Mobile robot control method and system

Abstract

The invention discloses a mobile robot control method and system, and relates to the technical field of mobile robot intelligent motion control, and the method comprises the steps: collecting environment feature data and motion state data; performing preprocessing and feature fusion to generate standardized state data; constructing a reinforcement learning training environment in an offline simulation environment, and obtaining a convergent reinforcement learning strategy model; inputting the current robot standardized state data into the reinforcement learning strategy model, and obtaining and outputting a preliminary motion control instruction and a reference trajectory; a prediction control model MPC is established, current robot standardized state data are input, short-time motion prediction and optimization are carried out, and a final control instruction meeting constraint conditions is obtained through solving; instructions are distributed to the four Mecanum wheel drivers, and the rotating speed and the direction of each wheel are adjusted. According to the invention, high-precision trajectory tracking, adaptive obstacle avoidance and energy efficiency optimization of the mobile robot in a complex dynamic environment are realized, and the system stability and the intelligent level are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile robot intelligent motion control, and in particular to a mobile robot control method and system. Background Art

[0002] With the widespread application of mobile robots in smart warehousing, logistics distribution, industrial manufacturing, and other fields, omnidirectional mobile robots based on Mecanum wheels have become a hot topic in academia and industry due to their flexible motion capabilities and spatial adaptability. In recent years, with the improvement of sensor accuracy and the rapid development of artificial intelligence technology, autonomous perception, path planning, and motion control methods for mobile robots have continued to advance. Traditional motion control methods are mainly based on PID, fuzzy control, and classical kinematic / dynamic modeling. Some cutting-edge research has begun to introduce intelligent algorithms such as deep learning and reinforcement learning to improve robots' adaptability to complex environments and their decision-making intelligence. In particular, in real-world scenarios such as dynamically changing environments, dense obstacles, and delicate operations, higher requirements are placed on the autonomy, robustness, and precision of robot control strategies.

[0003] However, existing mobile robot control technologies generally have a number of limitations. On the one hand, traditional motion control methods rely on precise model assumptions and manual parameter adjustment, making it difficult to cope with dynamic disturbances and high-dimensional inputs in complex unstructured environments, resulting in insufficient trajectory tracking accuracy and global planning capabilities. On the other hand, although the end-to-end control method based on single reinforcement learning has certain adaptive capabilities, it faces practical problems such as a lack of training samples, slow convergence speed, and the inability to guarantee safety under extreme working conditions. It is difficult to meet the high stability and real-time requirements of industrial-grade deployment. In addition, some existing fusion control strategies have obvious shortcomings in the effective fusion of kinematic constraints and multi-source environmental perception data, and the coordination of global planning and local optimization. They are unable to achieve efficient, accurate and safe control of omnidirectional mobile robots under complex tasks.

[0004] Therefore, there is an urgent need for an innovative mobile robot control method that can integrate multimodal perception data, combine global intelligent decision-making with local safety optimization, and have high robustness and real-time performance, so as to break through the bottleneck of existing technology and achieve high-precision autonomous control of the robot's omnidirectional motion in complex scenarios. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed, and the name of the present invention is a mobile robot control method and system.

[0006] Therefore, the technical problems solved by the present invention are: the existing mobile robot motion control methods have difficulties in fusing multi-source perception data to achieve global and local collaborative optimization, lack of adaptability to complex dynamic environments, insufficient motion control accuracy and safety, and how to achieve high-precision, robust and adaptive motion control of omnidirectional mobile robots based on Mecanum wheels in complex environments.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides a method for controlling a mobile robot, comprising collecting characteristic data of an environment in which a positioning object is located and motion state data of a robot body, to obtain the environmental characteristic data and the motion state data respectively; Preprocess and fuse the environmental feature data and motion state data to generate standardized state data; Based on the standardized state data as input, a reinforcement learning training environment is built in an offline simulation environment, and a converged reinforcement learning strategy model is obtained through iterative training; Collect and preprocess the current robot's standardized state data in real time, input it into the reinforcement learning strategy model, and obtain and output preliminary motion control instructions and reference trajectories; Establishing a predictive control model (MPC) based on the robot's kinematic and dynamic constraints. The MPC model takes preliminary motion control instructions and a reference trajectory as reference inputs, and simultaneously inputs the current robot's standardized state data, performs short-term motion prediction and optimization, and solves for the final control instructions that meet the constraints. The final control instructions are distributed to the four Mecanum wheel drivers to adjust the speed and direction of each wheel to achieve three-degree-of-freedom omnidirectional motion of the robot.

[0008] As a preferred embodiment of the mobile robot control method of the present invention, the collecting of characteristic data of the environment in which the positioning object is located and the motion state data of the robot body includes collecting information on obstacle distribution, ground type, spatial geometry, ambient light intensity, and dynamic changes in the environment to obtain environmental characteristic data; The robot's position information, posture information, speed information and the rotation speed information of the four driving wheels are collected to obtain motion state data.

[0009] As a preferred solution of the mobile robot control method described in the present invention, the preprocessing and feature fusion of environmental feature data and motion state data include, respectively, performing denoising, filtering, outlier removal, normalization and time series synchronization processing on the environmental feature data and motion state data, and performing feature splicing and dimension unification on the processed data according to a preset format, and fusing them into standardized state data.

