A conflict-based multi-vehicle path planning optimization method
By generating a set of feasible paths before vehicle movement and optimizing them during the runtime phase, and combining the concepts of conflict graphs and drive-away trees, the problems of low efficiency and computational complexity in multi-vehicle path planning are solved, achieving efficient and highly adaptable path planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI ZHENHUA HEAVY IND
- Filing Date
- 2022-12-29
- Publication Date
- 2026-05-12
AI Technical Summary
Existing multi-vehicle path planning methods are inefficient and computationally complex in resolving spatiotemporal conflicts. They cannot effectively distinguish the cost of conflict resolution, resulting in unreasonable path planning. Furthermore, they rely on prior knowledge of specific scenarios or have limited effectiveness.
A conflict-based multi-vehicle path planning optimization method is adopted. By generating a set of feasible paths before the vehicles move and continuously optimizing them during the vehicle operation phase, the shortest path is generated using the Dijkstra algorithm. The path planning is optimized by combining the concepts of conflict graph and drive-away tree, thereby reducing computational pressure and the formation of conflict loops.
It improves the adaptability and efficiency of path planning, reduces computational complexity, ensures the feasibility and near-optimality of path planning, avoids unnecessary computation and path optimization, is highly adaptable, and does not depend on map layout and business knowledge.
Smart Images
Figure CN116796913B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to automated warehousing and logistics park technologies, and more specifically, to a conflict-based multi-vehicle route planning optimization method. Background Technology
[0002] Automated warehousing and logistics parks refer to application scenarios where automated handling equipment, such as Automated Guided Vehicles (AGVs) and automated forklifts, performs material handling operations within a fixed space. To ensure the safety and efficiency of multiple vehicles operating within the same space, pre-planning vehicle routes in both time and space becomes a key issue that needs to be addressed. For example... Figure 1 The image shows a typical warehouse scenario with multiple vehicles operating simultaneously, including... Figure 1 The gray areas are impassable, the white areas are passable, circles represent the starting positions of vehicles, stars represent the target positions, numbers represent vehicle numbers, and lines connecting circles to stars represent the planned vehicle paths. When a vehicle travels through a white square on the path, it must occupy at least one square. Therefore, if two or more vehicles need to occupy the same square at the same time, it is defined as a spatiotemporal conflict. The problem then becomes how to plan a reasonable path for each vehicle, ensuring that all vehicles do not experience spatiotemporal conflicts while minimizing travel time.
[0003] Theoretically, the multi-vehicle optimal route planning problem is an NP-complete problem, meaning there is no single optimal solution. Therefore, the industry's focus is on finding approximate optimal solutions under specific industry constraints. Currently, the most common approaches include:
[0004] 1) Use a single-vehicle path planning algorithm with time windows (such as Dijkstra, A*, and their variants) to plan the shortest path in space and time for each vehicle. This type of method is relatively simple and fast, using time windows to resolve conflicts with other vehicles. However, it considers factors rather coarsely, usually employing greedy algorithms such as setting priorities and planning order to solve potentially complex problems between multiple vehicles. It is not very efficient overall, often performing unnecessary detours for minor avoidance maneuvers.
[0005] 2) The multi-vehicle path planning method, which has emerged in recent years, considers the possible changes in the paths of other vehicles while planning their own paths, seeking the optimal combination in a multi-vehicle game. A representative method in this category is Conflict-Based Path Planning (CBS). Its idea is to focus on conflicts between vehicles (two or more vehicles passing through the same area at the same time) and process them sequentially according to time. Each time a conflict is addressed, one side of the conflict is attempted to avoid it in time and space using a time-window-based path planning method. Each avoidance generates a new situation, and each situation is scored according to the number of conflicts that need to be addressed. Through iteration, the method converges towards a path with fewer conflicts and where conflicts occur in the future, ultimately finding a completely conflict-free path. This method can find a good comprehensive approximate optimal path, but it has two drawbacks: firstly, there are too many similar situations, leading to low search efficiency (a typical example is...). Figure 2 As shown, there are too many similar situations; secondly, the cost of conflict resolution was not distinguished, that is, although the previous decisions reduced the number of conflicts, they led to an awkward situation where the cost of resolving the remaining conflicts was too high (e.g. Figure 3 As shown, car number 1 will resolve conflicts a and b at minimal cost by going first, but this will lead to a very high cost in resolving c. Although some improvements based on prior knowledge have been proposed, they are either limited to a specific application scenario or have limited effectiveness. Summary of the Invention
[0006] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a conflict-based multi-vehicle path planning optimization method that effectively avoids the two major problems in existing technologies and possesses the characteristics of wide adaptability and high retrieval efficiency.
