A multi-robot hierarchical space-time optimization path planning method based on conflict resolution

Abstract

This invention discloses a hierarchical spatiotemporal optimization path planning method for multi-robots based on conflict resolution, belonging to the field of multi-robot path planning technology. The method includes: constructing a semantic topology and grid-based hierarchical graph of the multi-robot operating environment to obtain preset robot paths; detecting and classifying robot path conflicts; calculating the dynamic priorities of robots with path conflicts; evaluating the total cost of each conflict resolution strategy for robot path conflicts; obtaining the optimal conflict resolution strategy for each robot based on the principle of minimizing total cost; selecting robots to be processed based on their dynamic priorities; and replanning the paths of the robots to be processed based on their optimal conflict resolution strategies and preset paths. This invention can perform the lowest-cost replanning for low-priority robots, ultimately eliminating robot conflicts.

Description

Technical Field

[0001] This invention relates to the field of multi-robot path planning technology, and more specifically, to a multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution. Background Technology

[0002] With the rapid development of robotics technology in China, intelligent parking systems based on autonomous parking robots are gradually becoming an effective solution to the parking problem. Autonomous parking robots are automated parking devices that use mobile vehicle carriers to reach the target location. The entire parking process is unmanned. When parking, there is no need to reserve space for passengers to get in and out of the vehicle, indirectly increasing land utilization. Statistics show that autonomous parking can increase parking capacity by more than 30%. At the same time, unmanned parking reduces problems such as manually searching for parking spaces, greatly improving parking efficiency.

[0003] However, due to the limited capabilities of a single robot, it is difficult to complete complex tasks. Therefore, intelligent parking systems often require multiple robots to collaborate in vehicle storage and retrieval. Multi-robot path planning is far more complex than single-robot path planning. When multiple mobile robots exist in a scenario, even if each robot has obstacle avoidance strategies, traffic congestion and blockages can easily occur. A multi-robot path planning system must not only generate an optimized path from the starting point to the target point for each robot, but also avoid collisions between the robot and obstacles in the surrounding environment, as well as between the robot and other robots, preventing deadlocks and enabling multiple robots to successfully complete the designated task through collaboration. This is especially important because parking environments often involve random and dynamic adjustments, such as the movement of other vehicles and the occupancy of parking spaces. The path planning algorithm needs to be able to accurately and in real time perceive and update map data and incorporate it into the path planning considerations, making dynamic adjustments to the path planning; at intersections, multiple robots may arrive at or approach the intersection at the same time, so it is necessary to consider the yield rules between robots, determine the passing order of robots, and allocate appropriate time intervals to avoid intersection congestion and accidents; finally, the algorithm needs to generate high-quality paths within a limited time to meet the real-time requirements.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the problems in related technologies, this invention proposes a multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution, in order to overcome the aforementioned technical problems existing in the current related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows:

[0007] A hierarchical spatiotemporal optimization path planning method for multi-robots based on conflict resolution, the method comprising:

[0008] S1. Construct a semantic topology and grid hierarchical graph of a multi-robot operating environment, and use a graph search algorithm to generate a preset robot path from the starting node to the target node in the semantic topology and grid hierarchical graph.

[0009] S2. Based on the pre-configured classification rules, the collision type of the robot path is initially detected, and the pre-trained deep learning model is used to correct the initial detection of the collision type of the robot path, so as to obtain the detection and classification results of the robot path collision.

[0010] S3. Calculate the dynamic priority of robots with path conflicts based on the degree of entanglement of robot paths and the path execution time.

[0011] S4. Based on the detection and classification results of robot path conflicts, the pre-established conflict type resolution strategy template library and cost function, evaluate the total cost of each conflict resolution strategy for robot path conflicts; and obtain the optimal conflict elimination strategy for each robot according to the principle of minimizing total cost.

[0012] S5. Select robots to be processed based on their corresponding dynamic priorities; replan the paths of the robots to be processed based on the optimal conflict resolution strategy and preset paths.