[0010] As a preferred solution of the mobile robot control method described in the present invention, the reinforcement learning strategy model that obtains convergence through iterative training includes setting up a reinforcement learning training system including robot state, action space and reward function in a simulation environment, using standardized state data as input, iteratively training the robot motion control strategy through a reinforcement learning algorithm, and continuously optimizing the strategy parameters until the cumulative reward value reaches the preset convergence condition, thereby finally obtaining a reinforcement learning strategy model that can be used for robot motion control.

[0011] As a preferred solution of the mobile robot control method described in the present invention, the real-time collection and preprocessing of the current robot standardized state data, during the actual operation of the robot, periodically collects the current environmental feature data and motion state data, and generates the standardized state data of the current cycle according to the preset preprocessing process; the standardized state data of the current cycle is input into the reinforcement learning strategy model, and the reinforcement learning strategy model outputs the corresponding preliminary motion control instructions and reference trajectory information.

[0012] As a preferred embodiment of the mobile robot control method of the present invention, the establishment of an MPC model based on the robot's kinematic and dynamic constraints includes establishing, based on the structural parameters, motion mode, and force relationship of the mobile robot body, a kinematic model describing the relationship between the robot's displacement, velocity, acceleration, and the rotational speed of each wheel, and a dynamic model describing the robot's dynamic behavior. Based on the kinematic model and dynamic model, the robot position, posture, velocity and acceleration are selected as state variables to construct a predictive control optimization problem. Position constraints, posture constraints, velocity constraints, acceleration constraints, wheel speed constraints and driving force constraints are set in the optimization problem.

[0013] As a preferred solution of the mobile robot control method described in the present invention, the solution to obtain the final control instruction that meets the constraint conditions includes: in each control cycle, using the preliminary motion control instruction and reference trajectory as target input, and using the standardized state data of the current cycle as state input, using the predictive control model MPC to predict the motion state and optimize the control sequence in the future time period based on the set physical constraints, using a numerical optimization algorithm to solve the optimal control instruction sequence, and selecting the first control instruction from the optimal control instruction sequence as the final control instruction output of this cycle to drive the robot to actually move.

[0014] As a preferred embodiment of the mobile robot control method described in the present invention, the method of distributing the final control command to the four Mecanum wheel drivers includes decomposing the final control command into velocity components for movement along the X-axis, movement along the Y-axis, and rotation around the vertical axis according to the kinematics of the robot chassis; calculating the target speed and rotation direction of each drive wheel using an inverse kinematics algorithm based on the installation parameters and rolling direction of the four Mecanum wheels; and sending the target speed and rotation direction of each drive wheel to the corresponding driver, thereby controlling the four Mecanum wheels to work in coordination and achieving three-degree-of-freedom omnidirectional motion of the robot within a plane.

[0015] In a second aspect, an embodiment of the present invention provides a mobile robot control system, comprising: Data acquisition module: collects characteristic data of the environment in which the positioning object is located and the motion state data of the robot body, and obtains environmental characteristic data and motion state data respectively; Data preprocessing and feature fusion module: preprocesses and fuses environmental feature data and motion state data to generate standardized state data; Reinforcement learning strategy model training module: Based on standardized state data as input, a reinforcement learning training environment is constructed in an offline simulation environment, and a converged reinforcement learning strategy model is obtained through iterative training; Reinforcement learning reasoning and command generation module: This module collects and preprocesses the current robot's standardized state data in real time, inputs it into the reinforcement learning strategy model, and obtains and outputs preliminary motion control commands and reference trajectories. Predictive control model optimization module: This module establishes a predictive control model (MPC) based on the robot's kinematic and dynamic constraints. The MPC takes preliminary motion control instructions and reference trajectories as input, along with the current robot's standardized state data, to perform short-term motion prediction and optimization, ultimately obtaining the final control instructions that satisfy the constraints. Chassis drive distribution and execution module: distributes the final control instructions to the four Mecanum wheel drives, adjusts the speed and direction of each wheel, and realizes the robot's three-degree-of-freedom omnidirectional motion.

[0016] The beneficial effects of the present invention are as follows: By integrating multi-source environmental feature data with robot motion state data to construct standardized state data input, this invention achieves precise perception of the robot and its environment. This control strategy, which combines reinforcement learning with a predictive control model (MPC), combines global autonomous decision-making capabilities with local safety optimization performance. This not only improves the robot's motion control accuracy and path tracking robustness in complex and dynamic environments, but also effectively ensures the system's real-time performance and operational safety. Compared to existing technologies, this invention enables efficient autonomous control of Mecanum-wheeled omnidirectional mobile robots in multi-dimensional, highly dynamic scenarios, significantly expanding the robot's adaptability and engineering application value in demanding applications such as intelligent warehousing and automated handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 This is an overall flow chart of a mobile robot control method provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0019] Example 1, with reference to Figure 1 , as an embodiment of the present invention, provides a mobile robot control method, comprising: S1: Collect characteristic data of the environment where the positioning object is located and the motion state data of the robot body to obtain environmental characteristic data and motion state data respectively.

