A mobile robot simulation training platform and device

Abstract

The application relates to the field of robot simulation and intelligent control technology, and discloses a mobile robot simulation training platform and equipment, which comprises a hardware layer, an algorithm layer and an interaction layer; the hardware layer is used for constructing a three-dimensional simulation environment comprising a mobile robot model capable of carrying a task load, a task scene model and environmental variables; the algorithm layer is integrated with a path planning algorithm, a target recognition algorithm and a motion control algorithm; and the interaction layer provides a parameter adjustment interface. The platform working process is as follows: firstly, the robot and the environment are initialized; then, a target is identified and an optimal path is planned through target detection and path planning; then, the robot is driven to move and real-time obstacle avoidance is performed through motion control and obstacle avoidance; finally, performance evaluation and parameter optimization are performed. Through real parameter modeling and closed-loop simulation training, the application solves the problems of high traditional development cost, long cycle and distorted simulation results, and provides reliable support for the research and development and algorithm verification of mobile robots.

Description

Technical Field

[0001] This application relates to the field of robot simulation and intelligent control technology, specifically a mobile robot simulation training platform and equipment. Background Technology

[0002] The development of traditional mobile robots (such as robots used for tasks like object retrieval and site cleaning) has long relied on repeated testing and debugging of physical equipment. Taking a tennis ball-retrieving robot as an example, this method suffers from high R&D costs, significant hardware wear and tear, long testing cycles, and low efficiency. More critically, most existing robot simulation platforms employ simplified kinematic models and idealized sensor parameters, lacking high-precision modeling of the robot's actual physical dimensions, precise sensor parameters (such as camera distortion, LiDAR point cloud noise, and IMU errors), and complex environmental variables (such as dynamic lighting conditions, ground friction and adhesion coefficients, and target object distribution density). This significant deviation between the model and reality prevents the effective transfer of algorithm training and testing results from simulation environments to real robots and their working scenarios, severely limiting the practical value of simulation platforms in the development and verification of robot perception, planning, and control algorithms.

[0003] Furthermore, existing simulation platforms are typically designed for robots with a single, fixed form and task (such as a fixed grasping arm), lacking simulation support for modular mobile robot platforms with interchangeable payloads. This makes it impossible to conduct efficient algorithm iteration and performance evaluation in a unified simulation environment when developing next-generation or multi-functional robots (e.g., a ball-collecting module or a cleaning module mounted on the same chassis), further exacerbating the gap between virtual simulation and physical deployment.

[0004] Therefore, there is an urgent need for a general simulation training platform that can accurately simulate real robot dynamics, sensor characteristics and their interaction with complex environments, and support modular task payloads, in order to accelerate the process of mobile robots from algorithm development to physical application. Summary of the Invention

[0005] The purpose of this application is to provide a mobile robot simulation training platform and equipment to solve the technical problems mentioned in the background art.

[0006] To achieve the above objectives, this application discloses the following technical solution: a mobile robot simulation training platform and equipment, comprising: The hardware layer is used to construct a three-dimensional simulation environment, which includes a mobile robot model capable of carrying task payloads, a task scenario model, and environmental variables. The algorithm layer integrates path planning algorithms, target recognition algorithms, and motion control algorithms; The interaction layer provides a parameter adjustment interface; The simulation training platform and equipment are configured to perform the following steps: Robot and Environment Initialization: Import the mobile robot model and the task scenario model from the model library, and set the sensor parameters and environmental variables; Target detection and path planning: Identify the location of the task target using at least one environmental perception sensor, and then call the path planning algorithm to plan the optimal path; Motion control and obstacle avoidance: The motion control algorithm is used to drive the robot to move and avoid obstacles in real time based on sensor data; Performance evaluation and parameter optimization: Record task success rate and path length metrics, and automatically or manually adjust the parameters of the algorithm layer based on these metrics.

[0007] Optionally, the path planning algorithm includes: A dynamic window is constructed, which is jointly determined by speed limit constraints, dynamic constraints, and security constraints. Trajectory simulation: For each velocity pair in the dynamic window, predict the trajectory in future time intervals based on the robot's kinematics model; Cost function calculation: For each simulated trajectory, a cost function is calculated, which includes a target orientation term, an obstacle avoidance term, and a velocity preference term. Choose the speed pair with the lowest cost as the control output.