[0013] Furthermore, a semantic topology and grid hierarchical graph of the multi-robot operating environment are constructed, and a graph search algorithm is used to generate a preset robot path from the starting node to the target node in the semantic topology and grid hierarchical graph, including:

[0014] Import the work environment grid map and feature information, and perform spatiotemporal rasterization on the map. Use the robot width information and the safe distance between robots as the spatial resolution of the work environment grid map, and determine the time interval of the grid occupied according to the robot's travel speed.

[0015] The semantic information of objects in the robot's working environment is labeled and clustered, and a semantic road network map of the working environment is formed based on the semantic information of objects.

[0016] The semantic road network map of the work environment is merged with the grid map of the work environment to obtain a semantic topology and a grid hierarchical map. In the semantic topology and grid hierarchical map, the shortest path from the starting node to the target node generated based on the graph search algorithm is used as the robot's preset path.

[0017] Furthermore, based on pre-configured classification rules, preliminary detection of robot path conflict types is performed, and a pre-trained deep learning model is used to refine the preliminary detection of robot path conflict types, resulting in the following robot path conflict detection and classification results:

[0018] Using pre-configured classification rules, preliminary detection and classification results of robot path conflict types are obtained;

[0019] The environmental perception module acquires environmental sensor data of the multi-robot operating environment, and combines it with the robot's state information to obtain input data. It also acquires different types of collision labels of the robots as output data.

[0020] Using input and output data, a pre-configured multilayer MLP model is trained, and the parameters of the multilayer MLP model are optimized.

[0021] By using the detection and classification results output by the multi-layer MLP model, the preliminary detection and classification results are corrected through weighted adjustment to obtain the detection and classification results of robot path conflicts.

[0022] Furthermore, using pre-configured classification rules, preliminary detection and classification results of robot path conflict types are obtained, including:

[0023] When the robot paths intersect, the type of robot path conflict is determined to be a fixed-point conflict.

[0024] When a group of robots simultaneously traverses an edge and exchanges nodes, the type of robot path conflict is determined to be a fixed-point conflict.

[0025] If the node where the robot is located is occupied in the next moment, the type of robot path conflict is determined to be a following conflict.

[0026] When a group of robots forms a loop, the type of robot path conflict is determined to be a loop conflict.

[0027] Furthermore, based on the degree of entanglement of robot paths and the path execution time, the corresponding dynamic priorities of robots with path conflicts are calculated, including:

[0028] Obtain the total number of times each robot path conflicts with other paths, and assign priorities to robots based on the principle that the more conflicts there are, the lower the priority of the robot.

[0029] When the total number of conflicts in a group of robots is the same, priority is assigned to the robots based on the principle that the shorter the path execution time, the lower the priority of the robot.

[0030] When the total number of conflicts and the execution time of the robot group are the same, the avoidance party and the priority party are randomly selected.

[0031] Furthermore, the pre-established conflict type resolution strategy template library includes:

[0032] Based on the different conflict types of the robot, formulate corresponding conflict resolution strategies; and establish a conflict type resolution strategy template library based on the conflict resolution strategies.

[0033] Furthermore, based on the detection and classification results of robot path conflicts, the pre-established conflict type resolution strategy template library, and the cost function, the total cost of each conflict resolution strategy for robot path conflicts is evaluated, including:

[0034] A cost function is established by defining different robot path conflict types and conflict resolution strategies; the weights in the cost function are adjusted according to pre-defined logic.

[0035] Based on the detection and classification results of robot path conflicts and the pre-established conflict type resolution strategy template library, several conflict type resolution strategies are obtained; the total cost corresponding to each conflict type resolution strategy is calculated using the cost function.

[0036] Furthermore, the weights in the cost function are adjusted according to pre-defined logic, including:

[0037] When the obstacle density is higher than the preset obstacle density threshold, the safety weight is increased;

[0038] When the obstacle density is lower than the preset obstacle density threshold, the path efficiency weight is increased.

