Path planning method and system for mobile robot

Abstract

The invention discloses a path planning method and system for a mobile robot, and relates to the technical field of path planning methods, and the method comprises the steps: determining a path planning model based on optimized point cloud data, environment complexity and the resolution of a visibility graph; in the determination of the path planning model, the optimized path is determined based on the two-way search of the starting point and the end point, the accuracy of the path planning model is ensured, rapid and accurate path planning is realized based on the two-way search of the starting point and the end point, and the accuracy of the optimized path is ensured. Collecting a plurality of path nodes, and dynamically optimizing each path node; according to the method, the obstacle features are collected according to detection of the optimized path, adaptive filtering is performed on the obstacle features, dynamic optimization of the path nodes and adaptive filtering of the obstacle features are compatible, so that the obstacles which have small influence on path planning can be adaptively filtered, and the calculation burden of path planning can be remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning methods, and in particular to a path planning method and system for a mobile robot. Background Art

[0002] With the development of science and technology, mobile robots are gradually being used in our daily lives and are able to perform corresponding path planning on the road. Mobile robots plan paths based on their current position and target position. In existing technologies, existing path planning technologies mainly rely on a single type of sensor data (such as lidar or camera). The limitations of these sensors often lead to data noise and blind spots in environmental perception, resulting in low accuracy of the paths output by mobile robots in complex environments. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art, and the present invention provides a path planning method and system for a mobile robot.

[0004] An embodiment of the present invention provides a path planning method for a mobile robot, comprising: Determining optimized point cloud data based on multimodal data; the multimodal data includes point cloud data from a laser radar, visual data from a visual sensor, and thermal imaging data from a thermal imaging sensor; Determine the path planning model based on the optimized point cloud data, environment complexity, and visibility map resolution; In the determined path planning model, the optimized path is determined based on a bidirectional search between the starting point and the end point; Collect multiple path nodes and dynamically optimize each path node; Obstacle features are collected based on the detection of the optimized path and adaptively filtered.

[0005] An embodiment of the present invention provides a path planning system for a mobile robot. The path planning system for a mobile robot is applied to the above-mentioned path planning method for a mobile robot. The path planning system for a mobile robot includes: A point cloud data module is used to determine optimized point cloud data based on multimodal data; the multimodal data includes point cloud data from a lidar, visual data from a visual sensor, and thermal imaging data from a thermal imaging sensor; A path planning module that determines the path planning model based on the optimized point cloud data, environment complexity, and visibility map resolution; A path module, configured to determine an optimized path based on a bidirectional search between a starting point and an end point in the path planning model; Path node module, used to collect multiple path nodes and dynamically optimize each path node; The obstacle module is used to collect obstacle features based on the detection of the optimized path and to perform adaptive filtering on the obstacle features.

[0006] Compared with the prior art, the present invention has the following beneficial effects: In an embodiment of the present invention, the method in the embodiment of the present invention is used to determine optimized point cloud data based on multimodal data; the multimodal data includes point cloud data of a lidar, visual data of a visual sensor, and thermal imaging data of a thermal imaging sensor; a path planning model is determined based on the optimized point cloud data, environmental complexity, and the resolution of a visibility map; in the determined path planning model, the optimized path is determined based on a two-way search of a starting point and an end point, which is compatible with the overall consideration of the optimized point cloud data, environmental complexity, and the resolution of the visibility map, thereby ensuring the accuracy of the path planning model, and realizing fast and accurate path planning based on a two-way search of a starting point and an end point, thereby ensuring the accuracy of the optimized path.

[0007] Therefore, multiple path nodes are collected and each path node is dynamically optimized; obstacle features are collected based on the detection of the optimized path, and the obstacle features are adaptively filtered. This is compatible with the dynamic optimization of path nodes and the adaptive filtering of obstacle features, so as to adaptively filter out obstacles that have little impact on path planning, which can significantly reduce the computational burden of path planning while ensuring the accuracy and robustness of path planning, especially in dynamic and complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a flow chart of a path planning method for a mobile robot in an embodiment of the present invention; Figure 2 1 is a flow chart of step S11 in the path planning method for a mobile robot in an embodiment of the present invention; Figure 3 1 is a flow chart of step S12 in the path planning method for a mobile robot in an embodiment of the present invention; Figure 4 1 is a flow chart of step S13 in the path planning method for a mobile robot in an embodiment of the present invention; Figure 5 1 is a flow chart of step S14 in the path planning method for a mobile robot in an embodiment of the present invention; Figure 6 1 is a flow chart of step S15 in the path planning method for a mobile robot in an embodiment of the present invention; Figure 7 It is a schematic diagram of the structural composition of the path planning system of the mobile robot in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0010] See also Figures 1 to 7 A mobile robot path planning method is applied to a mobile robot path planning scenario; the mobile robot path planning method includes: Step S11: Determine optimized point cloud data based on the multimodal data; the multimodal data includes point cloud data from the laser radar, visual data from the visual sensor, and thermal imaging data from the thermal imaging sensor; Step S12: determining a path planning model based on the optimized point cloud data, the environment complexity, and the resolution of the visibility map; Step S13: In the path planning model, an optimized path is determined based on a bidirectional search between a starting point and an end point; Step S14: collecting multiple path nodes and dynamically optimizing each path node; Step S15: Obstacle features are collected based on the detection of the optimized path, and the obstacle features are adaptively filtered; refer to Figure 2 In step S11, optimized point cloud data is determined based on the multimodal data; the multimodal data includes point cloud data from the laser radar, visual data from the visual sensor, and thermal imaging data from the thermal imaging sensor; In the specific implementation process of the present invention, the specific steps are: S111: The mobile robot is equipped with a laser radar, a visual sensor, and an imaging sensor, and collects point cloud data from the laser radar, visual data from the visual sensor, and thermal imaging data from the thermal imaging sensor, and performs multimodal data fusion on the point cloud data from the laser radar, the visual data from the visual sensor, and the thermal imaging data from the thermal imaging sensor to generate fused point cloud data; S112: Smoothing the fused point cloud data and removing outliers in the fused point cloud data to determine optimized point cloud data.

[0011] In an embodiment of the present application, the mobile robot is configured with a laser radar, a visual sensor, and an imaging sensor, and collects point cloud data of the laser radar, visual data of the visual sensor, and thermal imaging data of the thermal imaging sensor. Multimodal data fusion is performed on the point cloud data of the laser radar, the visual data of the visual sensor, and the thermal imaging data of the thermal imaging sensor to generate fused point cloud data, thereby ensuring the accuracy of the fused point cloud data.