[0007] To achieve the above objectives, the present invention adopts the following technical solution:
[0008] A conflict-based multi-vehicle path planning optimization method:
[0009] Before the vehicle moves, the search expectation is relaxed, a set of feasible paths is generated, and the set of feasible paths is continuously optimized until the vehicle moves.
[0010] Preferably, the relaxed search expectation further includes:
[0011] Dijkstra's algorithm is used to search from the endpoint to the starting point, recording the distance from each node to the endpoint and the relationship between the parent nodes. In subsequent searches, the shortest path from any node to the endpoint is directly queried.
[0012] Preferably, generating the set of feasible paths further includes:
[0013] S1. Generate the shortest spatial path for each vehicle individually, and then estimate the time window for each node.
[0014] S2. Add the next path to the spatiotemporal conflict graph according to priority. When two vehicles pass through the same node, construct time constraints according to the first-come, first-served approach.
[0015] S3. Determine whether step S2 is feasible. If yes, end the process; otherwise, proceed to step S4.
[0016] S4. Cancel the routes of newly added vehicles and add them to the list of vehicles to be re-planned;
[0017] S5. Determine if there are any paths that have not been added to the conflict list. If yes, return to step S2; otherwise, proceed to step S6.
[0018] S6. Perform feasible route planning for each vehicle to be re-planned.
[0019] Preferably, in step S1, the feasible path planning for a single vehicle further includes:
[0020] S101. Pack the directed connected graph G and the conflict tree consisting of the starting point and other paths into search nodes, and save them to the weight priority list open_list with the current shortest path as the weight.
[0021] S102. Determine if the weight priority list open_list is empty. If it is, it means that the path does not exist, and the process ends. If not, proceed to step S103.
[0022] S103. Select the best search node in the weighted priority list open_list, take the path node C, and query the shortest path R2 from C to the destination.
[0023] S104. Try to add the shortest path R2 to the directed connected graph G;
[0024] S105. Determine if there is a strongly connected component. If yes, proceed to step S106. If no, combine the shortest path R1 from the starting point to path node C and the path R2 from path node C to the end point into the best path and add it to the directed connected graph G to generate a new directed connected graph G.
[0025] S106. Determine whether the subsequent spatial node N of path node C has been obtained. If yes, proceed to step S107. If no, store the current search node in the searched list close_list.
[0026] S107. Check if the searched list close_list already exists. If it exists, proceed to step S108. If it does not exist, return to step S106.
[0027] S108. Continue to determine whether there is a weight priority list open_list. If there is, relax the repeated nodes and return to step S106. If there is no such node, proceed to step S109.
[0028] S109. The path from the starting point to the subsequent spatial node N is R1, and we attempt to add it to the directed connected graph G.
[0029] S110. Determine if a strong connection component exists. If yes, add the vehicle associated with the strong connection component to the driving tree. If no, proceed to step S111.
[0030] S111. Determine if there are any vehicles without a path to be driven away. If yes, proceed to step S112. If no, combine the new subsequent spatial node N, the directed connected graph G, and the driving tree to form a new search node. Calculate the weight based on the shortest path of the subsequent spatial node N, add it to the weight priority list open_list, and then return to step S106.
