Industrial scene-oriented wheel-foot humanoid robot path planning and navigation method

Abstract

The invention relates to the technical field of robot navigation, and discloses an industrial scene-oriented wheel-foot humanoid robot path planning and navigation method, which comprises the following steps of S1, acquiring environment information and self state information of a robot; s2, performing time sequence alignment and coordinate transformation on the acquired environment information and self state information; s3, recognizing a dynamic obstacle according to the time sequence to obtain a fusion environment map; s4, performing dynamic obstacle avoidance and path re-planning based on a local dynamic planning algorithm; s5, whether a wheel type mode or a foot type mode is started or not is judged according to the environmental information terrain flatness and the ground gradient parameters; and S6, feeding back according to errors and energy consumption in the navigation execution process. According to the invention, by adopting an optimization strategy combining dynamic path re-planning and reinforcement learning, the flexibility of path planning and the capability of coping with a complex environment are improved, and the energy consumption and the obstacle avoidance distance are effectively reduced, so that the robot can stably operate in a wider scene.

Description

Technical Field

[0001] This invention relates to the field of robot navigation technology, and in particular to a path planning and navigation method for wheeled humanoid robots for industrial scenarios. Background Technology

[0002] "Industrial settings" typically refer to environments and locations related to industrial production, manufacturing, and automation. This includes various equipment, facilities, machinery, sensors, and workflows, all of which are closely related to activities such as large-scale production, logistics, equipment maintenance, and quality control.

[0003] Wheel-legged humanoid robots combine wheeled and legged locomotion mechanisms. They possess the high-speed, energy-efficient mobility of wheeled robots and the flexibility and adaptability to complex terrain of legged robots. Wheel-legged humanoid robots typically need to operate in the same type of industrial task for extended periods. Through reinforcement learning and historical data analysis, path planning and navigation methods can continuously adjust parameters, enabling the robot to gradually optimize its path, reduce energy consumption, and improve work efficiency and stability in repetitive tasks.

[0004] In related technologies, robot path planning mostly relies on fixed parameters and static maps, lacking the ability to adapt to dynamic changes in the environment. When moving obstacles, terrain changes, or task objectives are adjusted in the external environment, the system cannot update the planning results in time, which can easily lead to path failure, untimely obstacle avoidance, or excessive energy consumption, thus significantly reducing the stability of robot operation and the efficiency of task completion in complex industrial scenarios. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a path planning and navigation method for wheeled humanoid robots in industrial scenarios. It aims to improve the lack of adaptive ability to dynamic changes in the environment. When there are moving obstacles, terrain changes or task objectives in the external environment, the system cannot update the planning results in time, which can easily lead to path failure, untimely obstacle avoidance or excessive energy consumption.

[0006] In a first aspect, the present invention provides the following technical solution: a path planning and navigation method for wheeled humanoid robots in industrial scenarios, comprising the following steps: S1. By installing lidar, vision camera, inertial measurement unit and odometry sensor on the wheel-legged humanoid robot, the robot's environmental information and its own status information can be obtained; S2. Then, the collected environmental information and self-state information are time-series aligned and coordinate transformed, and the robot's pose estimation is obtained through extended Kalman filter or factor graph optimization algorithm. S3. Generate a geometric map based on LiDAR data, generate a semantic map based on visual information, and identify dynamic obstacles based on time series to obtain a fused environment map; S4. Use the global path planning algorithm to obtain the initial path from the starting point to the target point, and use the local dynamic planning algorithm to perform dynamic obstacle avoidance and path replanning. S5. Based on environmental information such as terrain flatness and ground gradient parameters, determine whether to enable wheeled or legged mode, and control the robot to perform the corresponding movements. S6. Drive the robot to move based on control commands, and adjust the path tracking error feedback according to the real-time positioning results. At the same time, use reinforcement learning algorithm to dynamically adjust the path planning parameters based on the error and energy consumption feedback during the navigation process.