[0012] At this time, the mobile robot is equipped with lidar, vision sensors and imaging sensors. The lidar measures the distance by emitting laser beams and receiving the reflected signals, thereby generating point cloud data; this data contains the three-dimensional position and shape information of objects in the environment; vision sensors (such as cameras) capture images of the environment and provide rich texture, color and shape information; this information is crucial for identifying objects, scenes and obstacles; thermal imaging sensors generate thermal images by detecting infrared energy radiated from the surface of an object, showing the temperature distribution of the object; this is very useful for detecting hidden heat sources, heat source distribution and dynamic changes.

[0013] Collect point cloud data from lidar, visual data from vision sensors, and thermal imaging data from thermal imaging sensors. At the same time, ensure that all sensors collect data at the same or similar time points. At the same time, since different sensors may have different coordinate systems and resolutions, their data need to be converted to the same coordinate system; this can be achieved by using known spatial features (such as wall corners, ground landmarks, etc.) for alignment.

[0014] Furthermore, multimodal data fusion is performed on the point cloud data of the lidar, the visual data of the vision sensor, and the thermal imaging data of the thermal imaging sensor to generate fused point cloud data. At the same time, the data from different sensors are weighted averaged to generate fused data. This method is simple but may not fully utilize the correlation between the data.

[0015] At this point, in order to fuse the data from different sensors, we first define the data of each sensor as: P_L: LiDAR point cloud data, usually 3D point data generated by the LiDAR scanning the environment; P_V: Visual sensor data, usually image data taken by a camera; P_T: Thermal imaging sensor data, usually thermal images generated by a thermal imaging camera.

[0016] In a multi-sensor environment, we perform weighted fusion to improve the final data quality. For weighted fusion, we define three weight coefficients: α, β, and γ, corresponding to the lidar, visual sensor, and thermal imaging sensor, respectively. The sum of these weight coefficients is 1: α + β + γ = 1.

[0017] Next, we use the following weighted sum formula to fuse the sensor data: P_f = α × P_L + β × P_V + γ × P_T; Here, P_f represents the fused point cloud data, and α, β, and γ are the weighting coefficients for the sensor data. This allows us to adaptively adjust these weighting coefficients based on the characteristics of different sensors and environmental requirements. For example, in conditions of poor ambient lighting, we might prefer to increase the weight (γ) of thermal imaging sensor data to improve system stability and accuracy. In complex environments, such as fog or low light, LiDAR and vision sensors may struggle to accurately capture the shape and position of objects, while thermal imaging sensors can provide clearer object outlines. By adjusting the weighting coefficients, we can increase the weight (γ) of thermal imaging data, ensuring that the system still obtains reliable environmental data under these complex conditions.

[0018] Therefore, the fused point cloud data is smoothed and the outliers in the fused point cloud data are removed to determine the optimized point cloud data, thereby ensuring the accuracy of the optimized point cloud data.

[0019] At this point, the fused point cloud data is smoothed. Smoothing is intended to reduce noise and outliers in the point cloud data and improve the accuracy and consistency of the data. At the same time, commonly used filtering algorithms include Gaussian filtering, mean filtering, and median filtering. These algorithms smooth the data by calculating the average or weighted average of the neighboring points around each point, and adjust according to the density and noise level of the point cloud; ensure that the smoothing process does not excessively blur details or introduce new errors.

[0020] Gaussian filtering: We use the Gaussian function to smooth each point cloud coordinate. Let the Gaussian function be G(x), and its formula is as follows: G(x) = (1 / (σ × sqrt(2 × π))) × exp(-((x - μ)^2) / (2 × σ^2))

[0021] Where x is the coordinate of the current point cloud; μ is the mean of the Gaussian distribution, representing the expected position of the point cloud; σ is the standard deviation, indicating the degree of data dispersion; and exp is the natural exponential function, describing the smooth transition of point cloud coordinates. Gaussian filtering removes outliers that violate statistical laws, making the point cloud data more uniform and smooth in space, reducing the impact of noise points on environmental modeling.

[0022] At the same time, the outlier processing is designed to identify and remove outliers in the point cloud data, which may be caused by sensor noise, occlusion or erroneous measurement; in order to further enhance the reliability of the data, the system will also detect and remove possible outliers based on the spatial distribution characteristics of the point cloud data. By setting a reasonable threshold, points that are significantly different from the distribution of most point clouds are eliminated to further improve the accuracy of the point cloud data. Application example: In some environments (for example, forests or urban environments), LiDAR may be blocked by objects such as trees and buildings, resulting in the appearance of some noise points. Gaussian filtering and outlier detection methods can effectively remove these noise points to ensure that the quality of the data is not affected in the subsequent path planning and environment modeling processes.

[0023] In an example of this application, it is assumed that we have obtained a point cloud data containing an indoor environment through fusion in step S111, and now smoothing and separation processing is performed; the Gaussian filtering algorithm is selected for smoothing; the radius of the filter is set to 0.1 meters and the standard deviation is 0.05 meters; this means that for each point, we will consider the neighboring points within 0.1 meters around it and use the weighted average method to calculate the new position; after applying Gaussian filtering, the noise and small fluctuations in the point cloud data are effectively reduced, and the overall data is smoother and more consistent.

[0024] A distance-based method was chosen to identify outliers. A distance threshold, such as 0.5 meters, was set, meaning that if a point was greater than 0.5 meters from its nearest neighbor, the point was considered an outlier. After applying the outlier removal algorithm, multiple outliers were successfully identified and removed, likely due to sensor noise or occlusion. After removing the outliers, the point cloud data became cleaner and more accurate. In this example, smoothing and outlier removal optimized the fused point cloud data, reducing the impact of noise and outliers and improving data accuracy and consistency. This provided a more reliable representation of the environment for subsequent path planning, helping to improve the robot's navigation performance and safety.

[0025] Therefore, through the data fusion of lidar, visual sensors and thermal imaging sensors, we can fully utilize the advantages of each sensor and make up for the shortcomings of a single sensor, especially in complex environments, and the fusion weights can be dynamically adjusted according to the actual environment. The Gaussian filtering method is used to remove abnormal data points and improve the smoothness of the point cloud data. At the same time, the outlier detection method is combined to eliminate outliers, further improving the accuracy and reliability of the data. Adaptive weight coefficient: By adjusting the weight coefficient, the system can adaptively optimize the data fusion scheme according to the data quality of different sensors and the complexity of the environment, avoiding the performance problems caused by fixed weight settings. Through this multimodal data fusion and optimization processing method, the present invention can effectively improve the accuracy of environmental modeling and noise suppression capabilities, and is particularly suitable for complex environmental scenes. It has important application value in the fields of autonomous driving, robot navigation, etc.

