controller used by the robot using a LADRC controller parameter optimization algorithm based on gradient descent to obtain the optimal bandwidth, and to establish an optimal LADRC controller using the optimal bandwidth; Boundary setting module, used to define the boundary conditions of robot path planning, including the position coordinates of the starting and ending points, and the corresponding speed and / or acceleration parameters; The data acquisition module is used to continuously collect the robot's current posture estimation data, current target posture data, and current surrounding environment information through the lidar and camera, and filter the data to eliminate obvious errors or redundant information, sorting them according to the motion time sequence as a known posture sequence. The current surrounding environment information includes whether there are obstacles and the obstacle location data collected at times; The path tracking module is used to construct a cubic spline interpolation function. The cubic spline interpolation function uses boundary conditions and known posture sequences to perform interpolation processing to generate a continuous and smooth path curve. The next target posture including the next position coordinates and motor speed is extracted from it and given to the optimal LADRC controller. The optimal LADRC controller controls the robot to track to the next target posture and repeatedly calls the data acquisition module until the robot reaches the end point.

6. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call instructions stored in the memory to execute the control method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the control method according to any one of claims 1 to 4 is implemented.

Images