[0008] Optionally, the construction of the dynamic window includes: Speed ​​constraint: The robot's speed is limited by a minimum linear velocity. Maximum linear velocity Minimum angular velocity and maximum angular velocity velocity set for: ; Dynamic constraints: at the current speed Below, based on maximum linear acceleration and maximum angular acceleration During the control cycle Within reach of speed range for: ; Safety constraints: Calculate the maximum speed at which safe braking is possible. ,in safe speed set to distance to the nearest obstacle for: ; The final dynamic window is .

[0009] Optionally, the expression for the prediction result of the trajectory simulation is: in, This represents the robot's current position. The robot's current orientation angle. To control the cycle, For the current linear velocity being evaluated, This represents the angular velocity currently being evaluated.

[0010] Optionally, the cost function is: in, , The Euclidean distance from the endpoint of the trajectory to the target point; , The minimum distance to the nearest obstacle on the trajectory; , Linear velocity; , and These are the preset weighting coefficients.

[0011] Optionally, the target recognition algorithm includes: Bounding box prediction: Each candidate box outputs a 4D vector. The coordinates can be converted to actual coordinates using the following formula: ,in, The coordinates of the center of the predicted bounding box. To predict the width and height of the bounding box, The coordinates of the top left corner of the grid. For the corresponding width and height of the anchor frame, For the Sigmoid function; Confidence prediction: Confidence score for each prediction box This indicates the probability that the box contains the target; Classification prediction: For each bounding box, predict C classification scores for each category. ,in The original output for each category; Final output: Final score for each predicted box .

[0012] Optionally, the loss function of the target recognition algorithm is: Total loss ,in, For bounding box regression loss, For confidence loss, For classifying losses, , and These are preset weighting coefficients; Bounding box regression loss for: ,in, The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box. This represents the Euclidean distance between the center of the predicted bounding box and the center of the ground truth bounding box. The minimum diagonal length of the bounding box. This is an indicator of aspect ratio consistency. for The weighted terms; Confidence loss for: ,in, This indicates that a target exists. Indicates no target; Classification loss for: ,in, For category The true label.

[0013] Optionally, the motion control algorithm includes: Establish a vehicle coordinate system with the origin at the center of the vehicle chassis, the x-axis pointing forward, and the y-axis pointing to the left. The angle of counterclockwise rotation around the z-axis; Obtain the velocity vector output by the upper-level control ,in For forward speed, The lateral velocity, This is the command for angular velocity around the z-axis; The inverse kinematics model transforms the velocity vector output by the upper-level control into the desired angular velocity vector of the four omnidirectional wheels. The conversion formula is: ,in, Where is the radius of the wheel. The inverse kinematics matrix: , The distance from the center of the vehicle body to the centers of the front and rear wheels. The distance from the center of the vehicle body to the center of the left and right wheels; The forward kinematic model is based on the actual measured wheel speed. Thrust reverser speed: ,in, For positive kinematics: ; PID control: Desired wheel speed for each wheel Compared with the actual measured wheel speed residual The motor drive signal for each wheel ,in, This is the proportionality coefficient. Integral coefficient, Differential coefficients.

[0014] Optionally, the robot model includes: A camera, mounted on top of the robot, is used to identify the location of the task target in real time; LiDAR, mounted on the robot chassis, is used for obstacle avoidance and dynamic obstacle detection; IMU sensors are used to measure the robot's attitude angles.

[0015] Optionally, the environmental variables include light intensity, ground friction coefficient, ground adhesion coefficient, and target distribution density.

[0016] Beneficial Effects: The mobile robot simulation training platform and equipment of this application overcome the inherent defects of traditional physical testing and simplified simulation by constructing a simulation training platform integrating hardware, algorithm, and interaction layers, and executing a closed-loop training process based on real parameter initialization. This technology can simulate robot dynamics, sensor noise, and complex environmental interactions with high fidelity in a virtual environment, reducing the cost and improving the efficiency of testing and optimization of path planning, target recognition, and motion control algorithms. Simultaneously, it ensures the reliable transfer and deployment of simulation training results to real robot systems, providing a reliable infrastructure for the efficient development of mobile robots. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a structural block diagram of the mobile robot simulation training platform and equipment provided in the embodiments of this application; Figure 2 A flowchart illustrating the simulation training process of the mobile robot simulation training platform and equipment provided in this application embodiment. Detailed Implementation