[0026] refer to Figure 3 , in step S12, a path planning model is determined based on the optimized point cloud data, the environment complexity, and the resolution of the visibility graph; In the specific implementation process of the present invention, the specific steps are: S121: When constructing a corresponding path based on the optimized point cloud data, collecting the types of multiple obstacles detected by the mobile robot and the pass areas, and determining the environment complexity according to the types of multiple obstacles detected by the mobile robot and the pass areas; S122: Triggering adjustment of the spatial resolution of the visibility map according to the complexity of the environment, so that the spatial resolution adapts to different environmental complexities, and determining a path planning model based on the environmental complexity, the spatial resolution of the visibility map, and the training of the optimized point cloud data; In an embodiment of the present application, when constructing a corresponding path based on the optimized point cloud data, the types of multiple obstacles detected by the mobile robot and the passage areas are collected, and the environmental complexity is determined based on the types of multiple obstacles detected by the mobile robot and the passage areas; the overall consideration of the types of multiple obstacles detected by the mobile robot and the passage areas is compatible to ensure the accuracy of the environmental complexity.

[0027] At this time, the mobile robot uses its onboard sensors (such as lidar, cameras, infrared sensors, etc.) to scan the surrounding environment, identify and record the types of obstacles; the types of obstacles can include static obstacles (such as walls, furniture, buildings, etc.) and dynamic obstacles (such as pedestrians, vehicles, pets, etc.); while identifying obstacles, the robot also needs to determine which areas are passable; this is usually done by calculating the gaps between obstacles or based on pre-set map information; the passable area can be a continuous open space or a complex network composed of multiple small areas.

[0028] The robot calculates the number of obstacles per unit area or the proportion of obstacles in the area to assess the density of obstacles. The higher the obstacle density, the higher the environmental complexity. The robot analyzes the width, connectivity, and variability of the passage area. Narrow passages, disconnected areas, or frequently changing passage areas will increase the complexity of the environment. Combining the analysis results of obstacle density and passage area, the robot gives a comprehensive environmental complexity score or classification. This score can be quantitative (such as a numerical range) or qualitative (such as simple, medium, complex).

[0029] Specifically, assume that a mobile robot performs a navigation task in a shopping mall; the robot scans static obstacles such as walls, shelves, and lounge chairs, as well as dynamic obstacles such as pedestrians and shopping carts through lidar; at the same time, the robot identifies passable areas such as aisles, elevator entrances, and rest areas; in the shopping mall, especially during peak hours, the pedestrian density is high, and shopping carts and shelves also occupy a considerable amount of space; the robot calculates that the number of obstacles per unit area is high, so the obstacle density is high; although there are multiple aisles and elevator entrances in the shopping mall, the aisle width in some areas (such as areas with dense shelves) is narrow, and due to the frequent movement of pedestrians and shopping carts, the connectivity and variability of the passable areas are also high; based on the above analysis, the robot assesses the environmental complexity of the shopping mall as "complex"; this means that when planning a path, the robot needs to choose the path more carefully to avoid collisions with obstacles and ensure the connectivity and safety of the path.

[0030] Furthermore, in this invention, we construct a hierarchical visibility graph and adjust its resolution based on the complexity of the environment to improve the efficiency and accuracy of path planning. This method is applicable to various types of automated systems, such as unmanned vehicles and robotic navigation, and is particularly well-suited for environments with dynamic changes and dense physical obstacles.

[0031] Environmental complexity is assessed by comprehensively considering the density of obstacles and the spatial properties of the traffic area. The environmental complexity index C_env is a key parameter for quantifying the overall environmental complexity and is calculated using the following formula: Environmental complexity formula: C_env = (sum from i=1 to N of C_obstacle_i / N) + (sum from j=1 to Mof A_area_j / M) Where: N is the total number of obstacles in the environment. M is the total number of defined traversable areas. C_obstacle_i is the geometric complexity of the i-th obstacle, which may include characteristics such as its size, shape, and layout. A_area_j is the area of ​​the j-th traversable area. Larger traversable areas correspond to smaller contributions to C_env.

[0032] Furthermore, the spatial resolution of the visibility graph is adjusted according to the size of the environmental complexity, so that the spatial resolution adapts to different environmental complexities. The path planning model is determined based on the environmental complexity, the spatial resolution of the visibility graph, and the training of the optimized point cloud data. This is compatible with the overall consideration of the environmental complexity, the spatial resolution of the visibility graph, and the training of the optimized point cloud data, ensuring the accuracy of the path planning model.

[0033] At this point, based on the environmental complexity determined in step S121, the robot has a preliminary understanding of the environment; environmental complexity can be a numerical score or a classification label (such as simple, medium, complex); according to the different environmental complexity, the robot dynamically adjusts the spatial resolution of the visibility graph; in complex environments, in order to improve the accuracy and flexibility of path planning, the robot adopts a higher spatial resolution, that is, increases the number of nodes and the density of edges in the visibility graph; in this way, the robot can more accurately capture detailed information in the environment, such as narrow passages, tiny gaps between obstacles, etc.; on the contrary, in simple environments, the robot adopts a lower spatial resolution to reduce the amount of calculation and increase the planning speed.

[0034] The robot collects optimized point cloud data, which contains the geometric information of the environment and obstacle information; at the same time, based on the adjusted visibility map, the robot generates a series of training samples, including the starting point, end point and possible paths; based on the complexity of the environment and the spatial resolution, the robot selects or designs a suitable path planning model; the model can be search-based (such as A×, Dijkstra, etc.), sampling-based (such as RRT, PRM, etc.), or learning-based (such as deep reinforcement learning, neural networks, etc.); in complex environments, the robot may be more inclined to use learning-based methods because these methods can handle more complex environments and dynamic changes.

[0035] The robot performs training using the prepared training data and the selected model. The training process may involve multiple iterations to optimize the model's parameters and performance. During training, the robot also considers the impact of spatial resolution to ensure that the model works effectively at different resolutions. After training, the robot verifies the model using a validation dataset to ensure its performance in the actual environment. The validation process may include testing the model's path planning capabilities in a simulated environment, as well as limited testing in the actual environment.

[0036] Specifically, assume that a mobile robot is performing path planning in a complex warehouse environment; the warehouse environment contains many tall shelves, narrow aisles, and dynamically changing inventory; the robot evaluates the environment complexity as "complex" and adjusts the spatial resolution of the visibility graph accordingly; in the narrow aisles and densely packed shelves of the warehouse, the robot uses a higher spatial resolution to capture detailed information in these areas; while in the open areas of the warehouse, the robot uses a lower spatial resolution to reduce the amount of computation.

[0037] Based on the complexity of the environment and the adjusted spatial resolution, the robot selected a path planning model based on deep reinforcement learning. This model is capable of handling complex environments and dynamic changes, and can generate effective paths on visibility maps of different resolutions. The robot uses the optimized point cloud data and the adjusted visibility map to generate training samples and train the model. During training, the robot continuously adjusts the model parameters to improve its path planning capabilities in complex environments. After training, the robot uses a validation dataset to verify the model to ensure its performance in the warehouse environment.