[0020] Collect information on obstacle distribution, ground type, spatial geometry, ambient light intensity, and dynamic changes in the environment to obtain environmental feature data; collect the robot's position information, posture information, speed information, and the rotation speed information of the four driving wheels to obtain motion state data.

[0021] The S1 step of the present invention clearly stipulates the specific content of collecting environmental feature data and robot body motion state data, which significantly enhances the implementation basis and engineering feasibility of subsequent control methods. Specifically, environmental feature data includes obstacle distribution (i.e., detecting the spatial coordinates and boundaries of obstacles in the environment through lidar or visual sensors), ground type (such as flat, ramp, steps, distinguished by force sensors or visual analysis), spatial geometry (such as shape information of walls, corners, openings, etc. in space), ambient light intensity (which can be monitored in real time by light sensors), and dynamic environmental change information (such as the appearance / disappearance of moving obstacles, personnel flow, etc., which relies on dynamic visual tracking). The motion state data of the robot body includes position information (usually obtained by encoders or fusion SLAM to obtain the robot's global / local coordinates), posture information (obtaining the robot's current angle of rotation or posture quaternion through IMU or gyroscope), speed information (the current linear velocity and angular velocity are calculated by encoders and IMUs), and the rotational speed information of the four drive wheels (the current rotational speed of each wheel is independently collected by wheel speed sensors or encoders).

[0022] The core of this step lies in the full collection and categorized, structured management of multimodal perception data, providing a high-dimensional and rich input foundation for subsequent feature fusion and intelligent control algorithms. Unlike existing approaches that rely solely on a single source (e.g., using odometry or monocular vision to estimate motion state), this method effectively improves the accuracy and robustness of state observation by jointly collecting multi-source heterogeneous data. For example, the combined collection of obstacle distribution and spatial structure significantly enhances the feasibility of obstacle avoidance and path planning. The simultaneous acquisition of information such as ground type and light intensity provides critical perception input for subsequent adaptive motion control (e.g., slip detection and compensation). The collection of independent drive wheel speeds not only directly monitors the actual performance of each wheel but also enables real-time detection of wheel discrepancies and mechanical failures, enhancing system health management and motion consistency.

[0023] S2: Preprocess and fuse the environmental feature data and motion state data to generate standardized state data.

[0024] The environmental feature data and motion state data are subjected to denoising, filtering, outlier elimination, normalization and time series synchronization respectively, and the processed data are subjected to feature splicing and dimension unification according to a preset format and fused into standardized state data.

[0025] In this embodiment of the present invention, environmental feature data and motion state data are subjected to denoising, filtering, outlier removal, normalization, and time-series synchronization, before being combined with feature concatenation and dimensionality to form standardized state data. This process addresses the heterogeneity of multi-source sensory data and the uncertainty inherent in actual acquisition, ensuring the validity and comparability of input data. Specifically, denoising processing (such as median filtering, Kalman filtering, etc.) can effectively eliminate instantaneous interference, sudden pulses and high-frequency noise in the collected signals; filtering processing smoothes the signal for specific noise patterns such as low-pass and high-pass to improve signal quality; outlier removal processing automatically removes abnormal points that exceed the normal distribution range by setting statistical thresholds, sliding windows or machine learning discrimination methods to prevent extreme errors from affecting the overall judgment; normalization processing unifies data of different dimensions and different value ranges to [0,1] or zero mean unit variance interval through linear normalization, Z-score standardization and other methods, which helps to eliminate dimensional interference between various features and make subsequent feature fusion and model input more efficient and robust; time series synchronization processing uses algorithms such as interpolation alignment or distribution alignment to ensure that multi-source data are effectively aligned at the same time point and the same sampling frequency to prevent information lag or dislocation.

[0026] Feature concatenation and dimensionality unification further combine all preprocessed features in a defined format (e.g., one-dimensional vectors, feature matrices), ensuring that each piece of standardized state data is complete, without loss, and compatible with subsequent algorithm input requirements. For high-dimensional perceptual features, methods such as principal component analysis (PCA) and linear discriminant analysis (LDA) can be used for dimensionality reduction and unification, effectively reducing redundancy and highlighting key features.

[0027] S3: Based on the standardized state data as input, a reinforcement learning training environment is constructed in an offline simulation environment, and a converged reinforcement learning strategy model is obtained through iterative training.

[0028] In a simulation environment, a reinforcement learning training system including robot state, action space and reward function is set up. With standardized state data as input, the robot motion control strategy is iteratively trained through the reinforcement learning algorithm, and the strategy parameters are continuously optimized until the cumulative reward value reaches the preset convergence condition. Finally, a reinforcement learning strategy model that can be used for robot motion control is obtained.

[0029] In this embodiment of the present invention, the essence of step S3 is to use standardized state data that has undergone rigorous preprocessing and feature fusion as input to build a controllable and highly realistic reinforcement learning training platform in a simulation environment. This training environment first sets the robot state (such as position, speed, posture, and environmental characteristics), the action space (the robot's optional control variables such as speed, steering, and wheel speed), and the reward function (a function used to measure the quality of each behavioral strategy), so that the training system can truly reflect the actual movement process and environmental interaction. Using standardized state data as input not only ensures the standardization and consistency of training samples, but also makes the reinforcement learning model highly adaptable in various environments and working conditions.