[0031] S112. Add unrouteed vehicles to the tree-driving system;
[0032] S113. Determine if there is a loop conflict when driving away the tree. If yes, return to step S106; otherwise, proceed to step S114.
[0033] S114. The driven vehicle recursively uses this path planning method to avoid the obstacle;
[0034] S115. Determine whether the avoidance was successful. If yes, proceed to step S116; otherwise, return to step S106.
[0035] S116. Calculate the cost, package the new path directed connected graph G and the drive tree into a new search node, calculate the weight of the shortest path of the subsequent spatial node N plus the avoidance cost, and add it to the weight priority list open_list.
[0036] Preferably, in step S6, if the vehicle's current path differs from the desired shortest path, then the following steps are performed:
[0037] S601. Divide a set of feasible paths into fixed time_steps in the time dimension to form a discretized spatiotemporal three-dimensional graph.
[0038] S602. Determine if there are any paths that need optimization. If yes, proceed to step S603; otherwise, end.
[0039] S603, Select any vehicle that needs optimization;
[0040] S604. Without changing the spatiotemporal paths of other vehicles, plan a path on the spatiotemporal three-dimensional map.
[0041] S605. Determine if there is a better path. If there is, update the path. If not, return to step S602.
[0042] The conflict-based multi-vehicle path planning optimization method provided by this invention also has the following beneficial effects:
[0043] 1) Overall, the approach of planning multiple feasible routes in the early stage and then optimizing them individually during the execution period alleviates the computational pressure of solving in the early stage;
[0044] 2) In the stage of solving the feasible solution set of multi-vehicle paths, retain as many spatial shortest paths as possible and resolve conflicts in the time dimension. This ensures the quality of the basic solution set and avoids unnecessary calculations.
[0045] 3) For tasks that require replanning feasible paths, the concept of conflict diagrams is creatively introduced to ensure that the newly planned path is feasible and that path planning efficiency is guaranteed.
[0046] 4) In the worst case (requiring multi-vehicle collaboration), the concept of a driving tree is introduced to ensure that multi-vehicle collaboration can be achieved by driving vehicles, and a feasible solution set is planned.
[0047] 4) Compared with industry methods, it is highly adaptable (does not rely on prior conditions such as map layout and business knowledge), efficient (avoids symmetric repeated retrieval, and uses feasible solutions to replace near-optimal solutions in the early stage), does not lose feasible solutions (there are still solutions in the worst case), and does not lose near-optimal solutions (optimization is carried out during the execution period). Attached Figure Description
[0048] Figure 1 This is a schematic diagram of an existing multi-vehicle work scenario in a warehouse;
[0049] Figure 2 This is a schematic diagram of a situation similar to the existing CBS;
[0050] Figure 3 This is a schematic diagram illustrating the mutual influence of existing conflicts.
[0051] Figure 4 This is a schematic diagram of the shortest path for existing reverse retrieval;
[0052] Figure 5 This is a schematic diagram of A* path retrieval in the multi-vehicle path planning optimization method of the present invention;
[0053] Figure 6 This is a schematic diagram illustrating the effect of the multi-vehicle path planning optimization method of the present invention;
[0054] Figure 7 This is a schematic diagram of the spatiotemporal path in the multi-vehicle path planning and optimization method of the present invention;
[0055] Figure 8 This is a schematic diagram of the spatiotemporal path relationships in the multi-vehicle path planning and optimization method of the present invention;
[0056] Figure 9 This is a schematic diagram of the vehicle driving relationship in the multi-vehicle path planning optimization method of the present invention;
[0057] Figure 10 This is a flowchart illustrating the multi-vehicle path planning optimization method of the present invention;
[0058] Figure 11 This is a flowchart illustrating the process of generating a feasible path set in the multi-vehicle path planning optimization method of the present invention.
[0059] Figure 12 This is a flowchart illustrating the vehicle avoidance path planning process in the multi-vehicle path planning optimization method of the present invention.