[0007] Preferably, in step S2, the robot state estimate is obtained by weighted updating of the predicted state and the observed state; The predicted state is determined by the control input, while the observed state is determined by the output of the multimodal sensor. The fusion algorithm employs either the extended Kalman filter algorithm or the nonlinear factor graph optimization algorithm.

[0008] Preferably, step S3 includes the following steps: S301. Establish a multi-layer map structure that includes a geometric layer, a semantic layer, a dynamic layer, and a frequency domain layer; S302. Merge the maps of each layer according to preset weight parameters to obtain a unified environment model; The fusion process is weighted based on geometric feature consistency, semantic matching degree, and obstacle movement frequency characteristics.

[0009] Preferably, step S4 includes the following steps: S401. Use a heuristic search algorithm to search the global map to obtain the initial path; S402. Smooth the initial path. S403. During the local planning process, linear velocity and angular velocity control quantities are generated based on the dynamic window method combined with reinforcement learning strategy. S404. Detect obstacle positions using information acquired by the acquisition module, and update the local path according to the obstacle avoidance cost function; The cost function comprehensively considers four indicators: path orientation, distance to obstacles, speed constraints, and risk factors.

[0010] Preferably, step S5 includes the following steps: S501. Calculate the standard deviation of ground height and terrain gradient through the terrain sensing module; S502. When the standard deviation of ground height is less than the first threshold and the terrain gradient is less than the second threshold, select the wheeled movement mode. S503. When the above conditions are not met, select the foot-walking mode; S504. During mode switching, a model predictive control algorithm is used to perform rolling optimization of the control input to achieve smooth transition and stable path tracking.

[0011] Preferably, step S6 includes the following steps: S601. Perform attitude adjustment based on the speed and attitude commands output by the path planning; S602. Real-time monitoring of robot positioning error and attitude deviation through feedback module; S603. When the positioning error exceeds the set threshold, local path replanning is triggered. S604: Control signals are fed back to the motion controller via the communication bus for navigation control.

[0012] Preferably, step S6 further includes the following step: An adaptive parameter update strategy based on reinforcement learning is adopted, which comprehensively considers four indicators: path smoothness, energy consumption, obstacle avoidance distance and task completion time through the reward function; The search heuristic parameters and control weights in path planning are automatically adjusted based on historical task data, which is used to learn and adjust path planning based on historical task data.

[0013] Secondly, this invention provides the following technical solution: a path planning and navigation system for wheeled humanoid robots in industrial scenarios, the system comprising: The sensing module is used to collect environmental information and robot status data; The positioning module is used to perform spatiotemporal fusion of multi-source sensor data and output robot pose estimation. The map building module is used to create multi-layered maps (geometric, semantic, dynamic, and frequency domain) and perform weighted fusion. The path module is used to perform global path planning and local dynamic obstacle avoidance planning; The switching module is used to adaptively select wheeled or legged movement modes based on terrain features; The navigation feedback module is used to execute control commands, detect navigation deviations and perform feedback corrections, and dynamically adjust path planning and control parameters based on task execution results, learning and adjusting path planning.

[0014] As a further technical feature of the above: The sensing module includes a lidar, a depth camera, an inertial measurement unit, and an odometer sensor, and is aligned with a unified clock through a time synchronization module for synchronous acquisition of multi-source data. The switching module is equipped with a terrain feature analysis unit and a model prediction control unit. The terrain feature analysis unit is used to calculate the ground height difference and slope information, and the model prediction control unit is used to generate the optimal control command in wheeled or footed mode.

[0015] Thirdly, the invention provides the following technical solution: a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned path planning and navigation method for wheeled humanoid robots in industrial scenarios.

[0016] Fourthly, the present invention provides the following technical solution: a readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned path planning and navigation method for wheeled humanoid robots in industrial scenarios.