[0038] At the same time, based on the estimated environmental complexity C_env, this paper proposes an adaptive resolution adjustment strategy. This strategy is used to determine the spatial resolution used when constructing the visibility graph to ensure an optimal balance between graph accuracy and computational efficiency. The resolution adjustment formula is as follows: R_env = k × (C_env)^(1 / p) Where k is a constant used to adjust the baseline resolution based on actual application requirements. p is a weighting factor for environmental complexity, which adjusts the impact of environmental complexity on resolution. The resolution R_env gradually increases with increasing C_env. This ensures that the visibility graph provides more detailed information in areas of higher environmental complexity, thereby guiding more accurate path planning.

[0039] The core innovation of the present invention lies in dynamically adjusting the resolution of environmental modeling through a hierarchical and adaptive approach, which not only reduces the demand for computing resources, but also improves the accuracy and response speed of path planning in complex environments. For example, in an urban traffic environment, complex intersection areas are accurately modeled through high-resolution maps, while lower resolutions are used on open highways, effectively improving the overall operating efficiency of the system. In addition, the method of adaptively adjusting the resolution enables the system to better respond to sudden changes in the environment, such as changes in road conditions caused by temporary construction areas or emergencies. The system can quickly adjust the resolution to adapt to new environmental conditions, ensuring the stability and reliability of the navigation system. Through practical applications, the present invention has been proven to be able to provide efficient and accurate navigation support under various environmental conditions, greatly improving the user experience and the safety of the system.

[0040] refer to Figure 4 In step S13, in the path planning model, an optimized path is determined based on a bidirectional search between a starting point and an end point; In the specific implementation process of the present invention, the specific steps are: S131: In the path planning model, a start point and an end point are collected, and a bidirectional search is performed from the start point and the end point. A first path and a second path are determined based on the bidirectional search of the start point and the end point, and an optimized path is determined based on the synthesis of the first path and the second path. S132: In the bidirectional search between the starting point and the end point, the search for the starting point and the search for the end point will meet in the middle, then the path search is completed, and multiple path nodes are marked in the optimized path. The state of each path node s_i will be dynamically adjusted according to the environmental information obtained in real time, including the adjustment of the search step size Δs_i.

[0041] The cost of the path search C_total is calculated as follows: C_total = sum from i=1 to N of (d(s_i, s_(i+1)) + w × h(s_i)) Where: d(si, si+1)) is the actual distance between path node si and the adjacent path node si+1; w is the weight of the heuristic function, which helps balance the actual distance and the estimated distance to the destination; h(si) is the heuristic estimate, that is, the estimated distance from the current path node si to the destination. By dynamically adjusting w and h(si), we can adapt to various environmental changes and optimize the path search efficiency.

[0042] In an embodiment of the present application, in the path planning model, the starting point and the end point are collected, and a bidirectional search is performed from the starting point and the end point. The first path and the second path are determined based on the bidirectional search of the starting point and the end point, and the optimized path is determined based on the synthesis of the first path and the second path; the first path and the second path are determined by being compatible with the bidirectional search of the starting point and the end point, and the accuracy of the optimized path is guaranteed based on the overall consideration of the synthesis of the first path and the second path.

[0043] The present invention achieves fast and accurate path planning by combining adaptive bidirectional path search with a dynamic optimization algorithm. This approach is particularly suitable for dynamically changing environments, such as autonomous driving and advanced robotic navigation systems, where environmental uncertainty and complexity require path planning systems that are not only efficient but also dynamically adaptable to environmental changes.

[0044] At the same time, in the path planning model, it is first necessary to clarify the starting point (starting position) and the end point (target position); this information is usually provided by the robot's task instructions or user input; the position information of the starting point and the end point can be expressed as coordinate points in two-dimensional or three-dimensional space; starting from the starting point, the robot uses a certain search algorithm (such as A×, Dijkstra, etc.) to expand outward to build a search tree or search graph; each node in the search tree / graph represents a possible position, and the edge represents the movement from one position to another; at the same time, starting from the end point, the robot also uses the same or different search algorithm to backtrack inward to build another search tree / graph; the purpose of bidirectional search is to approach from the starting point and the end point at the same time to reduce the search space and improve the efficiency of path planning.

[0045] When the starting point search and the end point search meet at an intermediate node, a path from the starting point to the end point has been found. This path is composed of the portion from the starting point search to the meeting node (the first path) and the portion from the end point search to the meeting node (the second path). In some cases, there may be multiple meeting paths, and the robot can choose the shortest, safest, or least expensive path. Furthermore, after combining the first and second paths, optimization may be required, such as smoothing the path, removing redundant nodes, and adjusting the path length. The optimized path should be as short, smooth, and safe as possible, while avoiding collisions with obstacles.

[0046] Specifically, imagine a mobile robot performing a handling task in a warehouse environment, moving from shelf A (the starting point) to shelf B (the end point) to pick up an item. The robot obtains the precise coordinates of shelf A and shelf B through its positioning system. Starting from shelf A, the robot uses the A× algorithm to search, while simultaneously searching from shelf B using the Dijkstra algorithm. Both algorithms consider path length and obstacle information. During the search, the starting and ending search paths converge at a warehouse aisle intersection. The paths before this intersection constitute the first and second paths, respectively. After combining the first and second paths, the robot discovers some unnecessary twists and turns and redundant nodes. To optimize the path, the robot uses a smoothing algorithm to remove these twists and redundant nodes, resulting in a shorter, smoother path. Ultimately, the robot moves from shelf A to shelf B along the optimized path, successfully completing the handling task. During this process, the bidirectional search algorithm significantly improves the efficiency of path planning, while path optimization ensures that the robot can perform its task safely and efficiently.

[0047] Furthermore, in the bidirectional search between the starting point and the end point, the search at the starting point and the search at the end point will meet in the middle, completing the path search and marking multiple path nodes in the optimized path. The state of each path node s_i will be dynamically adjusted according to the environmental information obtained in real time, including the adjustment of the search step size Δs_i.

[0048] The cost of the path search C_total is calculated as follows: C_total = sum from i=1 to N of (d(s_i, s_(i+1)) + w × h(s_i)) Where: d(si, si+1)) is the actual distance between path node si and the adjacent path node si+1; w is the weight of the heuristic function, which helps balance the actual distance and the estimated distance to the destination; h(si) is the heuristic estimate, that is, the estimated distance from the current path node si to the destination. By dynamically adjusting w and h(si), we can adapt to various environmental changes and optimize the path search efficiency.

[0049] At this time, in the bidirectional search process between the starting point and the end point, the search tree (or search graph) of the starting point continues to expand outward, while the search tree (or search graph) of the end point continues to backtrack inward; when the two meet at an intermediate node, it means that a valid path from the starting point to the end point has been found; at this point, the bidirectional search process ends.