[0060] Figure 13 This is a flowchart illustrating the optimization of the path set in the multi-vehicle path planning optimization method of the present invention. Detailed Implementation
[0061] To better understand the above-mentioned technical solutions of the present invention, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0062] Combination Figure 10 As shown, the conflict-based multi-vehicle path planning optimization method provided by this invention consists of two stages:
[0063] In the first stage, before the vehicle moves, the search expectation is relaxed to generate a set of feasible paths. In the second stage, the set of feasible paths is continuously optimized to find the Pareto optimum and is continuously optimized until the vehicle moves.
[0064] In the first phase, the focus shifts from the earliest conflicts to the conflict cycles that could lead to deadlock. Specifically, this involves isolating conflict resolution methods in both time and space. This essentially preserves the shortest path space as much as possible, avoiding conflicts by waiting in time, thus significantly reducing the search space. If adjustments in the time dimension alone cannot meet the requirements, a driving action is needed. This is achieved through recursive avoidance attempts using a driving relationship tree. This ensures that expanding the search scope is only necessary in the worst-case scenario, thus avoiding deadlock. Figure 2 The case study contains numerous invalid state calculations.
[0065] Furthermore, during path planning, the iterative optimization direction is not solely focused on reducing conflicts, but rather prioritizes whether existing conflicts will form deadlock loops. This approach considers related conflicts holistically, effectively characterizing the importance and cost of resolving them. This avoids the awkward situation where resolving earlier conflicts leads to more severe conflicts in subsequent paths, or where resolving later conflicts requires backtracking and adjusting previously resolved conflicts. Figure 3 The example shown.
[0066] 4. In the process of finding a set of feasible paths, excessive waiting is inevitable (due to the principle of resolving conflicts in the time dimension while minimizing changes to the path space). After solving for the set of feasible paths, and ensuring that the paths of other vehicles do not worsen, path planning based on time windows is used to optimize the vehicle's own path, i.e., to seek the Pareto optimal path. Multiple vehicles can iterate and attempt concurrently, and this process can be carried out during travel without affecting the efficiency of the overall path planning. Thus, by delaying optimization, the computational pressure is alleviated.
[0067] like Figure 6 As shown, vehicle 1 is an automated guided vehicle (AGV), vehicle 2 is a forklift, and location A is a dedicated forklift work area, which vehicle 1 cannot pass through. This conflicting situation (such as...) Figure 6 As shown in (a), by trying different routes in sequence using time windows, a detour route will be planned (e.g., Figure 6 As shown in (b). Using CBS path planning, either prior constraints are used to find a detour route, or a large number of conflict scenarios are calculated before a reasonable plan can be devised. This invention, however, can calculate the optimal plan by simply driving car 1 two steps while car 2 is planning its route (e.g., ...). Figure 6 (as shown in (c)).
[0068] Relaxed search expectations further include:
[0069] For each transport task, without considering other vehicles, Dijkstra's algorithm is used to search from the destination to the starting point, recording the distance from each node to the destination and the relationship between the parent nodes. The search node space only needs to be expanded to 130% of the optimal path (adjustable), that is, some search nodes are expanded for backup. In subsequent searches, the shortest path from any node to the destination is directly queried. For example... Figure 4 As shown, the plan starts from the pentagram at the end point and moves towards the circle at the starting point, saving the shortest distance at the top left corner of the node, while the arrow points to the parent node.
[0070] When adjusting the planned path later, the A* algorithm is used to plan from the starting point to the destination, and the above results are used as an estimate of the destination. For example... Figure 5As shown, the plan extends from the starting circle to the ending pentagram. The shortest distance from the starting point is stored in the lower right corner of the node. The parent node relationships pointing to the starting point and the ending point are omitted and not labeled in the diagram.