[0017] The present invention has the following beneficial effects: 1. In this invention, by adopting an optimization strategy that combines dynamic path replanning and reinforcement learning, the robot can automatically adjust its path planning parameters according to real-time environmental changes in complex industrial scenarios, thereby improving the robot's autonomous adaptability. This not only enhances the flexibility of path planning and its ability to cope with complex environments, but also effectively reduces energy consumption and obstacle avoidance distance, enabling the robot to operate stably in a wider range of scenarios.

[0018] 2. In this invention, by introducing adaptive wheel-foot mode switching technology, the most suitable motion mode is automatically selected according to the ground gradient and terrain flatness. Unlike the common fixed mode selection method, it can dynamically adjust according to the actual terrain, so that the robot can move in the optimal way in different environments, avoiding control instability caused by terrain incompatibility.

[0019] 3. In this invention, the adaptive parameter update mechanism driven by reinforcement learning adjusts the path planning and control weights based on historical task data, continuously optimizing the robot's performance in various tasks. Compared with the limitations of relying solely on preset parameters, it can continuously optimize parameters by learning from historical data, thereby improving the robot's performance and efficiency when performing tasks over a long period of time. Attached Figure Description

[0020] Figure 1 This is a flowchart of the path planning and navigation method for wheeled humanoid robots in industrial scenarios proposed in this invention. Figure 2 This is a flowchart illustrating the map construction process of the path planning and navigation method for wheeled humanoid robots in industrial scenarios proposed in this invention. Figure 3This is a flowchart illustrating the path planning process of the wheeled humanoid robot path planning and navigation method for industrial scenarios proposed in this invention. Figure 4 This is a flowchart showing the mode switching process of the path planning and navigation method for wheeled humanoid robots in industrial scenarios proposed in this invention. Figure 5 This is a navigation feedback flowchart for the path planning and navigation method for wheeled humanoid robots in industrial scenarios proposed in this invention. Figure 6 This is a system architecture diagram of the path planning and navigation system for wheeled humanoid robots in industrial scenarios proposed in this invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1 Reference Figures 1-5 In the first embodiment of the present invention, the present invention provides a path planning and navigation method for wheeled humanoid robots in industrial scenarios, including the following steps: S1. By installing lidar, vision camera, inertial measurement unit and odometry sensor on the wheel-legged humanoid robot, the robot's environmental information and its own status information can be obtained; S2. Then, the collected environmental information and self-state information are time-series aligned and coordinate transformed, and the robot's pose estimation is obtained through extended Kalman filter or factor graph optimization algorithm. S3. Generate a geometric map based on LiDAR data, generate a semantic map based on visual information, and identify dynamic obstacles based on time series to obtain a fused environment map; S4. Use the global path planning algorithm to obtain the initial path from the starting point to the target point, and use the local dynamic planning algorithm to perform dynamic obstacle avoidance and path replanning. S5. Based on environmental information such as terrain flatness and ground gradient parameters, determine whether to enable wheeled or legged mode, and control the robot to perform the corresponding movements. S6. Drive the robot to move based on control commands, and adjust the path tracking error feedback according to the real-time positioning results. At the same time, use reinforcement learning algorithm to dynamically adjust the path planning parameters based on the error and energy consumption feedback during the navigation process.

[0023] In S2, the robot state estimate is obtained by weighted updating of the predicted state and the observed state; The predicted state is determined by the control input, while the observed state is determined by the output of the multimodal sensor. The fusion algorithm employs either the extended Kalman filter algorithm or the nonlinear factor graph optimization algorithm.