[0030] Reinforcement learning algorithms typically include architectures such as deep Q-networks (DQNs), policy gradient methods, and actor-critic algorithms. Through repeated interactions with a simulated environment, the agent selects actions based on its current state. The environment returns a new state and reward value, and the model continuously adjusts policy parameters based on accumulated rewards. The iterative training process dynamically optimizes the decision network structure based on continuous experience collection and policy updates, achieving adaptive policy adjustments. Model convergence criteria typically include cumulative rewards stabilizing at a preset threshold, the policy network output no longer changing significantly, or the completion rate of the simulation task reaching a specified target.

[0031] The offline simulation training platform can significantly reduce the time and cost of actual robot physical experiments, avoid equipment damage caused by trial and error under extreme conditions, and improve training efficiency and sample diversity. At the same time, by fully leveraging the high-dimensional characteristics of standardized state data, the trained policy model is compatible with a variety of complex scenarios (such as dense obstacles, dynamic environmental changes, etc.), improving the generalization and robustness of decision-making capabilities. Compared with existing data-driven methods based on static strategies or relying on a large number of real-world experiments, the S3 process of the present invention not only improves the engineering deployability of the control policy model, but also lays the theoretical and data foundation for subsequent online deployment and adaptive adjustment, significantly expanding the technical boundaries and application breadth of intelligent mobile robot motion control.

[0032] S4: Collect and preprocess the current robot's standardized state data in real time, input it into the reinforcement learning strategy model, and obtain and output preliminary motion control instructions and reference trajectories.

[0033] During the actual operation of the robot, the current environmental feature data and motion state data are periodically collected, and the standardized state data of the current cycle is generated according to the preset preprocessing process; the standardized state data of the current cycle is input into the reinforcement learning strategy model, and the reinforcement learning strategy model outputs the corresponding preliminary motion control instructions and reference trajectory information.

[0034] In the embodiment of the present invention, step S4 emphasizes that during the actual deployment and operation phase of the robot, the system can periodically (e.g., every 10ms, 20ms, or 50ms sampling period) collect the latest environmental feature data and robot body motion state data, and generate standardized state data for the current cycle through a preprocessing process consistent with the offline training phase (including denoising, normalization, time series synchronization, etc.). Here, "standardized state data for the current cycle" not only includes external features such as obstacle distribution, ground type, spatial geometry, light intensity, and dynamic changes in the environment collected in real time, but also covers the robot's own internal motion state, such as position information, posture information, speed information, and wheel speed. This comprehensive, high-frequency data input provides the information basis for the system to achieve adaptive decision-making in complex and dynamic environments.

[0035] This "standardized state data of the current cycle" is passed as input to the offline-trained reinforcement learning policy model in this step. The reinforcement learning policy model is essentially a function mapper trained through a deep network. It can quickly infer the optimal preliminary motion control instructions (such as target speed, steering angle, trajectory point sequence, etc.) and reference trajectory based on the current input state. Among them, the preliminary motion control instructions are usually global instructions for the robot chassis, such as specifying the three-degree-of-freedom speed or spatial posture transformation of the next cycle, and the reference trajectory is the sequence of spatial path points that the robot is expected to pass through in the short term. The model inference process does not require complex online search or model retraining, thus ensuring high real-time performance and deployment efficiency.

[0036] This step is to achieve efficient and seamless connection between the deep reinforcement learning strategy and the actual motion process of the robot, and to respond dynamically and in real time to the environment and changes in the robot itself. Compared with traditional control schemes that rely solely on manual rules or static models, the present invention ensures that the strategy model can generate adaptive preliminary control instructions under various working conditions by inputting high-dimensional, standardized multi-source data in real time, significantly improving trajectory tracking accuracy, environmental adaptability and motion robustness. In addition, this real-time reasoning architecture can also be deeply integrated with the downstream predictive control module to provide accurate and real-time reference input for subsequent physical constraint optimization and execution control. This mechanism has shown extremely high engineering practical value and technology promotion prospects in practical applications such as dynamic obstacle avoidance, collaborative motion, and energy consumption optimization, which fully proves the feasibility and advancement of the content of the present invention.

[0037] S5: Establish a predictive control model MPC based on the robot's kinematic and dynamic constraints. The predictive control model MPC takes the preliminary motion control instructions and reference trajectory as reference inputs, and simultaneously inputs the current robot's standardized state data to perform short-term motion prediction and optimization, and solves the final control instructions that meet the constraints.

[0038] Based on the structural parameters, motion mode and force relationship of the mobile robot body, a kinematic model describing the relationship between the robot's displacement, velocity, acceleration and the rotation speed of each wheel, as well as a dynamic model describing the robot's dynamic behavior, are established. Based on the kinematic model and dynamic model, the robot position, posture, velocity and acceleration are selected as state variables to construct a predictive control optimization problem. Position constraints, posture constraints, velocity constraints, acceleration constraints, wheel speed constraints and driving force constraints are set in the optimization problem.