[0071] Determining whether multiple spatiotemporal paths are feasible: such as Figure 7 As shown, two paths travel towards each other. The solid curve represents the vehicle starting constraint (vehicles already on a node must occupy it first), and the dashed line represents the adjustable constraint on the passage order. However, in this diagram, both dashed lines will result in a cycle, corresponding to cycles L1 and L2 on the diagram. This means that these two paths are impassable. Conversely, if all vehicles passing through the same node can find an order that prevents the path from forming a cycle, then the path set is traversable.
[0072] If vehicles are driven away, the driving relationships form a tree structure, such as... Figure 9 As shown, if car A drives car B, and car B drives car C, then the driven car cannot drive the source car again; that is, a loop cannot be formed. Figure 9 As shown by the dashed line.
[0073] like Figure 11 As shown, a set of feasible paths is generated. First, the shortest spatial path is planned for each vehicle, and then the time window for each node is estimated. This process does not consider the positions and paths of other vehicles, so it can be processed in parallel. Then, the paths are added to the conflict graph (e.g., ...) in order of priority. Figure 8 As shown in the diagram, vehicles passing through the same spatial node are added with time constraints in a first-come, first-served order. If a newly added path causes a conflict cycle (using Tarjan's algorithm in graph theory to calculate strongly connected components; if they exist, there is a conflict cycle; if there are no strongly connected components, there is no conflict cycle), the path is canceled and added to the list of vehicles to be replanned. This process continues until all vehicles have been processed (this step primarily aims to reduce the scale using a greedy approach). Finally, for each vehicle that needs replanning, a new feasible path is planned sequentially. The specific steps are as follows:
[0074] S1. Generate the shortest spatial path for each vehicle individually, and then estimate the time window for each node.
[0075] S2. Add the next path to the spatiotemporal conflict graph according to priority. When two vehicles pass through the same node, construct time constraints according to the first-come, first-served approach.
[0076] S3. Determine whether step S2 is feasible. If yes, end the process; otherwise, proceed to step S4.
[0077] S4. Cancel the routes of newly added vehicles and add them to the list of vehicles to be re-planned;
[0078] S5. Determine if there are any paths that have not been added to the conflict list. If yes, return to step S2; otherwise, proceed to step S6.
[0079] S6. Perform feasible route planning for each vehicle to be re-planned.
[0080] Feasible route planning for a single vehicle, such as Figure 12 As shown, the framework of this method is A*, but the search nodes not only consider spatiotemporal nodes, but also the conflict graph formed with other paths, and the driving relationship tree generated with other vehicles. The calculation method of other path weights, etc., is as follows. Figure 5 As shown. The specific steps are as follows:
[0081] S101. Pack the directed connected graph G and the conflict tree consisting of the starting point and other paths into search nodes, and save them to the weight priority list open_list with the current shortest path as the weight.
[0082] S102. Determine if the weight priority list open_list is empty. If it is, it means that the path does not exist, and the process ends. If not, proceed to step S103.
[0083] S103. Select the best search node in the weighted priority list open_list, take the path node C, and query the shortest path R2 from C to the destination.
[0084] S104. Try to add the shortest path R2 to the directed connected graph G;
[0085] S105. Determine if there is a strongly connected component. If yes, proceed to step S106. If no, combine the shortest path R1 from the starting point to path node C and the path R2 from path node C to the end point into the best path and add it to the directed connected graph G to generate a new directed connected graph G.
[0086] S106. Determine whether the subsequent spatial node N of path node C has been obtained. If yes, proceed to step S107. If no, store the current search node in the searched list close_list.
[0087] S107. Check if the searched list close_list already exists. If it exists, proceed to step S108. If it does not exist, return to step S106.
[0088] S108. Continue to determine whether there is a weight priority list open_list. If there is, relax the repeated nodes and return to step S106. If there is no such node, proceed to step S109.
[0089] S109. The path from the starting point to the subsequent spatial node N is R1, and we attempt to add it to the directed connected graph G.
[0090] S110. Determine if a strong connection component exists. If yes, add the vehicle associated with the strong connection component to the driving tree. If no, proceed to step S111.