[0024] Specifically, the predicted state and the observed state are updated using a weighted average. The predicted state is mainly calculated based on the control input, while the observed state is determined by multiple modal sensors, including vision, LiDAR, inertial measurement unit, and foot force sensor. By continuously performing prediction and update in the time dimension, the robot's state remains observable and controllable in a dynamic environment. The predicted state is generated based on kinematic and dynamic models and is predicted using the following formula: Suppose the robot is at time... The state vector is: ; in, Indicates the coordinates of the centroid position. For Euler angle orientation, For the linear velocity component, the control input vector is defined as follows: ; The prediction model is based on nonlinear state transition equations, corresponding to the measurements of acceleration and angular velocity, respectively. ; in, The process noise term is assumed to follow a zero-mean Gaussian distribution. , The motion function is nonlinear, determined by the dynamics of the robot chassis and the cooperative constraints of the wheels and legs. The observed state is output in real time by multimodal sensors, and after being unified with timestamps by the synchronization module, it forms an observation vector. ; in, Represents the observation model mapping function. To observe noise, satisfy In actual operation, the visual sensor provides the pose change, the IMU provides high-frequency inertial data, the lidar provides obstacle depth and feature point cloud, and the foot sensor provides feedback on contact status and support force distribution; State fusion algorithms can employ extended Kalman filtering or nonlinear factor graph optimization. When using EKF, the system first performs first-order linearization on the nonlinear state equations, and the update process is as follows: Predicting the covariance matrix: ; in, The state transition Jacobian matrix is ​​then used for observation updates: ; ; ; in, To observe the Jacobian matrix, The Kalman gain matrix is ​​used to achieve the optimal state estimate under noise and error disturbances during runtime. When the system operates in highly dynamic scenarios or in environments with irregular terrain, and the nonlinearity between prediction and observation is significant, the linearization error of EKF can lead to a decrease in convergence. In such cases, a nonlinear factor graph optimization algorithm can be used, which transforms state estimation into an optimization problem that minimizes the posterior error. ; in, Let be the residual function of each observed factor. The L2 norm is represented by the weighted covariance with noise. By constructing the time series graph results, the IMU integral constraints, visual odometry constraints, laser feature constraints, and foot contact constraints are uniformly encoded as graph nodes and factors, and the Gauss-Newton or Levenberg-Marquardt algorithms are used for iterative solution. After the fusion results are output, the system can obtain a comprehensive state estimate in real time, including attitude, velocity, position and their covariance information, providing input for the next stage of path tracking control. Due to the weighted update of multi-source information and the introduction of nonlinear constraints, the system can still maintain high positioning stability and navigation accuracy in complex industrial scenarios such as visual occlusion, vibration interference or uneven ground. In addition, as a further technical feature, a learning-based dynamic model compensation module is introduced in the prediction stage to correct the deviation of traditional dynamic equations. By using deep neural networks to model the mapping relationship between historical control inputs and state changes, the system's adaptability to unknown friction or ground damping is improved.

[0025] S3 includes the following steps: S301. Establish a multi-layer map structure that includes a geometric layer, a semantic layer, a dynamic layer, and a frequency domain layer; S302. Merge the maps of each layer according to preset weight parameters to obtain a unified environment model; The fusion process is weighted based on geometric feature consistency, semantic matching degree, and obstacle movement frequency characteristics.

[0026] Specifically, the geometric layer is generated by LiDAR and vision sensors to describe the spatial structure. Let the point cloud be... Occupy grid map ; The semantic layer extracts scene categories through a semantic segmentation network and generates... ; The dynamic layer describes the target motion intensity and is defined as follows: ; The frequency domain layer processes the time series data from the dynamic layer. Performing a fast Fourier transform yields: Used to identify periodic motion characteristics.

[0027] The unified environment model after integration is represented as follows: ; in, For the weight parameters, satisfying Weights are based on geometric feature consistency. Semantic matching degree Frequency domain characteristics Dynamic adjustment: The final fused model possesses both geometric accuracy and dynamic adaptability, and can be used for path planning, obstacle avoidance, and scene recognition. When scene changes are detected, the system can automatically adjust the weights of each layer.

[0028] S4 includes the following steps: S401. Use a heuristic search algorithm to search the global map to obtain the initial path; S402. Smooth the initial path. S403. During the local planning process, linear velocity and angular velocity control quantities are generated based on the dynamic window method combined with reinforcement learning strategy. S404. Detect obstacle positions using information acquired by the acquisition module, and update the local path according to the obstacle avoidance cost function; The cost function comprehensively considers four indicators: path orientation, distance to obstacles, speed constraints, and risk factors.