[0019] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] In this document, the term "comprising" is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0021] This embodiment provides a method such as Figure 1 The mobile robot simulation training platform and equipment shown can be used in various scenarios, including but not limited to: site cleaning, ball retrieval (tennis balls, pickles, etc.), etc. Specifically, the mobile robot simulation training platform and equipment in this embodiment includes: Hardware Layer: A 3D simulation environment is built in simulation software (such as Gazebo or Unity+ROS). This environment includes a mobile robot model capable of carrying task payloads (such as any matching ball-collecting module, cleaning module, etc.), a task scenario model (such as a corresponding tennis court, peak court, etc.), and environmental variables. In this embodiment, the robot model can be a differential-drive mobile platform (serving as the carrier of the task payload). This platform includes a universal mobile chassis and can carry different end effectors as task payloads (e.g., a mechanical claw for collecting balls). The model includes: a camera mounted on the top of the robot for real-time identification of the task target (e.g., a tennis ball); a lidar mounted on the robot (vehicle) chassis for obstacle avoidance and dynamic obstacle detection; and an IMU sensor for measuring the robot's attitude angles. Environmental variables include adjustable light intensity (affecting image recognition), ground friction coefficient (affecting wheel slippage), ground adhesion coefficient (used to simulate different ground conditions such as dry / wet, affecting wheel slippage and cleaning effectiveness), and task target distribution density (the number and location of randomly distributed balls / trash).

[0022] Algorithm layer: Integrates path planning algorithm, target recognition algorithm, and motion control algorithm. In this embodiment, the path planning algorithm is specifically the Dynamic Window (DAW) method, the target recognition algorithm is YOLOv5, and the motion control algorithm is PID-based wheel speed control.

[0023] The interaction layer provides a parameter adjustment interface (GUI), allowing users to directly adjust the weight coefficients (μ, η, ξ) of the DWA algorithm, the confidence threshold of YOLOv5, PID parameters, and various environmental variables (such as ground adhesion coefficient) via sliders or input boxes.

[0024] Among them, such as Figure 2 As shown, the simulation training platform and equipment are configured to perform the following steps: Robot and Environment Initialization: Import robot models, task scenario models, and selected task payload models (such as ball pickers) from the model library, and set sensor parameters and environmental variables (such as camera intrinsic parameters, lidar detection range, IMU noise parameters, and ground adhesion coefficient). Target detection and path planning: The camera identifies the location of the target (such as a tennis ball), and the path planning algorithm is called to plan the optimal path. That is, the camera captures RGB images, inputs them into the YOLOv5 network to obtain the target pixel coordinates, and transforms them into the position in the world coordinate system, which is used as the target point of the DWA algorithm. Motion control and obstacle avoidance: Motion control algorithms drive robot movement, and LiDAR enables real-time obstacle avoidance. Specifically, the DWA algorithm calculates the optimal speed command based on the robot's current pose, target point, and a real-time obstacle cost map generated by the LiDAR. This command is then converted into the desired wheel speed through kinematic calculations, and a PID controller drives the motors to ensure the actual wheel speed tracks the desired value. Performance evaluation and parameter optimization: Record task success rate (e.g., number of balls picked up) and path length metrics, and based on these metrics, return to the interaction layer to adjust algorithm parameters or environment variables for iterative optimization.

[0025] Based on the above, this embodiment overcomes the inherent shortcomings of traditional physical testing and simplified simulation by constructing a general simulation training platform integrating hardware, algorithm, and interaction layers, and executing a closed-loop training process based on real parameter initialization. This technology can simulate the dynamic characteristics, sensor noise, and complex environmental interactions of a multi-functional mobile robot with high fidelity in a virtual environment, reducing the cost and improving the efficiency of core algorithm testing and optimization. Simultaneously, it ensures the reliable transfer and deployment of simulation training results to real robot systems, providing a reliable infrastructure for the efficient development and functional expansion of mobile robots.

[0026] In one implementation, the path planning algorithm includes: Construct a dynamic window, which is jointly determined by speed limit constraints, dynamic constraints, and safety constraints; Trajectory simulation: For each velocity pair in the dynamic window, predict the trajectory in future time intervals based on the robot's kinematics model; Cost function calculation: For each simulated trajectory, the cost function is calculated, which includes a target orientation term, an obstacle avoidance term, and a velocity preference term. Choose the speed pair with the lowest cost as the control output.