[0050] After the encounter, the robot marks all the nodes on this path to form a complete path; these nodes are usually numbered in sequence, such as s_1, s_2, ..., s_n, where s_1 is the starting point and s_n is the end point; each node s_i contains its position information in space (such as coordinates) and its connection relationship with other nodes; as the robot moves along this path, the state of each node s_i may be dynamically adjusted based on the real-time environmental information obtained; state adjustments may include fine-tuning the node's position, adjusting the search step Δs_i (that is, the distance the robot needs to travel from the current node to the next node), and changes in the connection relationship between nodes; the search step Δs_i refers to the distance the robot needs to move from node si to node s(i+1); this distance can be adjusted based on environmental information; for example, in an area with dense obstacles, for safety reasons, the robot may reduce the search step to avoid obstacles more accurately; while in an open and obstacle-free area, the robot may increase the search step to improve movement efficiency.

[0051] Specifically, suppose a mobile robot performs a navigation task in a complex indoor environment and needs to move from room A (starting point) to room B (end point); the robot starts searching in room A using the A× algorithm and simultaneously starts searching in room B using the Dijkstra algorithm; during the search process, the two meet in the middle of a corridor, indicating that a valid path has been found.

[0052] The robot marks all nodes along this path, forming a path from room A to room B. These nodes include the starting point (room A's doorway), several key points in the corridor (such as turns and obstacle detours), and the end point (room B's doorway). As the robot moves along this path, it continuously acquires real-time environmental information through its sensors (such as lidar and cameras). For example, when the robot approaches an obstacle, it detects its position and size and adjusts the position of the next node (such as slightly offsetting the obstacle) and the search step size (such as reducing it to avoid collision). In spacious sections of the corridor, the robot may increase its search step size to quickly traverse the area. However, when approaching the doorway of room B, the robot reduces its search step size to accurately align with the doorway and avoid collisions with the doorframe. Ultimately, the robot successfully moves from room A to room B along this dynamically adjusted path, completing the navigation task. In this process, the bidirectional search algorithm improves the efficiency of path planning, while the dynamic adjustment of path nodes ensures that the robot can safely and flexibly cope with various challenges in complex environments.

[0053] refer to Figure 5,In step S14, multiple path nodes are collected and each path node is dynamically optimized; In the specific implementation process of the present invention, the specific steps are: S141: Collect multiple path nodes, determine corresponding surrounding obstacles based on perimeter detection of the multiple path nodes, and determine obstacle change events based on position changes of the surrounding obstacles; S142: When an obstacle change event occurs, triggering a corresponding dynamic optimization based on the corresponding path node, and adjusting the corresponding path portion in the dynamic optimization to update the optimized path in real time; In the path optimization of the path nodes, the path optimization formula is used: C_path = sum from i=1 to N of (d(s_i, s_(i+1)) × O_node(s_i)) Where: O_node(s_i) is the path node optimization coefficient, which is dynamically adjusted according to the importance of the path node s_i or its relative position to the obstacle to ensure the practicality and safety of the path.

[0054] In an embodiment of the present application, multiple path nodes are collected, corresponding surrounding obstacles are determined based on the peripheral detection of the multiple path nodes, and obstacle change events are determined based on the position changes of the surrounding obstacles; At this time, during the movement of the robot, it will continuously collect and update the information of multiple nodes on its path; these nodes are usually numbered according to time sequence or spatial position, such as s_1, s_2, ..., s_n, where n is the total number of path nodes; each path node contains the state information of the robot at that position, such as position coordinates, speed, direction, etc.

[0055] For each path node, the robot uses its equipped sensors (such as lidar, cameras, ultrasonic sensors, etc.) to scan and detect the surrounding environment; by analyzing sensor data, the robot can identify surrounding obstacles and determine the location, shape, size and other information of these obstacles; obstacles may be static (such as walls, trees, buildings, etc.) or dynamic (such as pedestrians, vehicles, other robots, etc.).

[0056] The robot continuously detects obstacles around each path node and records their location information. By comparing obstacle location information at different time points or locations, the robot can identify changes in obstacle positions. When the position of an obstacle changes significantly (such as sudden appearance, disappearance, movement, etc.), the robot will regard it as an obstacle change event. An obstacle change event may mean a change in environmental conditions, such as the sudden appearance of pedestrians, the movement of vehicles, the collapse of obstacles, etc.

[0057] Specifically, assume that a mobile robot is performing an inspection task in a factory workshop; in stage S141, the robot moves along a predetermined path and collects information on multiple path nodes; at path node s_3, the robot uses its lidar to scan the surrounding environment; it detects several static obstacles, such as walls, machinery and equipment, and shelves; at the same time, it also detects a dynamic obstacle, namely a moving worker; the robot continues to move and scans the surrounding environment again at path node s_5; at this time, it finds that the worker previously detected has moved to a new position; by comparing the obstacle position information of the two nodes s_3 and s_5, the robot recognizes the position change of the worker and regards it as an obstacle change event.

[0058] Based on this obstacle change event, the robot dynamically adjusts its subsequent path; it may choose a new path that avoids the worker, or adjust its movement speed and direction to ensure that it can continue to complete the inspection task while avoiding collisions. In this example, the robot continuously detects obstacles around path nodes and identifies changes in the position of obstacles, allowing it to flexibly respond to dynamic changes in the environment and ensure that the task is completed safely and efficiently.

[0059] Furthermore, when an obstacle change event occurs, a corresponding dynamic optimization is triggered based on the corresponding path node, and the corresponding path part is regulated in the dynamic optimization to update the optimized path in real time, thereby realizing real-time update of the optimized path.

[0060] At this time, during the movement of the robot, the sensors it is equipped with will continuously detect the surrounding environment, including the position and status of obstacles; when a significant change in the position or status of an obstacle is detected (such as sudden appearance, disappearance, change in movement direction, etc.), the robot will identify an obstacle change event; once an obstacle change event is detected, the robot will immediately search for information on the current path node or nearby path nodes; based on the information of these path nodes, the robot will trigger a dynamic path optimization mechanism; this mechanism may be based on preset rules, algorithms or machine learning models.

[0061] After the dynamic optimization mechanism is triggered, the robot analyzes the portion of the current path affected by the obstacle change event. Based on the analysis results, the robot adjusts the affected path portion. This may include replanning local paths, adjusting the positions of path nodes, changing the search step size or speed, etc. The purpose of the adjustment is to ensure that the robot can safely and efficiently bypass obstacles and continue to move along the optimized path.

[0062] After adjustment, the robot will generate a new, optimized path; this new path will be updated to the robot's path planning system in real time and become the basis for the robot's subsequent movement; at the same time, the robot will also record the results of this dynamic optimization for subsequent analysis and improvement.