[0091] S111. Determine if there are any vehicles without a path to be driven away. If yes, proceed to step S112. If no, combine the new subsequent spatial node N, the directed connected graph G, and the driving tree to form a new search node. Calculate the weight based on the shortest path of the subsequent spatial node N, add it to the weight priority list open_list, and then return to step S106.
[0092] S112. Add unrouteed vehicles to the tree-driving system;
[0093] S113. Determine if there is a loop conflict when driving away the tree. If yes, return to step S106; otherwise, proceed to step S114.
[0094] S114. The driven vehicle recursively uses this path planning method to avoid the obstacle;
[0095] S115. Determine whether the avoidance was successful. If yes, proceed to step S116; otherwise, return to step S106.
[0096] S116. Calculate the cost, package the new path directed connected graph G and the drive tree into a new search node, calculate the weight of the shortest path of the subsequent spatial node N plus the avoidance cost, and add it to the weight priority list open_list.
[0097] The following points should be noted in the above steps:
[0098] 1) Search nodes are based on spatial nodes, with additional conflict graph and driving tree information formed by other paths. Only when the conflict graph and driving tree are completely identical will it be determined whether the spatial nodes are identical, that is, relaxation is performed or the search is marked and added to the close_list.
[0099] 2) When calculating node weights, in addition to the distance to the starting point and the distance to the ending point, it is also necessary to calculate the additional costs incurred by all vehicles (i.e., those driven by this vehicle) in the driving tree being driven away.
[0100] 3) If the vehicle being driven does not have a mission, it can be directly planned to a feasible point and can complete the avoidance. If it has a mission, when planning the route, if it is necessary to avoid the obstacle, continue planning to the destination, and then the extra cost is the difference between the cost of the new route and the original route.
[0101] 4) If the site is very limited and there are indeed complex driving relationships, it may lead to excessively long search times. In this case, the depth of the driving tree can be limited to reduce overly complex paths.
[0102] After generating a set of feasible paths for all vehicles, all vehicles passing through the same spatial node now have a definite passage order, meaning there is a feasible solution for the passage process. However, this solution has room for optimization, for example:
[0103] 1) In actual traffic, there may be instances where vehicles do not arrive as expected or arrive early. As long as the order of modification does not cause a circular conflict on the multi-vehicle route conflict map, the change is allowed.
[0104] 2) such as Figure 13 As shown, if the vehicle finds that the current path differs significantly from the expected shortest path, it can replan the path without affecting other paths, optimizing both time and space (when generating feasible solutions, time-based waiting and avoidance methods were prioritized, potentially ignoring better spatial avoidance methods). Note that a spatiotemporal map is used here (when initially planning the spatial shortest path, a spatial map is used, without considering the time dimension). That is, a time dimension is extended onto the two-dimensional spatial map and divided into fixed time intervals, each interval being a time_step. A time window consists of one or more consecutive time_steps. This constructs a directed connected graph in three-dimensional spatiotemporal space, and path search can directly use ordinary A*. The specific steps are as follows:
[0105] S601. Divide a set of feasible paths into fixed time_steps in the time dimension to form a discretized spatiotemporal three-dimensional graph.
[0106] S602. Determine if there are any paths that need optimization. If yes, proceed to step S603; otherwise, end.
[0107] S603, Select any vehicle that needs optimization;
[0108] S604. Without changing the spatiotemporal paths of other vehicles, plan a path on the spatiotemporal three-dimensional map.
[0109] S605. Determine if there is a better path. If there is, update the path. If not, return to step S602.
[0110] Those skilled in the art should recognize that the above embodiments are merely illustrative of the present invention and are not intended to limit the present invention. Any variations or modifications to the above embodiments that are within the spirit and essence of the present invention will fall within the scope of the claims of the present invention.