[0039] When the robot density on the robot path is higher than the preset robot density threshold, the avoidance and coordination weights are increased.

[0040] When the robot density on the robot path is lower than the preset robot density threshold, the speed weight is increased.

[0041] When an emergency is detected, the weighting of security and response speed is increased.

[0042] Furthermore, based on the robots' corresponding dynamic priorities, the robots to be processed are selected as follows:

[0043] Compare the dynamic priorities of the robots and select the robot with the lowest priority as the robot to be processed.

[0044] Furthermore, based on the optimal conflict resolution strategy and preset path for the robot to be processed, path replanning for the robot to be processed includes:

[0045] Obtain the preset path of the robot to be processed; adjust the preset path of the robot to be processed to a replanned path by executing the best conflict elimination strategy of the robot to be processed.

[0046] The beneficial effects of this invention are as follows:

[0047] (1) This invention constructs a semantic road network graph-grid subgraph to store the semantic constraints, geometric constraints and policy constraints in the working environment in the form of a database, which simplifies the feasible area of ​​the robot's path planning; the object semantic constraints based on the graph model are used to divide the grid map into regions, so that the subsequent path conflict resolution is redistributed in the spatiotemporal grid subgraph, which simplifies the optimization space of the algorithm and improves the real-time performance of the algorithm.

[0048] (2) This invention employs a hierarchical spatiotemporal optimization path planning algorithm, balancing real-time computation with the existence of solutions through a spatiotemporally decoupled spatial path conflict screening and spatiotemporally joint conflict resolution strategy. The spatiotemporally decoupled spatial path conflict screening algorithm first detects the conflicts in the current situation and classifies them, then submits them to the spatiotemporally joint conflict resolution algorithm to plan a conflict-free path. Compared with multiple spatial planning algorithms such as reserved area and A*, when the number of robots is too large, invalid situations where robot paths do not exist are prone to occur; the spatiotemporally joint planning algorithm ensures the existence of multi-robot solutions, that is, in extreme cases, robots can be time-reallocated and executed sequentially. The method of this invention reduces the number of node expansions at the high level of the algorithm and the number of path replannings at the low level, is not affected by the working environment and the number of robots, and can ensure that multiple robots can complete the task.

[0049] (3) This invention provides a more explicit classification of path conflicts, proposing key opposing vertex conflicts, key intersecting vertex conflicts, and key edge conflicts, and designs resolution schemes for each conflict type. The method of this invention adds constraints at a higher level based on the conflict type, and at a lower level updates the set of shortest paths satisfying the constraints through path replanning or by waiting. The method of this invention can resolve current conflicts without triggering new avoidable conflicts, and can resolve multiple conflicts with a single path update. The method of this invention effectively reduces the overall computational cost of the algorithm and can quickly find suboptimal solutions for multi-robot path planning problems. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1This is a flowchart of a multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to an embodiment of the present invention;

[0052] Figure 2 A diagram of a multi-robot path planning system based on hierarchical spatiotemporal optimization with conflict resolution;

[0053] Figure 3 This is a schematic diagram of a semantic road network-raster hierarchical map;

[0054] Figure 4 A schematic diagram showing the mapping between conflict types and resolution strategies;

[0055] Figure 5 This is a flowchart of a multi-robot path planning algorithm. Detailed Implementation

[0056] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0057] According to embodiments of the present invention, a multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution is provided. The multi-robot autonomous parking path planning problem is often complex and nondeterministic, and may involve large-scale data and computation. The efficiency and real-time performance of the algorithm are challenges, requiring efficient search and optimization techniques to improve the performance of path planning algorithms. This will enable the autonomous parking system to be widely used in various parking lots, including commercial centers, public buildings, hospitals, and schools, bringing greater convenience and comfort to people's travel.