[0063] Specifically, suppose an autonomous vehicle is driving on a city road, and its path planning system has preset an optimal path from the starting point to the end point. During the driving process, the vehicle suddenly detects a pedestrian walking out from the roadside and about to cross the road. The appearance of this pedestrian constitutes an obstacle change event. The vehicle immediately searches for information about the current path node and triggers a dynamic path optimization mechanism. This mechanism is implemented based on preset obstacle avoidance rules and algorithms.

[0064] Through analysis, the vehicle determined that the portion of the current path affected by the pedestrian was the next straight road it would travel on; to avoid a collision with the pedestrian, the vehicle decided to adjust this portion of the path; the vehicle replanned a local path that allowed it to safely bypass the pedestrian; this new path included a left turn, bypassing the pedestrian, and then a right turn to return to the original path; the vehicle updated the optimized path in real time and continued driving along the new path; at the same time, the results of this dynamic optimization were also recorded for subsequent analysis and improvement; in this example, the autonomous vehicle successfully avoided a collision with the pedestrian by detecting obstacle change events, triggering the dynamic path optimization mechanism, regulating the affected path portion, and updating the optimized path in real time, ensuring safe and efficient driving.

[0065] In the path optimization of the path nodes, the path optimization formula is used: C_path = sum from i=1 to N of (d(s_i, s_(i+1)) × O_node(s_i)) Where: O_node(s_i) is the path node optimization coefficient, which is dynamically adjusted according to the importance of the path node s_i or its relative position to the obstacle to ensure the practicality and safety of the path.

[0066] Alternatively, this adaptive bidirectional search and dynamic optimization technology is crucial in autonomous vehicles. For example, in complex urban traffic environments, obstacles (such as pedestrians, other vehicles, and emergencies) frequently appear, and traditional one-way search path planning methods may not be able to respond to these dynamic changes in a timely manner. The bidirectional search method of the present invention can quickly find an effective path between the starting and ending points, and through dynamic optimization, it can adapt to unexpected situations, ensuring safe and smooth driving. This method is also effective in robot navigation, especially in warehouses or manufacturing scenarios with constantly changing environments, where robots need to adjust their paths in real time to avoid collisions and optimize transportation efficiency. By adjusting path nodes and search strategies in real time, the robot's operational efficiency and safety can be significantly improved. Combining bidirectional search with dynamic node optimization provides a real-time and effective path planning solution in complex and dynamic environments. The advantages of this method are its ability to quickly adapt to environmental changes, improve the real-time performance and accuracy of path planning, reduce computing resource consumption, and increase the robustness and adaptability of the system.

[0067] refer to Figure 6 , in step S15, obstacle features are collected based on the detection of the optimized path, and the obstacle features are adaptively filtered; In the specific implementation process of the present invention, the specific steps are: S151: Collecting the optimized path, collecting point cloud data of the obstacle based on obstacle detection of the optimized path, determining obstacle features based on the point cloud data of the obstacle, and determining the geometric complexity of the obstacle features based on the edge length and number of vertices of the obstacle features; S152: If the geometric complexity of the obstacle feature is less than a preset complexity threshold, adaptive filtering of the obstacle feature is triggered; if the geometric complexity of the obstacle feature is greater than the preset complexity threshold, retention of the obstacle feature is triggered; Calculate the set complexity of obstacles (C_obstacle), the geometric complexity formula is: C_obstacle = (L × V) / A_area Where: C_obstacle: The geometric complexity of the obstacle, indicating the complexity of the obstacle; L: The edge length of the obstacle, in meters (m); V: The number of vertices of the obstacle, indicating the geometric complexity of the obstacle; A_area: The area occupied by the obstacle, in square meters (m²); Obstacle filtering formula: F_filtered = {obstacle_i | C_obstacle_i>C_threshold} Where: F_filtered: the set of obstacles after filtering, that is, the retained obstacles; obstacle_i: the i-th obstacle; C_obstacle_i: the geometric complexity of the i-th obstacle; C_threshold: the complexity threshold, the set standard value, which determines which obstacles should be filtered.

[0068] In an embodiment of the present application, an optimized path is collected, point cloud data of the obstacle is collected based on obstacle detection of the optimized path, and obstacle features are determined based on the point cloud data of the obstacle. The geometric complexity of the obstacle features is determined based on the edge length and number of vertices of the obstacle features, thereby ensuring the accuracy of the geometric complexity of the obstacle features.

[0069] At this point, the system first obtains the optimized path information; these paths may be optimal or suboptimal paths calculated by the path planning algorithm based on factors such as the environment map, obstacle distribution, and task requirements; the optimized path is usually represented by a series of coordinate points or path segments, which constitute the trajectory that the robot or inspection equipment needs to follow.

[0070] When a robot or related equipment moves along an optimized path, it uses sensors (such as lidar, cameras, ultrasonic sensors, etc.) to monitor the surrounding environment in real time; these sensors can capture information about obstacles on the path and convert it into digital signals or data that can be used for subsequent processing.

[0071] For detected obstacles, the system uses sensors to collect their three-dimensional point cloud data; point cloud data is a collection of a large number of discrete three-dimensional coordinate points, which together describe the three-dimensional shape and position of the obstacle; the process of collecting point cloud data usually involves steps such as sensor scanning, data acquisition, and data preprocessing; by processing point cloud data, the system can extract the characteristic information of the obstacle; these features may include geometric features such as the obstacle's edges, vertices, planes, and surfaces, as well as spatial features such as the obstacle's size, direction, and position; the feature extraction process usually involves steps such as point cloud segmentation, feature recognition, and feature description.

[0072] In this step, the system calculates the geometric complexity of the obstacle based on the extracted obstacle feature information. Geometric complexity is an indicator that measures the complexity of the obstacle shape. It can be quantified by the edge length and number of vertices of the obstacle. Longer edge lengths and greater number of vertices usually mean a more complex obstacle shape and higher geometric complexity.

[0073] Specifically, suppose a robot is performing an inspection task in a warehouse environment with some shelves and goods placed in the warehouse as obstacles; the robot first calculates an optimized path from the starting point to the end point through a path planning algorithm; the robot obtains an optimized path that avoids most shelves and goods to ensure that the robot can complete the task safely and efficiently; while moving along the optimized path, the robot uses lidar to monitor the surrounding environment in real time; the lidar scans a pile of goods in front as an obstacle.

[0074] The lidar captures three-dimensional point cloud data of the cargo pile, which contains information such as its shape, position, and size. By processing this point cloud data, the system extracts the pile's edge and vertex features. The shape of the cargo pile approximates an irregular polyhedron with multiple edges and vertices. The system calculates the total edge length of the cargo pile to be approximately 10 meters, and the number of vertices to be approximately 15. Based on a preset complexity calculation formula (e.g., edge length weight 1 + vertex weight 2), the system calculates the cargo pile's geometric complexity to be high, indicating that the cargo pile's shape is relatively complex, requiring the robot to pay more attention and processing during navigation. Through this process, the system can accurately collect and optimize obstacle information along the path and formulate a corresponding navigation strategy based on the geometric complexity of the obstacles to ensure that the robot can complete the task safely and efficiently.

