Multi-agent path cooperative planning method and system in vehicle-road cooperative environment

Abstract

This invention provides a multi-agent path collaborative planning method and system in a vehicle-road cooperative environment, relating to the field of intelligent transportation technology. The method includes constructing a global spatiotemporal situation map by acquiring agent states, environmental perception, and road network topology information; extracting motion features to infer targets and time windows; calculating trajectory envelopes to identify overlapping areas and divide conflict domains; marking spatiotemporal channels within the conflict domains according to traffic priority to generate collaborative paths and occupy grids; generating independent paths in non-conflict domains; and verifying and adjusting connection conflicts before issuing the complete path. This method can achieve multi-agent path collaborative planning, avoid path conflicts, and improve traffic efficiency and safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method and system for multi-agent path collaborative planning in a vehicle-road cooperative environment. Background Technology

[0002] Multi-agent path collaborative planning in a vehicle-road cooperative environment is one of the core challenges of intelligent transportation systems. Current technologies typically employ centralized optimization or distributed coordination frameworks. Centralized methods are usually based on a global cost function, using mixed-integer linear programming or dynamic programming to solve for the joint path of all agents, theoretically achieving a globally optimal solution. Distributed methods rely on information exchange between agents, utilizing auction algorithms, negotiation mechanisms, or potential game models to gradually achieve locally or near-globally optimal path allocation. These methods perform reasonably well in static or low-dynamic scenarios, but face significant bottlenecks in the high-frequency, dynamic, and multi-constraint environment of vehicle-road cooperation.

[0003] However, current technologies suffer from drawbacks, including the trade-off between computational complexity and real-time performance. Centralized optimization suffers from exponential growth in the state space as the number of agents increases or the road network expands, resulting in solution times far exceeding the sub-second response windows required in practice, making it difficult to adapt to rapidly changing traffic flows. While distributed coordination reduces the computational load at individual points, it increases the number of iterative communications and is prone to getting trapped in local optima, leading to implicit conflicts in some agent paths—for example, at key nodes such as intersections and merging areas, the spatiotemporal overlap between agents is not adequately resolved, ultimately causing sudden braking, queuing, or traffic congestion. This approach neglects fine-grained alignment in the temporal dimension, and relying solely on spatial conflict detection often misses secondary conflicts caused by inconsistent arrival times of agents.

[0004] The shortcomings also include a lack of spatiotemporal coherence in conflict identification and handling. Most solutions break down path planning into independent sub-problems, first generating independent feasible paths for each agent, and then correcting conflicts through subsequent steps. This serial processing mode often leads to breaks in the connection at the boundary of the conflict domain—for example, when two agents pass through the same area at different time periods, due to limitations of static priorities or simple deadline judgments, subsequent adjustments will break the continuity of the original path, resulting in unpredictable misalignment between the planned independent path and the cooperating area. In addition, existing methods generally underutilize the real-time information provided by roadside sensing devices, and there is a lack of unified spatiotemporal representation between environmental perception and path planning, making it difficult to dynamically update the boundary of the conflict area from a global perspective. This can trigger new chain conflicts when agents change their paths in real time, reducing overall traffic efficiency and system robustness. Summary of the Invention

[0005] This invention provides a method and system for multi-agent path cooperative planning in a vehicle-road cooperative environment, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides a multi-agent path cooperative planning method in a vehicle-road cooperative environment, comprising:

[0007] The system acquires current state information of multiple intelligent agents, environmental perception information collected by roadside sensing devices, and road network topology information, and constructs a global spatiotemporal situation map in the form of a spatiotemporal grid.

[0008] Extract the temporal motion features of multiple agents, infer the target region and expected arrival time window of each agent, calculate the trajectory envelope of each agent, identify the agent combination whose trajectory envelope overlaps in the spatiotemporal dimension, and divide the spatiotemporal region involved in the agent combination into a conflict domain.