[0058] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to an embodiment of the present invention includes:

[0059] S1. Construct a semantic topology and grid hierarchical graph of a multi-robot operating environment, and use a graph search algorithm to generate a preset robot path from the starting node to the target node in the semantic topology and grid hierarchical graph.

[0060] S2. Based on the pre-configured classification rules, the collision type of the robot path is initially detected, and the pre-trained deep learning model is used to correct the initial detection of the collision type of the robot path, so as to obtain the detection and classification results of the robot path collision.

[0061] S3. Calculate the dynamic priority of robots with path conflicts based on the degree of entanglement of robot paths and the path execution time.

[0062] S4. Based on the detection and classification results of robot path conflicts, the pre-established conflict type resolution strategy template library and cost function, evaluate the total cost of each conflict resolution strategy for robot path conflicts; and obtain the optimal conflict elimination strategy for each robot according to the principle of minimizing total cost.

[0063] S5. Select robots to be processed based on their dynamic priorities; replan the paths of the robots to be processed based on the optimal conflict resolution strategy and preset paths.

[0064] To facilitate understanding of the above technical solutions of the present invention, the following further describes the above technical solutions of the present invention from the perspectives of architecture and principle, as follows:

[0065] This invention provides feasible, conflict-free paths for multi-robot parking systems while considering optimality. First, a semantic topology-grid hierarchical map is constructed. The semantic topology map stores key robot movement points and semantic constraints, as well as the offline empirical optimal paths for the corresponding grid subgraphs. When a multi-robot task is launched, a hybrid A* algorithm is used at the bottom layer to perform trajectory search in the grid, obtaining an initial trajectory. This improves the rationality of the planning while ensuring trajectory search efficiency. The overlapping areas of each robot in space are determined through the robot's pre-planned trajectory, enabling the extraction of the semantic subgraph with minimal robot interference. Conflict detection and classification are performed in the lower-level grid map corresponding to the semantic subgraph. The dynamic priority of conflicting multi-robots is calculated. Based on the conflict detection type and robot priority, the cost of conflict resolution strategies such as waiting in place, moving perpendicular to the original direction, and moving in the opposite direction is evaluated. The robot with the lowest priority is replanned in the grid subgraph using the lowest-cost approach, ultimately eliminating robot conflicts. The cost evaluation of conflict resolution strategies is achieved by defining different conflict types and their resolution strategies, establishing a cost function:

[0066]

[0067] in, For time cost, For energy consumption costs, To indicate the degree of path deviation, w1, and These are the weight coefficients for each item. Furthermore, the cost function described above is not the only cost function.

[0068] After identifying the environment, the environmental feature recognition module adjusts weights based on logic such as environmental complexity, path congestion, real-time changes, and task priority: when obstacle density is high, safety weights are increased to prioritize obstacle avoidance; when obstacles are sparse, path efficiency weights are increased to shorten travel time; when robots are densely packed on the path, avoidance and coordination weights are increased to reduce conflict; and when the path is empty, speed weights are increased. When unexpected situations are detected, safety and response speed weights are increased; if the task priority is high, task completion is prioritized. Specifically:

[0069] When the obstacle density is higher than the preset obstacle density threshold, the safety weight is increased [for example, when the obstacle density is >0.6 (more than 0.6 obstacles per square meter), the safety weight is increased to 1.5 times].

[0070] When the obstacle density is lower than the preset obstacle density threshold, the path efficiency weight is increased (for example, when the obstacle density is ≤0.3, the path efficiency weight is increased to 1.2 times).

[0071] When the robot density on the robot path is higher than the preset robot density threshold, the avoidance and coordination weights are increased (for example, when the robot density on the path is greater than the total number of robots * 0.7, the avoidance and coordination weights are increased to twice).

[0072] When the robot density on the robot path is lower than the preset robot density threshold, the speed weight is increased (for example, when the robot density on the path is less than the total number of robots * 0.3, the speed weight is increased to 1.5 times).