[0075] Furthermore, if the geometric complexity of the obstacle feature is less than a preset complexity threshold, adaptive filtering of the obstacle feature is triggered; if the geometric complexity of the obstacle feature is greater than the preset complexity threshold, retention of the obstacle feature is triggered; Calculate the set complexity of obstacles (C_obstacle), the geometric complexity formula is: C_obstacle = (L × V) / A_area Where: C_obstacle: The geometric complexity of the obstacle, indicating the complexity of the obstacle; L: The edge length of the obstacle, in meters (m); V: The number of vertices of the obstacle, indicating the geometric complexity of the obstacle; A_area: The area occupied by the obstacle, in square meters (m²); According to the above formula, more complex obstacles (those with longer edges or more vertices) will have a higher complexity value, C_obstacle. These obstacles may have a greater impact on path planning and should be retained. On the other hand, simple obstacles (those with shorter edges and fewer vertices) generally have a smaller impact on path planning and can be ignored.

[0076] Obstacle filtering formula: F_filtered = {obstacle_i | C_obstacle_i>C_threshold} Where: F_filtered: the set of obstacles after filtering, that is, the retained obstacles; obstacle_i: the i-th obstacle; C_obstacle_i: the geometric complexity of the i-th obstacle; C_threshold: the complexity threshold, the set standard value, which determines which obstacles should be filtered.

[0077] This adaptive filtering method allows the system to flexibly adjust the filtering criteria based on the characteristics of obstacles in different environments. Complex environments will retain more details, while simple environments will improve path planning efficiency by ignoring some small obstacles.

[0078] At this point, the system has extracted the obstacle features based on the obstacle's point cloud data and calculated the geometric complexity of these features. Geometric complexity is a quantitative indicator used to measure the complexity of the obstacle shape. The system will pre-set one or more complexity thresholds based on factors such as task requirements, robot capabilities, and environmental characteristics. These thresholds are used to determine whether the complexity of the obstacle features meets the standard that requires special treatment. The system will compare the calculated geometric complexity of the obstacle features with the preset complexity thresholds. This comparison process may be performed feature by feature or it may be a holistic assessment of the entire obstacle.

[0079] If the geometric complexity of the obstacle feature is less than the preset complexity threshold, it means that the obstacle feature is relatively simple and will not have much impact on the robot's navigation or task execution; at this time, the system will trigger the adaptive filtering mechanism to simplify or ignore the obstacle feature to reduce the amount of calculation and improve processing efficiency; if the geometric complexity of the obstacle feature is greater than the preset complexity threshold, it means that the obstacle feature is relatively complex and may have a greater impact on the robot's navigation or task execution; at this time, the system will retain all key information of the obstacle feature to ensure the accuracy and reliability of subsequent path planning, navigation or obstacle avoidance strategies.

[0080] Specifically, suppose a robot performs a navigation task in a complex indoor environment that contains obstacles of various shapes and sizes, such as furniture, walls, and pillars. The system uses lidar to scan an irregularly shaped table in front of it as an obstacle, and extracts features such as the edges and vertices of the table. Based on these features, the system calculates that the geometric complexity of the table is medium.

[0081] The system sets two complexity thresholds in advance: a low threshold and a high threshold; the low threshold is used to filter out obstacle features with very simple shapes, such as planes or straight lines; the high threshold is used to retain obstacle features with very complex shapes, such as three-dimensional structures with multiple curved edges and vertices; the system compares the geometric complexity of the table with the preset low and high thresholds; since the geometric complexity of the table is at a medium level, it is neither lower than the low threshold nor higher than the high threshold; although the geometric complexity of the table does not directly trigger the low or high threshold filtering or retention mechanism, the system can make decisions based on actual conditions; for example, if the table has little impact on the robot's navigation path and the robot's computing power is limited, the system can choose to trigger the adaptive filtering mechanism and simplify the feature representation of the table to reduce the amount of calculation; on the contrary, if the table has a greater impact on the robot's navigation path and requires precise planning of obstacle avoidance strategies, the system chooses to retain all key features of the table to ensure the accuracy and safety of navigation. Suppose in this example, the system decides to retain all the key features of the table; therefore, in the subsequent path planning or navigation process, the system will take into account the precise position and shape of the table to ensure that the robot can safely bypass the table and reach its destination; through such a process, the system can adaptively adjust the processing strategy according to the geometric complexity of the obstacle features to improve the robot's navigation efficiency and safety.

[0082] Optionally, one of the core innovations of this invention is an adaptive filtering method based on obstacle geometry. This method evaluates the geometric complexity of obstacles, including edge length and number of vertices, to adaptively filter out obstacles that have little impact on path planning. This method can significantly reduce the computational burden of the path planning algorithm while maintaining path planning accuracy and robustness, especially in dynamic and complex environments.

[0083] In practical applications, particularly in areas like autonomous driving and robotic navigation, the number of obstacles in the environment is often enormous, many of which have little impact on path planning. Using an adaptive filtering method based on geometric features, the system can filter out redundant obstacles, thereby reducing computational complexity and improving path planning efficiency. For example, in an autonomous driving system navigating urban roads, a vehicle may encounter numerous small obstacles (such as small rocks or roadside objects) that have minimal impact on the driving path. The filtering method of the present invention can effectively ignore these obstacles, thereby accelerating path calculation and improving real-time responsiveness. Furthermore, the adaptive filtering method based on obstacle geometric features of the present invention is also adaptable to dynamic environments. In practical applications, the position and shape of obstacles are constantly changing. For example, in indoor navigation, furniture, walls, and people may change at any time. By assessing the complexity of obstacles in real time and dynamically adjusting the filtering strategy, this method ensures that path planning is always performed in a simplified yet sufficiently accurate environment, thereby improving system robustness. The innovation of the present invention lies in the ability to dynamically and flexibly adjust the filtering criteria through the adaptive filtering method based on obstacle geometric features to remove obstacles that have little impact on path planning. This not only significantly reduces the amount of computation, but also improves real-time performance and system robustness while maintaining path planning accuracy. Compared with traditional obstacle handling methods, this invention can provide a more efficient and adaptable path planning solution in complex and dynamically changing environments.

[0084] See also Figure 7 , Figure 7 : is a schematic diagram of the structure of a path planning system for a mobile robot in an embodiment of the present invention; the path planning system for a mobile robot includes: A point cloud data module 21 is configured to determine optimized point cloud data based on multimodal data; the multimodal data includes point cloud data from a laser radar, visual data from a visual sensor, and thermal imaging data from a thermal imaging sensor; A path planning module 22 is configured to determine a path planning model based on the optimized point cloud data, the environment complexity, and the resolution of the visibility map; A path module 23 is configured to determine an optimized path based on a bidirectional search between a starting point and an end point in the path planning model; The path node module 24 is used to collect multiple path nodes and dynamically optimize each path node; The obstacle module 25 is used to collect obstacle features based on the detection of the optimized path and perform adaptive filtering on the obstacle features.