[0009] Based on the expected arrival time window and task urgency of each agent in the conflict domain, the passage priority values ​​are calculated and sorted. In order of sorting, the spatiotemporal channels of each agent are marked in the global spatiotemporal situation map. Based on the spatiotemporal channels, a cooperative path is generated for the current agent. The spatiotemporal grid occupied by the cooperative path is marked as occupied to obtain the cooperative path scheme. For agents in non-conflict domains, an independent path is generated based on the global spatiotemporal situation map to obtain the independent path scheme.

[0010] The cooperative path scheme and the independent path scheme are verified. When a connection conflict is detected at the boundary of the conflict domain, the passage priority and passage order of the agents are adjusted to obtain the complete path and send it to the corresponding agents.

[0011] In one optional embodiment, constructing a global spatiotemporal situation map in the form of a spatiotemporal grid includes:

[0012] Obtain road centerline data and road boundary data from road network topology information, perform spatial sampling on the road centerline data to obtain a sampling point sequence, and determine the grid size for spatial discretization based on the sampling point sequence and road boundary data;

[0013] The road network is spatially divided according to the grid size, generating a set of spatial grid units, and a spatial coordinate identifier is assigned to each spatial grid unit;

[0014] The system acquires dynamic and static obstacle location information from environmental perception information, matches the obstacle location information with spatial coordinate identifiers, determines the spatial grid cell occupied by the obstacle, and marks it as an obstacle grid.

[0015] Set the time interval step length, and generate a time slice sequence starting from the current moment according to the time interval step length;

[0016] A Cartesian product operation is performed on the set of spatial grid cells and the sequence of time slices to generate a unique spatiotemporal grid identifier for each spatiotemporal combination, thus constructing a three-dimensional spatiotemporal grid structure.

[0017] The spatial grid cells marked as obstacle grids are expanded on the time slice sequence to generate a set of spatiotemporal grids occupied by obstacles, and marked as impassable in the three-dimensional spatiotemporal grid structure to obtain a global spatiotemporal situation map.

[0018] In one optional embodiment, extracting the temporal motion features of multiple agents and inferring the target region and expected arrival time window for each agent includes:

[0019] The historical trajectory data of multiple intelligent agents is acquired and segmented according to a preset time interval. The position change and velocity change within each time period are extracted to obtain temporal motion features.

[0020] Construct a motion trend vector field, map the positional changes of continuous time periods in the temporal motion features to the motion trend vector field, and generate a motion trend direction distribution map;

[0021] Obtain road network topology information, spatially overlay the motion trend direction distribution map with the road network topology information, identify road network branches that are consistent with the dominant direction in the motion trend direction distribution map, and determine the intended road network branch;

[0022] Road network nodes are extracted along the intended road network branches and spatial clustering is performed to obtain node clusters. The spatial range covered by the node clusters is used as a set of candidate target regions.

[0023] Based on the velocity change, a velocity change trend curve is constructed, and the future velocity sequence is obtained by extending it in the time domain.

[0024] Based on road network topology information, calculate the shortest path distance from the current position of each agent to each candidate target area, combine the future velocity sequence to generate the arrival time distribution, and extract the candidate target area corresponding to the minimum arrival time as the target area.

[0025] The statistical characteristics of future velocity sequences are extracted to obtain the velocity fluctuation amplitude. Based on the velocity fluctuation amplitude and the shortest path distance, the arrival time confidence interval is generated, and the expected arrival time window is determined.

[0026] In one optional embodiment, the trajectory envelopes of each agent are calculated, combinations of agents whose trajectory envelopes overlap in the spatiotemporal dimension are identified, and the spatiotemporal regions involved in the combinations of agents are divided into conflict domains, including:

[0027] Obtain the target region and expected arrival time window for each agent, and generate a candidate trajectory set based on the current position and target region of each agent;

[0028] The curvature distribution of each candidate trajectory in the candidate trajectory set is extracted, and the curvature distribution is converted into a spatial expansion coefficient distribution. Based on the spatial expansion coefficient distribution, the candidate trajectories are non-uniformly expanded to obtain the trajectory spatial envelope.

[0029] The expected arrival time window is discretized to obtain multiple discrete moments. At each discrete moment, the spatial position of the trajectory spatial envelope is mapped to obtain the spatial occupancy status at each discrete moment. The spatial occupancy status at each discrete moment is organized in time order to form the trajectory envelope.