[0027] A feasible approach is to construct a dynamic window specifically including: Speed ​​constraint: The robot's speed is limited by a minimum linear velocity. Maximum linear velocity Minimum angular velocity and maximum angular velocity The robot speed is set according to the performance of the robot motor, and the speed set is... for: ; Dynamic constraints: The robot's real-time speed at the current moment, i.e., its current speed. Below, based on the maximum linear acceleration and maximum angular acceleration During the control cycle Within reach of speed range for: ; Safety constraints: Calculate the maximum speed at which safe braking is possible. ,in The safe speed set is the distance to the nearest obstacle (i.e., the nearest obstacle distance extracted from the current scan data of the lidar). for: ; The final dynamic window is .

[0028] Feasible approach is to use a differential kinematics model for trajectory simulation, predicting the trajectory within a future prediction time (usually set to 1-3 seconds). This involves using the differential model to predict a series of discrete pose points within the future prediction time for each set of (v, ω) within the window. The expression for the prediction result is as follows: in, This represents the robot's current position. The robot's current orientation angle. The control cycle (consistent with the simulation step size and the cycle of the underlying controller, for example, 0.1 seconds). For the current linear velocity being evaluated, This represents the angular velocity currently being evaluated.

[0029] A feasible approach is to use the following cost function: in, It is used to encourage the robot to move toward the target point. The Euclidean distance from the endpoint of the trajectory to the target point; Used to ensure that robots maintain a safe distance from obstacles. The minimum distance to the nearest obstacle on the trajectory; This is used to encourage robots to operate at higher speeds and improve efficiency. Linear velocity; , and To utilize the preset weighting coefficients in the interaction layer, for example, when security is a greater priority, the weighting can be increased. .

[0030] Based on the above, the DWA algorithm strictly considers the robot's kinematic constraints and real-time obstacle avoidance requirements through a dynamic window mechanism. The multi-objective balance of the cost function enables the planned trajectory to have safety, target orientation, and operational efficiency.

[0031] In one implementation, the target recognition algorithm employs the YOLOv5 algorithm, using a YOLOv5 model pre-trained on a dataset of approximately 10,000 target images (including simulated and real images). During the simulation, each frame captured by the camera is fed into the trained model for forward inference. The target recognition algorithm (i.e., the YOLOv5 algorithm) includes: Bounding box prediction: Each candidate box outputs a 4D vector. The coordinates can be converted to actual coordinates using the following formula: ,in, The coordinates of the center of the predicted bounding box. To predict the width and height of the bounding box, The coordinates of the top left corner of the grid. The width and height of the anchor boxes are pre-defined for the dataset through clustering. This is the Sigmoid function, which is used to constrain the network output within the range of (0,1) to ensure that the offset is reasonable; Confidence prediction: Confidence score for each prediction box This indicates the probability that the box contains the target; Classification prediction: For each bounding box, predict C classification scores for each category. ,in The original output for each category; Final output: Final score for each predicted box .

[0032] A feasible approach is to use the following loss function for the target recognition algorithm: Total loss ,in, For bounding box regression loss, For confidence loss, For classifying losses, , and These are preset weighting coefficients; Bounding box regression loss Using CIoU loss, it is: ,in, The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box. This represents the Euclidean distance between the center of the predicted bounding box and the center of the ground truth bounding box. The minimum diagonal length of the bounding box. This is an indicator of aspect ratio consistency. for The weighted terms; Confidence loss Using binary cross-entropy loss, it is: ,in, This indicates that a target exists. Indicates no target; Classification loss Using binary cross-entropy loss, it is: ,in, For category The true label.

[0033] Based on the above, the target recognition algorithm in this embodiment provides high-speed and high-precision target detection capabilities, ensuring that the robot can lock the target position in real time. Combined with the loss function, especially the model trained in the simulation environment, it guarantees detection accuracy and strong robustness to virtual scenes.