[0039] In an embodiment of the present invention, the kinematic model specifically refers to a mathematical expression that describes the mapping relationship between the robot's global coordinate position, local velocity, acceleration, and the independent rotational speed of each Mecanum wheel based on the robot's chassis structure, wheel arrangement, and mounting angle. For example, three-degree-of-freedom omnidirectional motion can be bidirectionally converted between wheel speed and chassis speed (linear velocity and angular velocity) using a coordinate transformation matrix. The dynamic model, based on the robot's mass, moment of inertia, and wheel-ground friction parameters, describes the dynamic response process of each state variable of the robot when subjected to an input driving force. This model not only takes into account the inertia of the robot body, but also encompasses non-ideal effects in actual operation such as friction and load changes.

[0040] This step introduces both kinematic and dynamic models to achieve a high-precision, multi-level description of the robot's actual motion process, which is significantly better than traditional control methods that only use simplified kinematic modeling. The robot's position, posture, velocity, and acceleration are used as core state variables, and directly controlled physical quantities such as wheel speed and driving force are set as optimization constraints, so that the control model can strictly limit the robot to operate within the physical limits and safe area. By fully introducing multi-dimensional constraints such as position, posture, velocity, and acceleration, high-precision trajectory tracking can be guaranteed, and theoretical guarantees can be provided for improving robustness and safety under actual complex working conditions.

[0041] Furthermore, in each control cycle, the preliminary motion control instructions and reference trajectory are used as target inputs, and the standardized state data of the current cycle is used as state input. The predictive control model MPC is used to predict the motion state and optimize the control sequence in the future time period based on the set physical constraints. The numerical optimization algorithm is used to solve the optimal control instruction sequence, and the first control instruction from the optimal control instruction sequence is selected as the final control instruction output of this cycle to drive the robot to actually execute motion.

[0042] It should also be noted that within each control cycle, the system receives preliminary motion control commands and reference trajectories (as desired targets) from the reinforcement learning module in real time. It then inputs the standardized, pre-processed, fused state data from the current cycle into the MPC optimizer. Using kinematic and dynamic models, combined with the aforementioned physical constraints, MPC predicts motion states for a set future timeframe (e.g., within 1-2 seconds), generating a set of optimal control command sequences covering all control moments (e.g., one decision point every 100 milliseconds, covering the next 10 steps).

[0043] During the optimization process, numerical optimization algorithms (such as quadratic programming, sequential quadratic programming, and interior-point methods) are used to solve the constrained control problem, ensuring that the predicted trajectory remains within physical and environmental boundaries at every moment. Ultimately, only the first instruction in the optimal control sequence is selected as the final execution instruction issued to the drive system for the current control cycle. This "rolling optimization-rolling execution" approach not only significantly enhances the system's adaptability to dynamic environmental changes but also allows for real-time correction of trajectory deviations caused by modeling errors or external disturbances.

[0044] This section aims to organically combine adaptive global decision-making with on-body safety control by synergizing MPC with reinforcement learning modules and utilizing a closed-loop optimization mechanism with physical constraints. This approach balances global planning capabilities with local physical feasibility, significantly improving the robot's control accuracy and motion safety in dynamic and complex environments. Compared to existing solutions that only use RL or pure MPC, this invention can better balance global optimization with real-time response, motion flexibility with safety margins, and enhance the feasibility and stability of project implementation.

[0045] S6: Distributes the final control instructions to the four Mecanum wheel drivers, adjusting the speed and direction of each wheel to achieve three-degree-of-freedom omnidirectional motion of the robot.

[0046] The final control command is decomposed into velocity components for movement along the X-axis, movement along the Y-axis, and rotation around the vertical axis according to the kinematics of the robot chassis. Based on the installation parameters and rolling direction of the four Mecanum wheels, an inverse kinematics algorithm is used to calculate the target speed and rotation direction of each drive wheel. These target speeds and rotation directions are then sent to the corresponding actuators, controlling the four Mecanum wheels to work together and achieve three-degree-of-freedom omnidirectional motion of the robot within a plane.

[0047] The core of step S6 is to convert the final control instructions obtained by MPC optimization into independent motion parameters of each Mecanum wheel that can be actually driven, so as to achieve omnidirectional motion capability with three degrees of freedom (X-axis translation, Y-axis translation, and rotation around the vertical axis). Specifically, the final control instructions usually include the linear velocity of the robot in the global coordinate system ( 、 ) and angular velocity ( ), these global instructions need to be mapped into the target speed and rotation direction of each wheel through the inverse kinematic solution of the robot chassis.

[0048] The arrangement of the four Mecanum wheels determines their contribution to the decomposition of different motion components. By establishing the inverse kinematic transformation matrix (for example: ; in is the kinematic coefficient matrix, For the The speed of the wheels, ∈{1,2,3,4}), which efficiently maps the global speed command into the target speed and direction of rotation for each drive wheel. The system then sends these target speeds and directions to the motor drivers via a real-time signal transmission interface, enabling the coordinated movement of the four Mecanum wheels.

[0049] This step uses a command allocation mechanism based on the inverse solution of global kinematics to ensure that the driving action of each wheel is strictly consistent with the overall desired motion of the robot, significantly improving the accuracy of trajectory tracking and posture control. Compared with traditional direct drive or simple speed distribution methods, this method fully considers the wheel arrangement, chassis parameters and mechanical effects, thereby effectively suppressing practical problems such as steering deviation, sideslip or single wheel failure, and improving system robustness. This distributed and refined drive command allocation also provides a basic interface for subsequent advanced control such as fault detection, active slip compensation and energy consumption optimization, greatly expanding the practical application boundaries and technical depth of omnidirectional mobile robots.