[0030] Mark the trajectory envelopes of each agent in the global spatiotemporal situation map, perform time-by-time spatial overlap detection on the trajectory envelopes of each agent, and identify trajectory envelope pairs that have spatial overlap at at least one discrete time.

[0031] Extract the trajectory envelope pairs and their corresponding agent identifiers and combine them into an agent combination. Obtain the trajectory envelope of each agent in the agent combination. Extract the spatially overlapping discrete moments and spatial regions in the trajectory envelopes and combine the spatial regions with their corresponding discrete moments to form a conflict domain.

[0032] In one optional embodiment, the curvature distribution of each candidate trajectory in the candidate trajectory set is extracted, the curvature distribution is converted into a spatial expansion coefficient distribution, and the candidate trajectories are non-uniformly spatially expanded based on the spatial expansion coefficient distribution to obtain the trajectory spatial envelope, which includes:

[0033] Discretize and sample each candidate trajectory in the candidate trajectory set to obtain a trajectory point sequence, calculate the curvature value of each trajectory point in the trajectory point sequence, and arrange the curvature values ​​of each trajectory point in the trajectory order to obtain the curvature distribution;

[0034] Obtain the physical size parameters of the agent, establish the conversion relationship between curvature values ​​and space occupancy requirements, map the curvature values ​​of each trajectory point in the curvature distribution to the corresponding space occupancy requirements of the trajectory point through the conversion relationship, determine the expansion radius of each trajectory point in combination with the physical size parameters, and arrange the expansion radii of each trajectory point in the trajectory order to obtain the space expansion coefficient distribution;

[0035] Extract the position coordinates of each trajectory point of the candidate trajectory, calculate the tangent vector of the candidate trajectory at each trajectory point, determine the normal direction of each trajectory point, and expand bidirectionally along the normal direction at each trajectory point with the trajectory point as the center according to the expansion radius corresponding to the spatial expansion coefficient distribution to generate the spatial expansion boundary of each trajectory point.

[0036] Connect the spatial extension boundaries of adjacent trajectory points on the candidate trajectory to form a closed envelope boundary, and use the spatial region enclosed by the closed envelope boundary as the trajectory spatial envelope.

[0037] In one optional embodiment, based on the expected arrival time window and task urgency of each agent within the conflict domain, passage priority values ​​are calculated and sorted. Spatiotemporal channels are then marked for each agent in the global spatiotemporal situation map according to the sorting order. A cooperative path is generated for the current agent based on the spatiotemporal channels, and the spatiotemporal grid occupied by the cooperative path is marked as occupied. The resulting cooperative path scheme includes:

[0038] Obtain the expected arrival time window and task urgency of each agent in the conflict domain, extract the time width and time center point of the expected arrival time window, calculate the time flexibility index based on the time width, calculate the time urgency index based on the time interval between the time center point and the current time, combine the time flexibility index and the time urgency index with the task urgency to generate the passage priority value of each agent, and sort the passage priority values ​​of each agent in descending order to obtain the priority ranking sequence.

[0039] Agents are selected sequentially according to priority. Candidate trajectories are extracted for the currently selected agent. The occupancy status of the spatiotemporal grid sequence corresponding to the candidate trajectory is queried in the global spatiotemporal situation map. Unoccupied spatiotemporal grids are selected as spatiotemporal channels. A passage path is determined for the current agent within the spatiotemporal channel. The spatiotemporal grid sequence corresponding to the passage path is marked as occupied, and the identity of the occupying agent and the occupation period are recorded.

[0040] Traverse all agents in the priority sorting sequence to complete the spatiotemporal channel marking, extract the spatiotemporal channel corresponding to each agent, convert the spatiotemporal channel of each agent into the corresponding agent's collaborative path, and use the set of collaborative paths of each agent as the collaborative path scheme.