Claims
1. A conflict-based multi-vehicle path planning optimization method, characterized in that: Before the vehicle moves, the search expectation is relaxed to generate a set of feasible paths. This set of feasible paths is continuously optimized until the vehicle moves. The generation of the feasible path set further includes: S1. Generate the shortest spatial path for each vehicle individually, and then estimate the time window for each node. S2. Add the next path to the spatiotemporal conflict graph according to priority. When two vehicles pass through the same node, construct time constraints according to the first-come, first-served approach. S3. Determine whether step S2 is feasible. If yes, end the process; otherwise, proceed to step S4. S4. Cancel the routes of newly added vehicles and add them to the list of vehicles to be re-planned; S5. Determine if there are any paths that have not been added to the conflict list. If yes, return to step S2; otherwise, proceed to step S6. S6. Perform feasible route planning for each vehicle to be re-planned. In step S1, the feasible path planning for a single vehicle further includes: S101. Pack the directed connected graph G and the conflict tree consisting of the starting point and other paths into search nodes, and save them to the weight priority list open_list with the current shortest path as the weight. S102. Determine if the weight priority list open_list is empty. If it is, it means that the path does not exist, and the process ends. If not, proceed to step S103. S103. Select the best search node in the weighted priority list open_list, take the path node C, and query the shortest path R2 from C to the destination. S104. Try to add the shortest path R2 to the directed connected graph G; S105. Determine if there is a strongly connected component. If yes, proceed to step S106. If no, combine the shortest path R1 from the starting point to path node C and the path R2 from path node C to the end point into the best path and add it to the directed connected graph G to generate a new directed connected graph G. S106. Determine whether the subsequent spatial node N of path node C has been obtained. If yes, proceed to step S107. If no, store the current search node in the searched list close_list. S107. Check if the searched list close_list already exists. If it exists, proceed to step S108. If it does not exist, return to step S106. S108. Continue to determine whether there is a weight priority list open_list. If there is, relax the repeated nodes and return to step S106. If there is no such node, proceed to step S109. S109. The path from the starting point to the subsequent spatial node N is R1, and we attempt to add it to the directed connected graph G. S110. Determine if a strong connection component exists. If yes, add the vehicle associated with the strong connection component to the driving tree. If no, proceed to step S111. S111. Determine if there are any vehicles without a path to be driven away. If yes, proceed to step S112. If no, combine the new subsequent spatial node N, the directed connected graph G, and the driving tree to form a new search node. Calculate the weight based on the shortest path of the subsequent spatial node N, add it to the weight priority list open_list, and then return to step S106. S112. Add unrouteed vehicles to the tree-driving system; S113. Determine if there is a loop conflict when driving away the tree. If yes, return to step S106; otherwise, proceed to step S114. S114. The driven vehicle recursively uses this path planning method to avoid the obstacle. S115. Determine whether the avoidance was successful. If yes, proceed to step S116; otherwise, return to step S106. S116. Calculate the avoidance cost, package the new path directed connected graph G and the drive-off tree into a new search node, calculate the weight of the shortest path of the subsequent spatial node N plus the avoidance cost, and add it to the weight priority list open_list.
2. The conflict-based multi-vehicle path planning optimization method according to claim 1, characterized in that, The relaxed search expectation further includes: Dijkstra's algorithm is used to search from the endpoint to the starting point, recording the distance from each node to the endpoint and the relationship between the parent nodes. In subsequent searches, the shortest path from any node to the endpoint is directly queried.
3. The conflict-based multi-vehicle path planning optimization method according to claim 1, characterized in that, In step S6, if the vehicle's current path differs from the desired shortest path, then the following steps are executed: S601. Divide a set of feasible paths into fixed time_steps in the time dimension to form a discretized spatiotemporal three-dimensional graph. S602. Determine if there are any paths that need optimization. If yes, proceed to step S603; otherwise, end. S603, Select any vehicle that needs optimization; S604. Without changing the spatiotemporal paths of other vehicles, plan a path on the spatiotemporal three-dimensional map. S605. Determine if there is a better path. If there is, update the path. If not, return to step S602.