[0050] Example 2 is the second embodiment of the present invention, which is different from the previous embodiment in that: If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art or the portion of the current technical solution, can be embodied in the form of a software product. The current computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0051] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0052] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0053] Example 3 is an embodiment of the present invention, which provides a mobile robot control system, including a data acquisition module, a data preprocessing and feature fusion module, a reinforcement learning strategy model training module, a reinforcement learning reasoning and instruction generation module, a predictive control model optimization module and a chassis drive allocation and execution module.

[0054] Data acquisition module: collects characteristic data of the environment in which the positioning object is located and the motion state data of the robot body, and obtains environmental characteristic data and motion state data respectively; Data preprocessing and feature fusion module: preprocesses and fuses environmental feature data and motion state data to generate standardized state data; Reinforcement learning strategy model training module: Based on standardized state data as input, a reinforcement learning training environment is constructed in an offline simulation environment, and a converged reinforcement learning strategy model is obtained through iterative training; Reinforcement learning reasoning and command generation module: This module collects and preprocesses the current robot's standardized state data in real time, inputs it into the reinforcement learning strategy model, and obtains and outputs preliminary motion control commands and reference trajectories. Predictive control model optimization module: This module establishes a predictive control model (MPC) based on the robot's kinematic and dynamic constraints. The MPC takes preliminary motion control instructions and reference trajectories as input, along with the current robot's standardized state data, to perform short-term motion prediction and optimization, ultimately obtaining the final control instructions that satisfy the constraints. Chassis drive distribution and execution module: distributes the final control instructions to the four Mecanum wheel drives, adjusts the speed and direction of each wheel, and realizes the robot's three-degree-of-freedom omnidirectional motion.

[0055] Example 4 is an embodiment of the present invention, which provides a mobile robot control method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation / comparative experiments.

[0056] A comparative experiment comparing the path tracking and dynamic obstacle avoidance performance of an omnidirectional mobile robot was conducted in a smart warehouse experimental area. The robot was equipped with four Mecanum wheels. Its sensor configuration included a LiDAR (for obstacle distribution), an RGBD vision module (for surface type and dynamic environment), an inertial measurement unit (IMU), and encoders (for velocity, wheel speed, and position information). The experimental area was filled with pre-set obstacles and dynamically moving obstacles to simulate complex routes between shelves and temporary obstacles.

[0057] The experiment was divided into two groups: the control group adopted the classic PID + segmented planning control, and the experimental group adopted the "standardized state data + reinforcement learning strategy + MPC collaborative optimization" control method proposed in this invention. All robots were first statically calibrated and initialized, and then completed multiple rounds of tasks under the unified starting and ending points and multi-segment and multi-inflection point routes. In each round, environmental features and body motion states were collected in real time, and standardized inputs were generated through data preprocessing and feature fusion. The experimental group had previously used the simulation environment of this scenario to complete the offline training of the reinforcement learning strategy model, deployed it on the robot side, and performed real-time inference during online operation and combined it with MPC to form dynamic optimal motion control instructions. Finally, each round of experiments fully recorded key parameters such as the robot's trajectory deviation, average tracking error, maximum obstacle avoidance reaction time, number of local path adjustments, and energy consumption.

[0058] Among them, comparison group-1 pursues a balance between accuracy and speed, and the specific parameters are: Proportional / integral / derivative gains: Kp 1.30, Ki 0.05, Kd 0.17; Maximum linear speed: 1.05 m / s; Maximum linear acceleration: 1.4 m / s²; Replanning period: 140 ms; Obstacle expansion radius: 0.19 m; Pure tracking forward range: 0.55 m; Maximum wheel speed: 230 rpm; Comparison Group-2 uses conservative obstacle avoidance, emphasizing stability (corresponding to slower speed and larger error in Table 1). Specific parameters are: Kp 1.10, Ki 0.06, Kd 0.15; Maximum linear speed: 0.92 m / s; Maximum linear acceleration: 1.1 m / s²; Replanning cycle: 180 ms (slower, resulting in delayed dynamic response); Obstacle expansion radius: 0.23 m (larger, more conservative detour); Front sight distance: 0.45 m; Maximum wheel speed: 220 rpm; Comparison group-3 is more aggressive tracking (corresponding to slightly faster speed and slightly lower error in Table 1). The specific parameters are: Kp 1.42, Ki 0.04, Kd 0.19; Maximum linear speed: 1.10 m / s; Maximum linear acceleration: 1.6 m / s²; Replanning period: 120 ms; Obstacle expansion radius: 0.17 m; Front sight distance: 0.60 m; Maximum wheel speed: 240 rpm.

[0059] The conservative configuration of comparison group 2 (larger expansion radius and slower replanning cycle) resulted in the lowest average speed, increased number of detours, and amplified maximum error; the higher speed and shorter replanning cycle of comparison group 3 improved the average error, but with slightly higher energy consumption; comparison group 1 was between the two.