[0029] Specifically, a heuristic search algorithm is first used to search the global map and generate an initial path from the starting point to the target point. During the path search process, the geometric structure and environmental features of the map are considered, and heuristic functions, including Manhattan distance or Euclidean distance, are used to evaluate the cost of each node. The generated initial path contains non-smooth turns or redundant path segments. To improve the smoothness and executability of the path, a curve smoothing algorithm, including B-spline or Bezier curve, is used to optimize the path, making the path more natural and reducing unnecessary vibrations and excessive rotations in the robot's movement. In actual operation, the robot is affected by dynamic obstacles, so local path planning is required based on the real-time environment. At this time, the dynamic window method and reinforcement learning strategy are combined to generate the robot's linear velocity in the current local environment. The dynamic window method limits the robot's speed and turning angle range to make the generated control quantity conform to the robot's current physical limitations, while reinforcement learning adjusts the control strategy based on environmental feedback to adapt to constantly changing obstacles. The robot monitors the position of obstacles ahead in real time through a data acquisition module. The system updates the local path based on the acquired obstacle information and calculates the corresponding obstacle avoidance cost function. The cost function comprehensively considers four main factors: Path orientation: The deviation between the path and the target direction, to prevent the robot from going astray; Obstacle distance: The shortest distance to an obstacle, ensuring you avoid it; Speed ​​constraint: Adjust the control quantity according to the robot's speed limit; Risk factors: Considering uncertainties in the environment, such as dynamic obstacles and external disturbances, to improve path safety, the cost function can be expressed as: Heading Error RiskFactor; in, These are the weighting coefficients. The minimum distance to the obstacle. The maximum speed of the robot; the Risk Factor represents the degree of uncertainty in a dynamic environment. By optimizing the cost function and updating the local path in real time, the robot can move smoothly and safely in complex environments. By generating a global path through heuristic algorithms, dynamically adjusting the local path, and combining real-time obstacle avoidance and control strategies, the system can efficiently cope with changing industrial scenarios, thus ensuring the safety and flexibility of path planning.

[0030] S5 includes the following steps: S501. Calculate the standard deviation of ground height and terrain gradient through the terrain sensing module; S502. When the standard deviation of ground height is less than the first threshold and the terrain gradient is less than the second threshold, select the wheeled movement mode. S503. When the above conditions are not met, select the foot-walking mode; S504. During mode switching, a model predictive control algorithm is used to perform rolling optimization of the control input to achieve smooth transition and stable path tracking.

[0031] Specifically, after completing path planning and local control, the system enters the motion mode adaptation stage, which automatically switches between wheeled and legged motion modes based on terrain features, thereby improving the robot's passability and stability in complex industrial environments. First, the standard deviation of ground height and terrain gradient are calculated using the terrain sensing module, which can be composed of LiDAR, depth camera, or force sensor. The standard deviation is calculated using the height data of the sampling points, as shown in the following formula: And estimate the terrain gradient based on the elevation difference between adjacent points. ; Among them, the height standard deviation It reflects the degree of undulation of the ground and is an important indicator for measuring the flatness of the terrain. :No. Ground elevation at each sampling point; The average height of all sampling points; The number of sampling points is such that, when the ground is flat, the height of each point is close to the average value. Smaller; if the ground is uneven, then If the value is relatively large, the system compares it with a first threshold to determine whether the wheel mode can be used; Topographic gradient : Indicates the slope or inclination of the ground. , These represent the rates of change of terrain height in the x and y directions, respectively. The larger the gradient, the steeper the terrain. When the gradient is less than a set threshold, it indicates that the slope is gentle and suitable for wheeled vehicles. Subsequently, when the system detects that the ground elevation standard deviation is less than the first threshold and the terrain gradient is less than the second threshold, it determines that the terrain is flat and selects the wheeled movement mode to obtain higher speed and energy efficiency. If either condition is not met, the system automatically switches to the legged walking mode to enhance terrain adaptability and stability. During the mode switching process, model predictive control is used, and the algorithm performs rolling optimization on the control input.