[0041] In one optional embodiment, verifying the cooperative path scheme and the independent path scheme, and when a connection conflict is detected at the boundary of the conflict domain, adjusting the passage priority and passage order of the agents to obtain the complete path and distribute it to the corresponding agents includes:

[0042] Extract the conflict domain exit time of each agent in the collaborative path scheme and the non-conflict domain entry time of the corresponding agent in the independent path scheme, calculate the time interval difference to determine the spatiotemporal connection error, and determine whether the spatiotemporal connection error exceeds the preset safe spatiotemporal interval threshold.

[0043] When the time limit is exceeded, identify the corresponding agent pair, obtain the expected arrival time window margin and task urgency weight of the agent pair, and recalculate the passage priority adjustment amount.

[0044] The adjustment strategy type is determined based on the priority adjustment amount. When the adjustment strategy type is time adjustment, the non-conflict domain entry time of low-priority agents is adjusted within the expected arrival time window margin, and independent path segments and cooperative path segments are regenerated. When the adjustment strategy type is sequence adjustment, the passage order of agent pairs is swapped and cooperative path segments and independent path segments are regenerated.

[0045] The adjusted collaborative path segments and independent path segments are spliced ​​together in spatiotemporal order to form a complete path. The continuity and uniqueness of the spatiotemporal grid occupancy in the complete path are verified. After the verification is passed, the complete path is sent to the corresponding agent.

[0046] A second aspect of the present invention provides a multi-agent path cooperative planning system in a vehicle-road cooperative environment, comprising:

[0047] The situation construction unit is used to acquire the current state information of multiple intelligent agents, environmental perception information collected by roadside perception devices, and road network topology information, and construct a global spatiotemporal situation map in the form of a spatiotemporal grid.

[0048] The conflict identification unit is used to extract the temporal motion features of multiple agents, infer the target region and expected arrival time window of each agent, calculate the trajectory envelope of each agent, identify the agent combination whose trajectory envelope overlaps in the spatiotemporal dimension, and divide the spatiotemporal region involved in the agent combination into a conflict domain.

[0049] The path generation unit is used to calculate and sort the passage priority values ​​based on the expected arrival time window and task urgency of each agent in the conflict domain. It then marks the spatiotemporal channels for each agent in the global spatiotemporal situation map according to the sorting order, generates a cooperative path for the current agent based on the spatiotemporal channels, marks the spatiotemporal grid occupied by the cooperative path as occupied, and obtains the cooperative path scheme. For agents in non-conflict domains, it generates independent paths based on the global spatiotemporal situation map and obtains the independent path scheme.

[0050] The verification and adjustment unit is used to verify the cooperative path scheme and the independent path scheme. When a connection conflict is detected at the boundary of the conflict domain, the unit adjusts the passage priority and passage order of the agents to obtain the complete path and sends it to the corresponding agents.

[0051] A third aspect of the present invention provides an electronic device, comprising:

[0052] processor;

[0053] Memory used to store processor-executable instructions;

[0054] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0056] In this embodiment of the invention, a global situation map is constructed using a spatiotemporal grid, and combined with temporal motion features and target area reasoning, to accurately identify conflict domains with overlapping trajectory envelopes. This significantly improves the global perception and prediction capabilities for the future motion states of multiple agents. The segmentation of conflict domains decouples path planning from overall to local collaboration, reducing the complexity of simultaneous computation in large-scale systems, fundamentally avoiding frequent global replanning, and effectively suppressing the waste of computational resources. The passage priority values ​​are sorted according to the expected arrival time window and task urgency. Within the conflict domain, spatiotemporal channels are sequentially marked for each agent, and collaborative paths are generated to ensure priority passage for high-urgency tasks while avoiding... Deadlocks and chaos caused by multiple vehicles simultaneously vying for the same spatiotemporal region are addressed. A marking mechanism for occupied spatiotemporal grids enables dynamic resource locking, with subsequent paths automatically avoiding already occupied areas, significantly reducing the probability of secondary conflicts and improving overall path generation efficiency and order. Non-conflict domain agents independently generate paths, processing them in parallel with conflict domain collaborative solutions, ensuring path continuity in conflict-free areas without sacrificing local optimality. The conflict domain boundary verification process independently detects connection conflicts, instantly adjusting passage priorities and order to ensure seamless spatiotemporal connection of global paths, eliminating potential risk points at boundaries, and giving the complete path dynamic robustness and executability. Attached Figure Description

[0057] Figure 1 A flowchart illustrating a multi-agent path collaborative planning method in a vehicle-road cooperative environment;

[0058] Figure 2 Generate a logic flowchart for the trajectory space envelope. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0061] Figure 1This is a flowchart illustrating the multi-agent path cooperative planning method in a vehicle-road cooperative environment according to an embodiment of the present invention.