[0073] When an unexpected situation is detected, the weights of safety and response speed are increased [for example, when an obstacle suddenly appears or the robot malfunctions, the safety weight is immediately increased to the highest level (weight coefficient 2), while the response speed weight is increased by 1.5 times to ensure that the system can respond quickly and adjust the path].

[0074] Simultaneously, in stable environments, the path smoothness weight is increased to reduce steering and braking, achieving efficient and stable operation. To adapt to different environmental changes, the system selects the strategy with the lowest total cost, and provides real-time feedback to continuously optimize system efficiency and robustness. Low-priority robots undergo the lowest-cost replanning in the grid subgraph. The overall system is a hierarchical spatiotemporal optimization path planning system based on conflict resolution, such as... Figure 2 As shown, the semantic road network-raster hierarchical map, such as Figure 3 As shown, the mapping between conflict types and resolution strategies is as follows: Figure 4 As shown, the multi-robot path planning algorithm flow is as follows: Figure 5 As shown, a hierarchical spatiotemporal optimization path planning method for multi-robots based on conflict resolution is proposed. Figure 5The process includes: calling the semantic road network graph - generating the shortest path for a single robot Hybrid A* - path conflict detection and classification, and calculating the priority of the conflicting robot when a conflict exists - calling the conflict type strategy template - generating the dynamic replanning motion trajectory - calculating the strategy cost and selecting the optimal resolution strategy - generating the dynamic replanning motion trajectory.

[0075] The specific steps of this invention are as follows:

[0076] (1) Import the work environment grid map and information on elements such as objects, entrances / exits, parking locations, and passage characteristics, and perform spatiotemporal rasterization on the map. Assume that robots i are defined, i=1, 2, 3, ..., m, where m is the total number of robots, with the same length and width dimensions (l, w). The smaller width information and the safe distance between robots are used as the spatial resolution of the grid map. The time interval ∆t of the grid occupied by the robot is determined according to the robot's driving speed. Then, the state of the robot's rear axis center point at time step k is:

[0077]

[0078] In the formula, x and y are the robot's position coordinates, and θ is the robot's orientation angle at this time.

[0079] The occupied grid information is as follows:

[0080]

[0081] Where R and T are rotation and translation matrices, This is the initial placeholder grid information.

[0082] (2) The geometric information and semantic information of the working environment are labeled, including road length, width, road signs, parking space types, and elements such as intersections and entrances / exits. Based on the semantic elements, a working environment road network map G=(V, E, F) is formed, where V represents the set of semantic nodes, and each semantic node maps to a connected region with the same semantic meaning in the grid map; E represents the set of connecting edges of each node of the robot, representing the grids where the regions mapped by different semantic nodes intersect and repeat; F represents the semantic node's constraint on the robot's passability. The semantic road network map extends from the starting node V0 to the target node V. g Preset the shortest path. Store the semantic road network map-raster map of the working environment in a discretized database format.

[0083] Furthermore, rasters with the same semantic information are clustered to construct a semantic road network map at the upper layer. Semantic map nodes define scene information and constraints, and preset paths based on experience are stored between nodes. A working environment road network map is formed based on semantic elements, with each semantic node mapping a connected region with the same semantic meaning in the raster map. The set of edges connecting robot nodes represents overlapping rasters in the regions mapped by different semantic nodes. Semantic nodes include constraints on robot mobility. The shortest path from the starting node to the target node is preset in the semantic road network map. The semantic road network map-raster submap of the working environment is stored in a discretized database format.