[0032] S6 includes the following steps: S601. Perform attitude adjustment based on the speed and attitude commands output by the path planning; S602. Real-time monitoring of robot positioning error and attitude deviation through feedback module; S603. When the positioning error exceeds the set threshold, local path replanning is triggered. S604: Control signals are fed back to the motion controller via the communication bus for navigation control.

[0033] Specifically, firstly, after completing environmental modeling, path planning, obstacle avoidance, and wheel-foot mode switching, the present invention enters the posture execution and closed-loop navigation stage, accurately converting the speed and posture commands output by the planning layer into underlying control signals, enabling the robot to perform stable posture adjustment and high-precision path tracking in complex industrial scenarios. Assuming the system operates with a fixed control cycle, real-time position and posture estimation is achieved through multi-source sensor fusion. The fused data includes IMU, odometry, lidar, and visual odometry information. Filtering algorithms are used to suppress sensor noise, thereby obtaining state estimation results. Secondly, the system continuously monitors positioning errors and attitude deviations. When the positioning error exceeds the set threshold (0.1–0.3 meters) or the attitude deviation exceeds the reasonable range (0.05–0.2 radians) and persists for several cycles, local path replanning is automatically triggered. The local planning module regenerates the short-term reference trajectory based on the updated environmental model and uses smooth interpolation to avoid control abrupt changes. If the current state is wheel-foot switching, the system automatically matches the corresponding dynamic model and constraints to ensure smooth transition and control continuity. Model predictive control is used for rolling optimization, comprehensively considering trajectory errors, control change rate, and obstacle avoidance constraints. It outputs control commands such as linear velocity, angular velocity, and gait parameters. The predicted step size is set to 10 to 30 steps to balance computational load and control real-time performance. The control signal is sent to the motion controller via a real-time communication bus. A timestamp and sequence number mechanism is used to synchronize the data. When a communication interruption or delay timeout is detected, a deceleration or stabilization strategy is automatically executed. The system security is improved through confirmation response and heartbeat mechanisms. Finally, the system also has a confidence-adaptive mechanism. When the filter confidence decreases or the localization covariance exceeds the threshold, the control parameters and trigger threshold will be dynamically adjusted. If necessary, the system will enter a speed limit or safety mode. If the system cannot return to normal for a long time, a rollback or manual intervention procedure will be triggered. Through closed-loop attitude control and adaptive replanning mechanism, robot attitude stabilization, path smoothing and safe obstacle avoidance can be achieved in dynamic industrial environments.

[0034] S6 also includes the following steps: An adaptive parameter update strategy based on reinforcement learning is adopted, which comprehensively considers four indicators: path smoothness, energy consumption, obstacle avoidance distance and task completion time through the reward function; The search heuristic parameters and control weights in path planning are automatically adjusted based on historical task data, which is used to learn and adjust path planning based on historical task data.

[0035] Specifically, firstly, an adaptive parameter update strategy based on reinforcement learning is introduced to dynamically optimize control performance under different task and environmental conditions. By constructing a comprehensive reward function, four key indicators, namely path smoothness, energy consumption, obstacle avoidance safety distance and task completion time, are evaluated in a weighted manner to guide the adaptive update of control parameters, so as to achieve a balance between safety, efficiency and stability in performance. Secondly, during long-term operation, the system continuously records historical task data, including feature information such as path trajectory, control input, energy consumption changes, and obstacle avoidance behavior. Based on reinforcement learning algorithms, the system performs offline or online analysis on this data to extract parameter optimization trends. The system then automatically adjusts the search heuristic parameters in the path planning stage and the weight allocation of the control layer, gradually converging to the optimal parameter combination that better matches the actual environmental characteristics and task requirements. Finally, through the aforementioned self-learning and parameter update mechanism, the system can maintain efficient path planning and stable control performance when facing different environmental complexities and changes in task objectives.