[0062] Multi-agent path cooperative planning methods in a vehicle-road cooperative environment include:

[0063] The system acquires current state information of multiple intelligent agents, environmental perception information collected by roadside sensing devices, and road network topology information, and constructs a global spatiotemporal situation map in the form of a spatiotemporal grid.

[0064] Extract the temporal motion features of multiple agents, infer the target region and expected arrival time window of each agent, calculate the trajectory envelope of each agent, identify the agent combination whose trajectory envelope overlaps in the spatiotemporal dimension, and divide the spatiotemporal region involved in the agent combination into a conflict domain.

[0065] Based on the expected arrival time window and task urgency of each agent in the conflict domain, the passage priority values ​​are calculated and sorted. In order of sorting, the spatiotemporal channels of each agent are marked in the global spatiotemporal situation map. Based on the spatiotemporal channels, a cooperative path is generated for the current agent. The spatiotemporal grid occupied by the cooperative path is marked as occupied to obtain the cooperative path scheme. For agents in non-conflict domains, an independent path is generated based on the global spatiotemporal situation map to obtain the independent path scheme.

[0066] The cooperative path scheme and the independent path scheme are verified. When a connection conflict is detected at the boundary of the conflict domain, the passage priority and passage order of the agents are adjusted to obtain the complete path and send it to the corresponding agents.

[0067] In one optional embodiment, constructing a global spatiotemporal situation map in the form of a spatiotemporal grid includes:

[0068] Obtain road centerline data and road boundary data from road network topology information, perform spatial sampling on the road centerline data to obtain a sampling point sequence, and determine the grid size for spatial discretization based on the sampling point sequence and road boundary data;

[0069] The road network is spatially divided according to the grid size, generating a set of spatial grid units, and a spatial coordinate identifier is assigned to each spatial grid unit;

[0070] The system acquires dynamic and static obstacle location information from environmental perception information, matches the obstacle location information with spatial coordinate identifiers, determines the spatial grid cell occupied by the obstacle, and marks it as an obstacle grid.

[0071] Set the time interval step length, and generate a time slice sequence starting from the current moment according to the time interval step length;

[0072] A Cartesian product operation is performed on the set of spatial grid cells and the sequence of time slices to generate a unique spatiotemporal grid identifier for each spatiotemporal combination, thus constructing a three-dimensional spatiotemporal grid structure.

[0073] The spatial grid cells marked as obstacle grids are expanded on the time slice sequence to generate a set of spatiotemporal grids occupied by obstacles, and marked as impassable in the three-dimensional spatiotemporal grid structure to obtain a global spatiotemporal situation map.

[0074] In one specific implementation, acquiring road centerline data and road boundary data from the road network topology information is the fundamental input for constructing a global spatiotemporal situation map. Road centerline data is typically stored as a set of polyline segments, describing the road's direction and curvature changes; road boundary data defines the lateral extent of the passable area in the form of polygons or bi-sided polylines. When spatially sampling the road centerline data, the sampling density is adaptively determined based on the road curvature: in curved areas with greater curvature, a smaller sampling interval is used to ensure that the sampling point sequence accurately describes the road shape; in straight sections with less curvature, a larger sampling interval is used to avoid redundant calculations. Let the sampling point sequence be... ,in The total number of sampling points, and each sampling point Includes two-dimensional plane coordinates .