[0084] (3) Classify the types of robot path conflicts, and design conflict resolution strategies such as waiting in place, moving perpendicular to the original direction, and moving in the opposite direction according to different conflict types, forming a conflict type-resolution strategy template library. Define the robot conflict type as c=[c(1), c(2), c(3), c(4)], representing fixed-point conflict, edge conflict, following conflict, and loop conflict in the robot path; assume that the robot can execute actions as a=[a(1), a(2), a(3), a(4)], representing waiting in place, moving straight, moving laterally, and rotating. The execution time cost of each action is cost=[cost1, cost2, cost3, cost4] time steps. The robot conflict resolution strategy is to execute actions that are inconsistent with the requirements of the original path in order to resolve the conflict, and the expression is:

[0085]

[0086] in, These represent waiting in place, traveling perpendicular to the original direction, and traveling against the original direction, respectively.

[0087] in:

[0088]

[0089]

[0090]

[0091] In the formula, w1, w2, and w3 are optimizable parameters for the time step of the selected action. Indicates the direction of rotation. This indicates the reverse action of the previous step. For example... Figure 3 As shown, a feasible elimination strategy mapping table is established based on the conflict type.

[0092] To improve the accuracy of conflict type classification, this invention combines an environmental perception module and a deep learning model (e.g., a multi-layer MLP model) to achieve dynamic identification and classification adjustment of conflict types. First, environmental perception sensor data and multi-robot state information are collected as input data, and the data is cleaned, normalized, and feature extracted, outputting conflict labels for different types. Next, the deep learning model is trained using historical data from multi-robot path planning and obstacle avoidance, continuously optimizing parameters to accurately classify different conflict types (vertex conflict, edge conflict, following conflict). Based on different conflict types, corresponding conflict resolution strategies are formulated, and a mapping table between conflict types and resolution strategies is established.

[0093] (4) When the planning system is running online, the underlying layer searches the semantic road network map based on the initial position and destination information of each individual robot to form an initial path. Specifically:

[0094] During online planning and execution, the robot receives initial and final position information from multiple robots. Based on a preset road network map and preset routes, it selects a preset road network path v0->, ...,->v for each robot. h ->v g Where -> indicates the direction of the path, v h ->v g Indicates from v h Walk to v g For each individual robot, the Hybrid A* algorithm is used to generate the shortest path for connectable semantic nodes in the grid subgraph, forming the initial path connecting the robot's initial and final positions. The Hybrid A* algorithm is a graph search algorithm.

[0095] (5) Perform path conflict detection and classification for multiple robots, and filter out sub-grid map paths with spatial conflicts to pass to the upper-level planner. Specifically:

[0096] Conflict detection is performed on repeated semantic nodes of multiple robots, and the robots perform actions at time intervals of ∆t. Given a sequence of actions, design a conflict model including collision detection and potential collision detection at time step k:

[0097] This indicates that at time step k, robot i and robot j share the same node.

[0098] This indicates that at time step k, robots i and j simultaneously traverse the edge and exchange nodes.

[0099] This indicates that robot i will occupy the current robot j node in the next moment because it is too close.

[0100] This indicates that multiple robots form a loop, and a deadlock will occur when following conflicts are prohibited.

[0101] in, Indicates the final action. This represents the path node of robot i at time step k. This represents the path node of robot j at time step k. This represents the path node of robot j at time step k+1. This represents the path node of robot i at time step k+1. Indicates at time step k, This represents the path node of robot i+1 at time step k+1. This represents the path node of robot i+2 at time step k. This represents the path node of robot j-1 at time step k+1.

[0102] (6) The upper-level planner determines the priority of robots based on the complexity of multi-robot path entanglement and outputs the robot conflict sub-paths with lower priority. Specifically:

[0103] To determine robot priority in a multi-robot system, the complexity of path entanglement is assessed. First, the number of conflicts between each path and other paths is determined. For each path pair, the number of times they occupy the same position at the same time is counted, and all conflict counts are summed to obtain the total conflict count. The frequency of conflicts with other paths is defined as the degree of entanglement between the two paths.

[0104] Conflict , =SUM k->K [C(1),C(2),C(3),C(4)]

[0105] In the formula, K represents the end point of the path, and k->K represents the process from the starting point to the end point.