[0036] Example 2: Reference Figure 6 In a second embodiment of the present invention, the present invention provides a path planning and navigation system for wheeled humanoid robots in industrial scenarios, the system comprising: The sensing module is used to collect environmental information and robot status data; The positioning module is used to perform spatiotemporal fusion of multi-source sensor data and output robot pose estimation. The map building module is used to create multi-layered maps (geometric, semantic, dynamic, and frequency domain) and perform weighted fusion. The path module is used to perform global path planning and local dynamic obstacle avoidance planning; The switching module is used to adaptively select wheeled or legged movement modes based on terrain features; The navigation feedback module is used to execute control commands, detect navigation deviations and perform feedback corrections, and dynamically adjust path planning and control parameters based on task execution results, learning and adjusting path planning.

[0037] Specifically, firstly, the system includes a perception module, which is used to collect environmental information and robot's own state data in real time. By fusing information from multiple sources such as vision, lidar, IMU, and force sensing, it provides a data foundation for subsequent localization and path decision-making. A localization module is set up to perform spatiotemporal fusion of multi-source sensor data and use filtering and optimization algorithms to output high-precision robot pose estimation results, thereby ensuring robust localization in complex industrial environments. Secondly, the map building module is used to build a multi-layered fusion map based on perception and positioning results, including geometric map, semantic map, dynamic map and frequency domain feature map. It also provides comprehensive environmental information support for path planning by representing the multi-dimensional environmental space through a weighted fusion strategy. The path module is responsible for executing global path planning and local dynamic obstacle avoidance planning. At the global level, it generates the optimal path based on the task objective and map information. At the local level, it adjusts obstacle avoidance according to real-time obstacles and dynamic changes to achieve efficient and safe navigation. Finally, the switching module adaptively selects either wheeled or legged movement mode based on terrain features, its own posture, and task requirements, achieving a balance between movement efficiency and stability under different terrain conditions. The navigation feedback module is used to execute control commands, monitor navigation deviations, and perform closed-loop corrections. At the same time, based on the feedback results during task execution, it dynamically adjusts path planning parameters and control weights through reinforcement learning and historical task data analysis. Through self-learning and continuous performance optimization, it achieves high-precision positioning in dynamic and complex industrial scenarios during use.

[0038] Example 3 In the third embodiment of the present invention, based on the same inventive concept, the present invention proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the path planning and navigation method for wheeled humanoid robots in industrial scenarios described in the above embodiments.

[0039] Example 4 In the fourth embodiment of the present invention, based on the same inventive concept, a computer device is proposed, comprising: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory to execute the path planning and navigation method for wheeled humanoid robots in industrial scenarios described in the above embodiment.

[0040] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0041] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention 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 the present invention should be included within the protection scope of the present invention.

Claims

1. A path planning and navigation method for wheeled humanoid robots in industrial scenarios, characterized in that, Includes the following steps: S1. By installing lidar, vision camera, inertial measurement unit and odometry sensor on the wheel-legged humanoid robot, the robot's environmental information and its own status information can be obtained; S2. Then, the collected environmental information and self-state information are time-series aligned and coordinate transformed, and the robot's pose estimation is obtained through extended Kalman filter or factor graph optimization algorithm. S3. Generate a geometric map based on LiDAR data, generate a semantic map based on visual information, and identify dynamic obstacles based on time series to obtain a fused environment map; S4. Use the global path planning algorithm to obtain the initial path from the starting point to the target point, and use the local dynamic planning algorithm to perform dynamic obstacle avoidance and path replanning. S5. Based on environmental information such as terrain flatness and ground gradient parameters, determine whether to enable wheeled or legged mode, and control the robot to perform the corresponding movements. S6. Drive the robot to move based on control commands, and adjust the path tracking error feedback according to the real-time positioning results. At the same time, use reinforcement learning algorithm to dynamically adjust the path planning parameters based on the error and energy consumption feedback during the navigation process.