[0085] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A path planning method for a mobile robot, characterized in that: include: Determining optimized point cloud data based on multimodal data; the multimodal data includes point cloud data from a laser radar, visual data from a visual sensor, and thermal imaging data from a thermal imaging sensor; Determine the path planning model based on the optimized point cloud data, environment complexity, and visibility map resolution; In the determined path planning model, the optimized path is determined based on a bidirectional search between the starting point and the end point; Collect multiple path nodes and dynamically optimize each path node; Obstacle features are collected based on the detection of the optimized path and adaptively filtered.

2. The path planning method for a mobile robot according to claim 1, wherein: The optimized point cloud data is determined based on the multimodal data; the multimodal data includes point cloud data from a laser radar, visual data from a visual sensor, and thermal imaging data from a thermal imaging sensor, including: The mobile robot is equipped with a laser radar, a visual sensor, and an imaging sensor, and collects point cloud data from the laser radar, visual data from the visual sensor, and thermal imaging data from the thermal imaging sensor, and performs multimodal data fusion on the point cloud data from the laser radar, the visual data from the visual sensor, and the thermal imaging data from the thermal imaging sensor to generate fused point cloud data; The fused point cloud data is smoothed and outliers in the fused point cloud data are removed to determine the optimized point cloud data.

3. The path planning method for a mobile robot according to claim 1, wherein: The path planning model is determined based on the optimized point cloud data, the environmental complexity, and the resolution of the visibility map, including: When constructing a corresponding path based on the optimized point cloud data, the types of multiple obstacles detected by the mobile robot and the passage areas are collected, and the environmental complexity is determined based on the types of multiple obstacles detected by the mobile robot and the passage areas; The spatial resolution of the visibility graph is adjusted according to the complexity of the environment, so that the spatial resolution adapts to different environmental complexities. The path planning model is determined based on the environmental complexity, the spatial resolution of the visibility graph, and the training of the optimized point cloud data.

4. The path planning method for a mobile robot according to claim 1, wherein: In the path planning model, determining the optimized path based on a bidirectional search between a starting point and an end point includes: In the path planning model, a starting point and an end point are collected, and a bidirectional search is performed from the starting point and the end point. A first path and a second path are determined based on the bidirectional search of the starting point and the end point, and an optimized path is determined based on the synthesis of the first path and the second path.

5. The path planning method for a mobile robot according to claim 4, characterized in that: In the path planning model, the optimized path is determined based on a bidirectional search between the starting point and the end point, and further includes: In a bidirectional search between the starting point and the end point, the search for the starting point and the search for the end point will meet in the middle, completing the path search and marking multiple path nodes in the optimized path. The state of each path node s_i will be dynamically adjusted based on the real-time environmental information, including the adjustment of the search step size Δs_i; The cost of the path search C_total is calculated as follows: C_total = sum from i=1 to N of (d(s_i, s_(i+1)) + w × h(s_i)) Where: d(si, si+1)) is the actual distance between path node si and the adjacent path node si+1; w is the weight of the heuristic function, which helps balance the actual distance and the estimated distance to the destination; h(si) is the heuristic estimate, that is, the estimated distance from the current path node si to the destination. By dynamically adjusting w and h(si), we can adapt to various environmental changes and optimize the path search efficiency.

6. The path planning method for a mobile robot according to claim 1, wherein: The collecting of multiple path nodes and dynamically optimizing each path node includes: Collect multiple path nodes, determine corresponding surrounding obstacles based on the surrounding detection of the multiple path nodes, and determine obstacle change events based on the position changes of the surrounding obstacles; When an obstacle change event occurs, the corresponding dynamic optimization is triggered based on the corresponding path node, and the corresponding path part is adjusted in the dynamic optimization to update the optimized path in real time.

7. The path planning method for a mobile robot according to claim 6, characterized in that: The collecting of multiple path nodes and dynamically optimizing each path node further includes: In the path optimization of the path nodes, the path optimization formula is used: C_path = sum from i=1 to N of (d(s_i, s_(i+1)) × O_node(s_i)) Where: O_node(s_i) is the path node optimization coefficient, which is dynamically adjusted according to the importance of the path node s_i or its relative position to the obstacle to ensure the practicality and safety of the path.

8. The path planning method for a mobile robot according to claim 1, wherein: The collecting obstacle features based on the detection of the optimized path and adaptively filtering the obstacle features include: Collecting the optimized path, collecting point cloud data of the obstacle based on obstacle detection of the optimized path, determining obstacle features based on the point cloud data of the obstacle, and determining the geometric complexity of the obstacle features based on the edge length and the number of vertices of the obstacle features; If the geometric complexity of the obstacle feature is less than the preset complexity threshold, adaptive filtering of the obstacle feature is triggered; if the geometric complexity of the obstacle feature is greater than the preset complexity threshold, retention of the obstacle feature is triggered.

9. The path planning method for a mobile robot according to claim 8, characterized in that: The collecting obstacle features based on the detection of the optimized path and adaptively filtering the obstacle features further includes: Calculate the set complexity of obstacles (C_obstacle), the geometric complexity formula is: C_obstacle = (L × V) / A_area Where: C_obstacle: The geometric complexity of the obstacle, indicating the complexity of the obstacle; L: The edge length of the obstacle, in meters (m); V: The number of vertices of the obstacle, indicating the geometric complexity of the obstacle; A_area: The area occupied by the obstacle, in square meters (m²); Obstacle filtering formula: F_filtered = {obstacle_i | C_obstacle_i > C_threshold} Among them: F_filtered: the set of obstacles after filtering, that is, the retained obstacles; obstacle_i: the i-th obstacle; C_obstacle_i: the geometric complexity of the i-th obstacle; C_threshold: the complexity threshold, the set standard value, which determines which obstacles should be filtered.

10. A path planning system for a mobile robot, characterized in that: The path planning system of the mobile robot is applied to the path planning method of the mobile robot according to any one of claims 1 to 9, and the path planning system of the mobile robot includes: A point cloud data module is used to determine optimized point cloud data based on multimodal data; the multimodal data includes point cloud data from a lidar, visual data from a visual sensor, and thermal imaging data from a thermal imaging sensor; A path planning module that determines the path planning model based on the optimized point cloud data, environment complexity, and visibility map resolution; A path module, configured to determine an optimized path based on a bidirectional search between a starting point and an end point in the path planning model; Path node module, used to collect multiple path nodes and dynamically optimize each path node; The obstacle module is used to collect obstacle features based on the detection of the optimized path and to perform adaptive filtering on the obstacle features.

Images