[0075] When determining the grid size for spatial discretization based on sampling point sequences and road boundary data, the average spacing between adjacent sampling points is calculated. And combined with the minimum lane width in the road boundary data The smaller of the two values ​​is taken as the upper limit of the reference grid size. Grid size The determination of the grid size must meet the following conditions: a single grid cell should not exceed half the minimum lane width laterally and should not exceed the average sampling interval longitudinally, to ensure that the grid division has sufficient spatial resolution to distinguish the occupancy status of adjacent lanes on the same road. In actual engineering, A fixed value between 0.5 meters and 2 meters is typically used, and this value is matched to the positioning accuracy of the roadside sensing equipment.

[0076] According to grid size The rectangular bounding box area covered by the road network is spatially divided, and divided along the horizontal axis into... Columns are divided along the vertical axis. Line, generate a set of spatial raster cells For each spatial grid cell Assign spatial coordinate identifiers ,in Indicates column index, Indicates the row index. There is a linear mapping between spatial coordinate identifiers and actual geographic coordinates: raster cells. The geographical coordinates of the center point are ,in and These are the horizontal and vertical coordinates of the bottom left corner of the bounding box. Grid cells outside the road network coverage area are directly marked as impassable and do not participate in subsequent path planning calculations.

[0077] After acquiring the location information of dynamic and static obstacles from the environmental perception information, the actual geographic coordinates of the obstacles are converted into corresponding spatial coordinate identifiers. For static obstacles, all spatial grid cells covered by their outline polygons are marked as obstacle grids; for dynamic obstacles, based on the position and velocity information collected by the roadside sensing device at the current moment, the spatial grid cells currently occupied by the obstacle are marked as obstacle grids. The location information of dynamic obstacles is usually given by the center coordinates and size parameters of the target detection box. The coordinates of the four vertices of the detection box are mapped one by one to the spatial coordinate identifiers, and all grid cells within the coverage area are marked. When the same spatial grid cell is covered by multiple obstacles at the same time, it is still marked only once to avoid duplicate processing.

[0078] Set time away from walk From the current moment Begin, according to Interval generation of time slice sequences ,in This represents the total number of time slices. The value of needs to take into account both the agent's maximum movement speed and the grid size: if the agent's maximum displacement within a single time step exceeds the size of a single grid cell, it may lead to obstacles being encountered during path planning. Therefore, it is required that... ,in This represents the maximum speed limit for all agents in the scenario. In vehicle-to-everything (V2X) scenarios, Typically, the duration is between 0.1 and 0.5 seconds. The time range is determined by the planned time domain length, generally covering a time range of 30 to 120 seconds in the future.

[0079] Set up spatial grid cells With time slice sequence Performing the Cartesian product operation yields the set of spacetime combinations. The total number of its elements is For each spatiotemporal combination Generate a unique spatiotemporal raster identifier Triples can be used It can be used directly as an identifier, or it can be mapped to a one-dimensional integer index through hash encoding for efficient storage and retrieval in memory. The constructed three-dimensional spatiotemporal raster structure can logically be regarded as a three-dimensional array, where the first two dimensions correspond to spatial positions and the third dimension corresponds to time slice indices. Each element stores the accessibility status of the spatiotemporal raster (accessible, inaccessible, or occupied).

[0080] When spatial grid cells marked as obstacle grids are expanded across a time slice sequence, the obstacle grids corresponding to static obstacles are marked as impassable in all time slices; that is, for any... Spacetime Grid All obstacles are set to an impassable state. The spatiotemporal extension of dynamic obstacles requires combining their motion prediction results: based on the velocity vector provided by the roadside sensing equipment, the position of the dynamic obstacle in each time slice is linearly extrapolated to obtain its position in each time slice. The system predicts the occupied spatial grid cells and marks the corresponding spatiotemporal grid cells as impassable. If the trajectory of a dynamic obstacle is uncertain, a certain safety buffer zone can be extended around the predicted location, and the spatiotemporal grid cells within the buffer zone can also be marked as impassable to improve the safety margin of the planning results. After the above marking is completed, each spatiotemporal grid cell in the three-dimensional spatiotemporal grid structure has a clear passage status, forming a global spatiotemporal situation map, which provides a unified spatiotemporal environment description basis for subsequent extraction of agent temporal motion features and division of conflict domains.