[0034] In one implementation, the motion control algorithm includes: Establish a vehicle coordinate system with the origin at the center of the vehicle chassis, the x-axis pointing forward, and the y-axis pointing to the left. The angle of counterclockwise rotation around the z-axis; Obtain the velocity vector output by the upper-level control ,in For forward speed, The lateral velocity, This is the command for angular velocity around the z-axis; The inverse kinematics model transforms the velocity vector output by the upper-level control into the desired angular velocity vector of the four omnidirectional wheels. Each element in the expected angular velocity vector This refers to the expected wheel speeds of the corresponding 1st, 2nd, 3rd, and 4th omnidirectional wheels, where... Where is the radius of the wheel. The inverse kinematics matrix: , The distance from the center of the vehicle body to the centers of the front and rear wheels. The distance from the center of the vehicle body to the centers of the left and right wheels. and Precise measurements and input are required during robot modeling; The forward kinematic model is based on the actual measured wheel speed. Thrust reverser speed: The vehicle speed is used to reflect the robot's actual motion state, where, For positive kinematics: The positive kinematics matrix is ​​the pseudo-inverse of the inverse kinematics matrix. It is used to infer the actual vehicle speed from the wheel speed feedback measured by the encoder, and is used for state updates or closed-loop control. PID control: desired wheel speed Compared with the actual measured wheel speed residual For each wheel, the actual measured wheel speed comes from the encoder mounted on each drive motor. The motor drive signal for each wheel... ,in, This is the proportionality coefficient. Integral coefficient, Differential coefficients.

[0035] Based on this, this embodiment achieves command decomposition through inverse kinematics, achieves precise tracking at the underlying level through PID control, and uses measured wheel speeds to estimate the vehicle speed in real time through forward kinematics, forming a complete closed loop of command-execution-feedback, ensuring the accuracy of motion control and the reliability of state perception.

[0036] Furthermore, it should be noted that although the above motion control algorithm is illustrated using a four-wheel omnidirectional mobile platform as an example, its control framework and methods are equally applicable to mobile robots with other numbers of wheels and other drive methods, such as two-wheel differential drive, three-wheel omnidirectional, and six-wheel MacPherson strut configurations. For different configurations, only the inverse kinematics matrix needs to be adjusted according to its specific kinematic model. Sum of normal kinematics matrix The structure and parameters remain the same, while the overall algorithm flow (coordinate establishment, velocity command reception, kinematic transformation, PID control, and velocity feedback) remains unchanged. Those skilled in the art can derive the corresponding kinematic matrices based on the actual robot configuration and apply them to this platform.

[0037] In summary, the mobile robot simulation training platform and equipment of this embodiment begin its workflow at the hardware layer by initializing the robot model and virtual task environment using real parameters. Subsequently, the algorithm layer initiates collaborative work: the target recognition algorithm continuously processes camera images to accurately locate the task target, providing target points for path planning; the path planning algorithm integrates the target position, real-time obstacle information from the LiDAR, and the robot's current speed and motion constraints to generate a safe and efficient collision-free trajectory online; the motion control algorithm decomposes the speed command into the desired rotational speed of each wheel through inverse kinematics and uses a PID controller to drive the motor for precise tracking. Simultaneously, the interaction layer allows developers to adjust algorithm parameters and environmental conditions in real time. Finally, by collecting data, the entire system is driven to undergo iterative optimization. This embodiment, through high-fidelity modeling across the entire chain and real-time interaction between the algorithm and the environment, achieves the training and verification of the robot's perception-decision-control full-stack capabilities in virtual space, effectively bridging the gap between simulation and real-world applications.

[0038] In the embodiments provided in this application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor may be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments may be performed by a computer program instructing the associated hardware. During implementation, the program may be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible to a computer. Computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0039] Finally, it should be noted that the above description is only a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A mobile robot simulation training platform and equipment, characterized in that, include: The hardware layer is used to construct a three-dimensional simulation environment, which includes a mobile robot model capable of carrying task payloads, a task scenario model, and environmental variables. The algorithm layer integrates path planning algorithms, target recognition algorithms, and motion control algorithms; The interaction layer provides a parameter adjustment interface; The simulation training platform and equipment are configured to perform the following steps: Robot and Environment Initialization: Import the mobile robot model and the task scenario model from the model library, and set the sensor parameters and environmental variables; Target detection and path planning: Identify the location of the task target using at least one environmental perception sensor, and then call the path planning algorithm to plan the optimal path; Motion control and obstacle avoidance: The motion control algorithm is used to drive the robot to move and avoid obstacles in real time based on sensor data; Performance evaluation and parameter optimization: Record task success rate and path length metrics, and automatically or manually adjust the parameters of the algorithm layer based on these metrics.