[0060] Experimental group 1 is a tracking-first trade-off (corresponding to the lower error in Table 1), with the following parameters: RL policy checkpoint: ckpt-178k (frozen inference, no additional exploration); MPC prediction horizon: N = 10 steps; Cost weights: tracking error weight 1.00, control increment weight 0.28, smoothing weight 0.22; Physical constraints: maximum wheel speed 240 rpm, maximum linear acceleration 1.5 m / s²; Safety margin (MPC internal expansion): 0.20 m; Experimental group 2 prioritizes energy efficiency and smoothness (corresponding to the lowest energy consumption, highest speed, but slightly increased error in Table 1). The specific parameters are: RL policy checkpoint: ckpt-195k (temperature inference T = 0.05, to improve perturbation robustness); MPC prediction time domain: N = 12 steps (longer look-ahead, good for speed maintenance); Cost weights: tracking error weight 0.90, control increment weight 0.35, smoothing weight 0.30; Physical constraints: maximum wheel speed 235 rpm, maximum linear acceleration 1.4 m / s²; Safety margin: 0.18 m; Experimental Group 3 prioritizes safety margins (corresponding to the highest completion rate, slightly higher error, and slightly higher energy consumption in Table 1). Specific parameters are: RL policy checkpoint: ckpt-210k (frozen inference); MPC prediction horizon: N = 10 steps; Cost weights: tracking error weight 1.20, control increment weight 0.30, smoothing weight 0.24; Physical constraints: maximum wheel speed 240 rpm, maximum linear acceleration 1.5 m / s²; Safety margin: 0.22 m (larger, more conservative for passing through narrow passages).

[0061] Experimental Group 1 prioritizes tracking accuracy and has the lowest average error. Experimental Group 2 increases the weights of smoothing and control costs and extends the prediction time domain, reducing energy consumption while maintaining speed, resulting in a slight increase in average error. Experimental Group 3 increases the safety margin, adopts a more conservative path, and has the highest completion rate, but with slightly higher energy consumption and error.

[0062] Table 1 summarizes the experimental data, showing the performance of a control group (traditional method) and an experimental group (the proposed solution) under the same test task. The units of each parameter are indicated in the table header. The results demonstrate the robot's technical advantages of high-precision path tracking and robustness in highly dynamic environments.

[0063] Table 1 Data reference table

[0064] As shown in Table 1, the mobile robot using the present invention's solution (experimental group) significantly outperformed the conventional control group in a number of key performance indicators, including path tracking accuracy, obstacle response speed, energy consumption, and task completion rate, in intelligent warehousing scenarios. The experimental group's average path tracking error was only 10.5 mm, a reduction of over 50% compared to the control group's 21.4 mm. The maximum deviation also dropped from an average of 57.8 mm in the control group to 27.2 mm, significantly improving high-precision motion control capabilities. The experimental group's average dynamic obstacle avoidance reaction time was 208 ms, an improvement of over 45% compared to the control group's 378 ms. This demonstrates that the reinforcement learning + MPC mechanism can more quickly adapt to sudden environmental changes and promptly modify motion strategies. The average number of local path adjustments was reduced by half, indicating that the robot no longer needs to make frequent and drastic path changes in complex environments, effectively improving motion continuity and safety.

[0065] Furthermore, the experimental group's average energy consumption was 18.6 Wh, approximately 18% lower than the control group's 22.8 Wh. This was attributed to improved energy-efficient coordinated control and reduced ineffective motion strategies under MPC optimization. In terms of task completion rate, the experimental group achieved or exceeded 99% almost throughout the entire process, while the control group experienced significant task failures and deviations from the track. These results fully demonstrate the robustness and adaptability of the proposed solution in highly dynamic and complex environments.

[0066] This significantly improved technical effect is rooted in the decision-making optimization capability of the present invention under the integrated collaboration of perception-feature fusion-deep reinforcement learning-model predictive control. Multi-source data fusion and standardized processing provide high-dimensional, accurate environmental and ontological perception inputs for intelligent algorithms, effectively reducing the drawbacks of traditional control that relies too much on model accuracy and parameters. The introduction of reinforcement learning significantly enhances the adaptive ability of motion strategies in changing scenarios, while the MPC link ensures the optimal physical constraints and energy consumption of motion execution. The comparison group relies solely on empirical parameter adjustment, optimization without physical constraints, and weak environmental adaptation. It is unable to maintain trajectory accuracy and energy consumption balance in complex environments, and is also susceptible to dynamic interference affecting task stability. Therefore, the tabular data directly reflects the technological breakthroughs of the present invention's solution in terms of accuracy, response speed, robustness, and energy efficiency, fully reflecting its creativity and engineering practical value.

[0067] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A mobile robot control method, characterized in that: include: Collect characteristic data of the environment in which the positioning object is located and motion state data of the robot body to obtain environmental characteristic data and motion state data respectively; Preprocess and fuse the environmental feature data and motion state data to generate standardized state data; Based on the standardized state data as input, a reinforcement learning training environment is built in an offline simulation environment, and a converged reinforcement learning strategy model is obtained through iterative training; Collect and preprocess the current robot's standardized state data in real time, input it into the reinforcement learning strategy model, and obtain and output preliminary motion control instructions and reference trajectories; Establishing a predictive control model (MPC) based on the robot's kinematic and dynamic constraints. The MPC model takes preliminary motion control instructions and a reference trajectory as reference inputs, and simultaneously inputs the current robot's standardized state data, performs short-term motion prediction and optimization, and solves for the final control instructions that meet the constraints. The final control instructions are distributed to the four Mecanum wheel drivers to adjust the speed and direction of each wheel to achieve three-degree-of-freedom omnidirectional motion of the robot.