2. The path planning and navigation method for wheeled humanoid robots in industrial scenarios according to claim 1, characterized in that, In step S2, the robot state estimate is obtained by weighted updating of the predicted state and the observed state; The predicted state is determined by the control input, while the observed state is determined by the output of the multimodal sensor. The fusion algorithm employs either the extended Kalman filter algorithm or the nonlinear factor graph optimization algorithm.

3. The path planning and navigation method for wheeled humanoid robots in industrial scenarios according to claim 1, characterized in that, S3 includes the following steps: S301. Establish a multi-layer map structure that includes a geometric layer, a semantic layer, a dynamic layer, and a frequency domain layer; S302. Merge the maps of each layer according to preset weight parameters to obtain a unified environment model; The fusion process is weighted based on geometric feature consistency, semantic matching degree, and obstacle movement frequency characteristics.

4. The path planning and navigation method for wheeled humanoid robots in industrial scenarios according to claim 1, characterized in that, S4 includes the following steps: S401. Use a heuristic search algorithm to search the global map to obtain the initial path; S402. Smooth the initial path. S403. During the local planning process, linear velocity and angular velocity control quantities are generated based on the dynamic window method combined with reinforcement learning strategy. S404. Detect obstacle positions using information acquired by the acquisition module, and update the local path according to the obstacle avoidance cost function; The cost function comprehensively considers four indicators: path orientation, distance to obstacles, speed constraints, and risk factors.

5. The path planning and navigation method for wheeled humanoid robots in industrial scenarios according to claim 1, characterized in that, S5 includes the following steps: S501. Calculate the standard deviation of ground height and terrain gradient through the terrain sensing module; S502. When the standard deviation of ground height is less than the first threshold and the terrain gradient is less than the second threshold, select the wheeled movement mode. S503. When the above conditions are not met, select the foot-walking mode; S504. During mode switching, a model predictive control algorithm is used to perform rolling optimization of the control input to achieve smooth transition and stable path tracking.

6. The path planning and navigation method for wheeled humanoid robots in industrial scenarios according to claim 1, characterized in that, S6 includes the following steps: S601. Perform attitude adjustment based on the speed and attitude commands output by the path planning; S602. Real-time monitoring of robot positioning error and attitude deviation through feedback module; S603. When the positioning error exceeds the set threshold, local path replanning is triggered. S604: Control signals are fed back to the motion controller via the communication bus for navigation control.

7. The path planning and navigation method for wheeled humanoid robots in industrial scenarios according to claim 1, characterized in that, S6 further includes the following steps: An adaptive parameter update strategy based on reinforcement learning is adopted, which comprehensively considers four indicators: path smoothness, energy consumption, obstacle avoidance distance and task completion time through the reward function; The search heuristic parameters and control weights in path planning are automatically adjusted based on historical task data, which is used to learn and adjust path planning based on historical task data.

8. A path planning and navigation system for wheeled humanoid robots in industrial scenarios, characterized in that, The system for the path planning and navigation method of a wheeled humanoid robot for industrial scenarios as described in any one of claims 1-7 includes: The sensing module is used to collect environmental information and robot status data; The positioning module is used to perform spatiotemporal fusion of multi-source sensor data and output robot pose estimation. The map building module is used to create multi-layered maps (geometric, semantic, dynamic, and frequency domain) and perform weighted fusion. The path module is used to perform global path planning and local dynamic obstacle avoidance planning; The switching module is used to adaptively select wheeled or legged movement modes based on terrain features; The navigation feedback module is used to execute control commands, detect navigation deviations and perform feedback corrections, and dynamically adjust path planning and control parameters based on task execution results, learning and adjusting path planning.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the path planning and navigation method for wheeled humanoid robots in industrial scenarios as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which, when executed by a processor, implements the path planning and navigation method for wheeled humanoid robots in industrial scenarios as described in any one of claims 1 to 7.

Images