[0106] Based on the priority rule of conflict resolution, and since multiple robots have multiple destinations, the sum of the conflict indices between the designed robot path and other robots in the scene is:

[0107] Score =SUM j≠i Conflict , )]

[0108] It can be seen that when the path The more conflicts a robot has with other paths, the lower its priority. The robot needs to perform avoidance and replanning. In particular, when the degree of path entanglement is equal, the path with the longer execution time has higher priority to avoid increasing the overall running time of the planning system. When the degree of entanglement is equal to the path time, the robots are determined to have the same priority and the avoidance path and the priority path are randomly selected.

[0109] (7) The upper-level planner calls the conflict type-resolution strategy template library, calculates the cost function of the resolution strategy, optimizes the strategy, performs spatiotemporal path reallocation and parameter optimization based on the conflict resolution strategy, and outputs a conflict-free resolution path. Specifically:

[0110] By calculating the comprehensive cost of conflict resolution schemes, the path with the minimum total cost is selected. The comprehensive cost considers time cost, energy cost, and path deviation to ensure high efficiency in multi-robot collaboration while maintaining safety. A spatiotemporal optimization algorithm is used to replan paths in the sub-grid map. Time reallocation is used to optimize low-priority robot conflict nodes in the grid sub-map. First, an initial path is planned for each robot in the sub-grid. Then, time is allocated according to the priority of conflict nodes, allowing low-priority robots to appropriately avoid high-priority robots. Finally, by optimizing the path completion time of each robot, the maximum completion time at conflict nodes is minimized, thereby improving the efficiency and coordination of overall path planning. The maximum completion time for replanning conflict path nodes for h conflicting robots is:

[0111]

[0112] Each of them To find the shortest time for the h-th conflicting robot to find a collision-free path within the sub-grid map, a single conflicting robot iterates through a constraint tree, resolving a pair of conflicts in each iteration based on the constraints, until a collision-free path is found. At this point, the optimization function replans the conflict nodes in the spatiotemporal space. Integrating the original path and the replanned path, a collision-free optimal time strategy path is finally generated for multiple robots. This indicates the optimization of weight parameters.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution, characterized in that, The method includes: S1. Construct a semantic topology and grid hierarchical graph of a multi-robot operating environment, and use a graph search algorithm to generate a preset robot path from the starting node to the target node in the semantic topology and grid hierarchical graph. Among them, a road network map of the working environment is formed based on semantic elements, and each semantic node maps to a connected region with the same semantic in the grid map; the set of connecting edges of each node of the robot represents the grids that intersect and repeat in the regions mapped by different semantic nodes; the semantic nodes include constraints on the robot's passability. S2. Based on the pre-configured classification rules, the collision type of the robot path is initially detected, and the pre-trained deep learning model is used to correct the initial detection of the collision type of the robot path, so as to obtain the detection and classification results of the robot path collision. S3. Calculate the dynamic priority of robots with path conflicts based on the degree of entanglement of robot paths and the path execution time. S4. Based on the detection and classification results of robot path conflicts, the pre-established conflict type resolution strategy template library and cost function, evaluate the total cost of each conflict resolution strategy for robot path conflicts; and obtain the optimal conflict elimination strategy for each robot according to the principle of minimizing total cost. S5. Select robots to be processed based on their dynamic priorities; replan the paths of the robots to be processed based on the optimal conflict resolution strategy and preset paths.

2. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 1, characterized in that, The construction of the semantic topology and grid hierarchical graph of the multi-robot operating environment, and the generation of the robot's preset path from the starting node to the target node using a graph search algorithm in the semantic topology and grid hierarchical graph, includes: Import the work environment grid map and feature information, and perform spatiotemporal rasterization on the map. Use the robot width information and the safe distance between robots as the spatial resolution of the work environment grid map, and determine the time interval of the grid occupied according to the robot's travel speed. The semantic information of objects in the robot's working environment is labeled and clustered, and a semantic road network map of the working environment is formed based on the semantic information of objects. The semantic road network map of the work environment is merged with the grid map of the work environment to obtain a semantic topology and a grid hierarchical map. In the semantic topology and grid hierarchical map, the shortest path from the starting node to the target node generated based on the graph search algorithm is used as the robot's preset path.

3. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 1, characterized in that, The pre-configured classification rules are used to initially detect the conflict types of the robot path, and a pre-trained deep learning model is used to correct the initial detection of robot path conflict types, resulting in the following robot path conflict detection and classification results: Using pre-configured classification rules, preliminary detection and classification results of robot path conflict types are obtained; The environmental perception module acquires environmental sensor data of the multi-robot operating environment, and combines it with the robot's state information to obtain input data. It also acquires different types of collision labels of the robots as output data. Using input and output data, a pre-configured multilayer MLP model is trained, and the parameters of the multilayer MLP model are optimized. By using the detection and classification results output by the multi-layer MLP model, the preliminary detection and classification results are corrected through weighted adjustment to obtain the detection and classification results of robot path conflicts.

4. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 3, characterized in that, The preliminary detection and classification results of robot path conflict types obtained using pre-configured classification rules include: When the robot paths intersect, the type of robot path conflict is determined to be a fixed-point conflict. When a group of robots simultaneously traverses an edge and exchanges nodes, the type of robot path conflict is determined to be a fixed-point conflict. If the node where the robot is located is occupied in the next moment, the type of robot path conflict is determined to be a following conflict. When a group of robots forms a loop, the type of robot path conflict is determined to be a loop conflict.

5. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 1, characterized in that, The calculation of the dynamic priority of robots with path conflicts based on the degree of entanglement and path execution time includes: Obtain the total number of times each robot path conflicts with other paths, and assign priorities to robots based on the principle that the more conflicts there are, the lower the priority of the robot. When the total number of conflicts in a group of robots is the same, priority is assigned to the robots based on the principle that the shorter the path execution time, the lower the priority of the robot. When the total number of conflicts and the execution time of the robot group are the same, the avoidance party and the priority party are randomly selected.

6. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 1, characterized in that, The pre-established conflict type resolution strategy template library includes: Based on the different conflict types of the robot, formulate corresponding conflict resolution strategies; and establish a conflict type resolution strategy template library based on the conflict resolution strategies.

7. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 1, characterized in that, The evaluation of the total cost of each conflict resolution strategy based on the robot path conflict detection and classification results, the pre-established conflict type resolution strategy template library, and the cost function includes: A cost function is established by defining different robot path conflict types and conflict resolution strategies; the weights in the cost function are adjusted according to pre-defined logic. Based on the detection and classification results of robot path conflicts and the pre-established conflict type resolution strategy template library, several conflict type resolution strategies are obtained; the total cost corresponding to each conflict type resolution strategy is calculated using the cost function.

8. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 7, characterized in that, The weights in the cost function are adjusted according to a pre-defined logic, including: When the obstacle density is higher than the preset obstacle density threshold, the safety weight is increased; When the obstacle density is lower than the preset obstacle density threshold, the path efficiency weight is increased. When the robot density on the robot path is higher than the preset robot density threshold, the avoidance and coordination weights are increased. When the robot density on the robot path is lower than the preset robot density threshold, the speed weight is increased. When an emergency is detected, the weighting of security and response speed is increased.

9. The multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 1, characterized in that, The step of selecting robots to be processed based on their corresponding dynamic priorities includes: Compare the dynamic priorities of the robots and select the robot with the lowest priority as the robot to be processed.

10. A multi-robot hierarchical spatiotemporal optimization path planning method based on conflict resolution according to claim 9, characterized in that, The path replanning for the robot to be processed, based on the optimal conflict resolution strategy and preset path, includes: Obtain the preset path of the robot to be processed; adjust the preset path of the robot to be processed to a replanned path by executing the best conflict elimination strategy of the robot to be processed.

Images