2. The mobile robot control method according to claim 1, wherein: The collecting of characteristic data of the environment in which the positioning object is located and the motion state data of the robot body includes collecting information on obstacle distribution, ground type, spatial geometry, ambient light intensity, and dynamic changes in the environment to obtain environmental characteristic data; The robot's position information, posture information, speed information and the rotation speed information of the four driving wheels are collected to obtain motion state data.

3. The mobile robot control method according to claim 2, wherein: The preprocessing and feature fusion of the environmental feature data and the motion state data include performing denoising, filtering, outlier elimination, normalization and time series synchronization on the environmental feature data and the motion state data, respectively, and performing feature splicing and dimension unification on the processed data according to a preset format to fuse them into standardized state data.

4. The mobile robot control method according to claim 3, wherein: The reinforcement learning strategy model that converges through iterative training includes setting up a reinforcement learning training system including robot state, action space and reward function in a simulation environment, using standardized state data as input, iteratively training the robot motion control strategy through a reinforcement learning algorithm, and continuously optimizing the strategy parameters until the cumulative reward value reaches a preset convergence condition, thereby finally obtaining a reinforcement learning strategy model that can be used for robot motion control.

5. The mobile robot control method according to claim 4, wherein: The obtaining and outputting of preliminary motion control instructions and reference trajectories includes, during the actual operation of the robot, periodically collecting current environmental feature data and motion state data, generating standardized state data of the current cycle according to a preset preprocessing process; inputting the standardized state data of the current cycle into a reinforcement learning strategy model, and the reinforcement learning strategy model outputting corresponding preliminary motion control instructions and reference trajectory information.

6. The mobile robot control method according to claim 5, wherein: The establishment of the predictive control model MPC based on the robot's kinematic and dynamic constraints includes establishing a kinematic model describing the relationship between the robot's displacement, velocity, acceleration, and the rotational speed of each wheel, and a dynamic model describing the robot's dynamic behavior, based on the structural parameters, motion mode, and force relationship of the mobile robot body; Based on the kinematic model and dynamic model, the robot position, posture, velocity and acceleration are selected as state variables to construct a predictive control optimization problem. Position constraints, posture constraints, velocity constraints, acceleration constraints, wheel speed constraints and driving force constraints are set in the optimization problem.

7. The mobile robot control method according to claim 6, wherein: The method of solving the final control instruction that satisfies the constraint conditions includes: in each control cycle, using the preliminary motion control instruction and the reference trajectory as the target input, and using the standardized state data of the current cycle as the state input, using the predictive control model MPC to predict the motion state and optimize the control sequence in the future time period based on the set physical constraints, using a numerical optimization algorithm to solve the optimal control instruction sequence, and selecting the first control instruction from the optimal control instruction sequence as the final control instruction output of this cycle to drive the robot to actually move.

8. The mobile robot control method according to claim 7, wherein: The method of distributing the final control command to the four Mecanum wheel drivers includes decomposing the final control command into velocity components of movement along the X-axis, movement along the Y-axis, and rotation about the vertical axis according to the kinematics of the robot chassis, calculating the target speed and rotation direction of each drive wheel based on the installation parameters and rolling direction of the four Mecanum wheels using an inverse kinematics algorithm, and sending the target speed and rotation direction of each drive wheel to the corresponding driver respectively, thereby controlling the four Mecanum wheels to work together to achieve three-degree-of-freedom omnidirectional motion of the robot within a plane.

9. A mobile robot control system, used to implement the mobile robot control method according to any one of claims 1 to 8, characterized in that: include: Data acquisition module: collects characteristic data of the environment in which the positioning object is located and the motion state data of the robot body, and obtains environmental characteristic data and motion state data respectively; Data preprocessing and feature fusion module: preprocesses and fuses environmental feature data and motion state data to generate standardized state data; Reinforcement learning strategy model training module: Based on standardized state data as input, a reinforcement learning training environment is constructed in an offline simulation environment, and a converged reinforcement learning strategy model is obtained through iterative training; Reinforcement learning reasoning and command generation module: This module collects and preprocesses the current robot's standardized state data in real time, inputs it into the reinforcement learning strategy model, and obtains and outputs preliminary motion control commands and reference trajectories. Predictive control model optimization module: This module establishes a predictive control model (MPC) based on the robot's kinematic and dynamic constraints. The MPC takes preliminary motion control instructions and reference trajectories as input, along with the current robot's standardized state data, to perform short-term motion prediction and optimization, ultimately obtaining the final control instructions that satisfy the constraints. Chassis drive distribution and execution module: distributes the final control instructions to the four Mecanum wheel drives, adjusts the speed and direction of each wheel, and realizes the robot's three-degree-of-freedom omnidirectional motion.

Images