2. The mobile robot simulation training platform and equipment according to claim 1, characterized in that, The path planning algorithm includes: A dynamic window is constructed, which is jointly determined by speed limit constraints, dynamic constraints, and security constraints. Trajectory simulation: For each velocity pair in the dynamic window, predict the trajectory in future time intervals based on the robot's kinematics model; Cost function calculation: For each simulated trajectory, a cost function is calculated, which includes a target orientation term, an obstacle avoidance term, and a velocity preference term. Choose the speed pair with the lowest cost as the control output.

3. The mobile robot simulation training platform and equipment according to claim 2, characterized in that, The construction of the dynamic window specifically includes: Speed ​​constraint: The robot's speed is limited by a minimum linear velocity. Maximum linear velocity Minimum angular velocity and maximum angular velocity velocity set for: ; Dynamic constraints: at the current speed Below, based on maximum linear acceleration and maximum angular acceleration During the control cycle Within reach of speed range for: ; Safety constraints: Calculate the maximum speed at which safe braking is possible. ,in safe speed set to distance to the nearest obstacle for: ; The final dynamic window is .

4. The mobile robot simulation training platform and equipment according to claim 2, characterized in that, The expression for the prediction result of the trajectory simulation is: in, This represents the robot's current position. The robot's current orientation angle. To control the cycle, For the current linear velocity being evaluated, This represents the angular velocity currently being evaluated.

5. The mobile robot simulation training platform and equipment according to claim 2, characterized in that, The cost function is: in, , The Euclidean distance from the endpoint of the trajectory to the target point; , The minimum distance to the nearest obstacle on the trajectory; , Linear velocity; , and These are the preset weighting coefficients.

6. The mobile robot simulation training platform and equipment according to claim 1, characterized in that, The target recognition algorithm includes: Bounding box prediction: Each candidate box outputs a 4D vector. The coordinates can be converted to actual coordinates using the following formula: ,in, The coordinates of the center of the predicted bounding box. To predict the width and height of the bounding box, The coordinates of the top left corner of the grid. For the corresponding width and height of the anchor frame, For the Sigmoid function; Confidence prediction: Confidence score for each prediction box This indicates the probability that the box contains the target; Classification prediction: For each bounding box, predict C classification scores for each category. ,in The original output for each category; Final output: Final score for each predicted box .

7. The mobile robot simulation training platform and equipment according to claim 6, characterized in that, The loss function of the target recognition algorithm is: Total loss ,in, For bounding box regression loss, For confidence loss, For classifying losses, , and These are preset weighting coefficients; Bounding box regression loss for: ,in, The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box. This represents the Euclidean distance between the center of the predicted bounding box and the center of the ground truth bounding box. The minimum diagonal length of the bounding box. This is an indicator of aspect ratio consistency. for The weighted terms; Confidence loss for: ,in, This indicates that a target exists. Indicates no target; Classification loss for: ,in, For category The true label.

8. The mobile robot simulation training platform and equipment according to claim 1, characterized in that, The motion control algorithm includes: Establish a vehicle coordinate system with the origin at the center of the vehicle chassis, the x-axis pointing forward, and the y-axis pointing to the left. The angle of counterclockwise rotation around the z-axis; Obtain the velocity vector output by the upper-level control ,in For forward speed, The lateral velocity, This is the command for angular velocity around the z-axis; The inverse kinematics model transforms the velocity vector output by the upper-level control into the desired angular velocity vector of the four omnidirectional wheels. The conversion formula is: ,in, Where is the radius of the wheel. The inverse kinematics matrix: , The distance from the center of the vehicle body to the centers of the front and rear wheels. The distance from the center of the vehicle body to the center of the left and right wheels; The forward kinematic model is based on the actual measured wheel speed. Thrust reverser speed: ,in, For positive kinematics: ; PID control: Desired wheel speed for each wheel Compared with the actual measured wheel speed residual The motor drive signal for each wheel ,in, This is the proportionality coefficient. Integral coefficient, Differential coefficients.

9. The mobile robot simulation training platform and equipment according to claim 1, characterized in that, The robot model includes: A camera, mounted on top of the robot, is used to identify the location of the task target in real time; LiDAR, mounted on the robot chassis, is used for obstacle avoidance and dynamic obstacle detection; IMU sensors are used to measure the robot's attitude angles.

10. The mobile robot simulation training platform and equipment according to claim 1, characterized in that, The environmental variables include light intensity, ground friction coefficient, ground adhesion coefficient, and target distribution density.

Images