[0081] The storage and update mechanism of the global spatiotemporal situation map also requires attention. Since roadside sensing devices continuously collect environmental information, the position and speed of dynamic obstacles change over time. Therefore, the global spatiotemporal situation map needs to support rolling updates: at fixed intervals, data corresponding to past time slices is discarded, and new time slices are appended to the end of the time axis, while the occupancy status of dynamic obstacles is updated based on the latest sensing data. This rolling window mechanism ensures that the global spatiotemporal situation map always reflects the latest environmental state at the current moment, while keeping memory usage within a fixed range, meeting the dual requirements of real-time performance and resource efficiency for vehicle-road cooperative systems.

[0082] In one optional embodiment, extracting the temporal motion features of multiple agents and inferring the target region and expected arrival time window for each agent includes:

[0083] The historical trajectory data of multiple intelligent agents is acquired and segmented according to a preset time interval. The position change and velocity change within each time period are extracted to obtain temporal motion features.

[0084] Construct a motion trend vector field, map the positional changes of continuous time periods in the temporal motion features to the motion trend vector field, and generate a motion trend direction distribution map;

[0085] Obtain road network topology information, spatially overlay the motion trend direction distribution map with the road network topology information, identify road network branches that are consistent with the dominant direction in the motion trend direction distribution map, and determine the intended road network branch;

[0086] Road network nodes are extracted along the intended road network branches and spatial clustering is performed to obtain node clusters. The spatial range covered by the node clusters is used as a set of candidate target regions.

[0087] Based on the velocity change, a velocity change trend curve is constructed, and the future velocity sequence is obtained by extending it in the time domain.

[0088] Based on road network topology information, calculate the shortest path distance from the current position of each agent to each candidate target area, combine the future velocity sequence to generate the arrival time distribution, and extract the candidate target area corresponding to the minimum arrival time as the target area.

[0089] The statistical characteristics of future velocity sequences are extracted to obtain the velocity fluctuation amplitude. Based on the velocity fluctuation amplitude and the shortest path distance, the arrival time confidence interval is generated, and the expected arrival time window is determined.

[0090] In one specific implementation, after acquiring historical trajectory data of multiple agents, the trajectory data is segmented according to a preset time interval. The preset time interval can be flexibly set according to the dynamic level of the scene. For example, a shorter time interval can be used in densely populated intersection areas to capture rapidly changing motion states, while the time interval can be appropriately extended in open road sections to reduce computational overhead. Within each time interval, the position change and velocity change of the agent are extracted: the position change reflects the spatial displacement direction and amplitude of the agent within that time interval, while the velocity change describes the acceleration and deceleration behavior characteristics of the agent within that time interval. The position change and velocity change of each time interval are arranged in chronological order to form a temporal motion feature sequence, which completely depicts the motion evolution process of the agent within the historical time period.

[0091] Based on temporal motion characteristics, a motion trend vector field is constructed. The positional changes over consecutive time periods are sequentially mapped into a two-dimensional spatial vector field. Each time period's positional change corresponds to a direction vector in the vector field, where the direction represents the agent's orientation and the amplitude represents the displacement. Statistical analysis of the direction vectors across all time periods generates a motion trend direction distribution map in the form of a direction histogram. The direction with the highest statistical frequency in the motion trend direction distribution map is the agent's dominant motion direction, reflecting the agent's primary movement intention over historical periods. For cases with multi-peak distributions, the top few peak directions can be extracted as candidate dominant directions, which are then further filtered using road network topology information.

[0092] After acquiring the road network topology information, the motion trend direction distribution map is spatially overlaid with the road network topology information for analysis. The road network topology information includes a structured description of the spatial orientation, connectivity, and node locations of each road segment. During spatial overlay, the angle deviation between the dominant direction in the motion trend direction distribution map and the orientation of each road network branch is calculated. Road network branches with an angle deviation below a preset threshold are identified as candidate branches consistent with the dominant direction. In scenarios with multiple candidate branches, the spatial reachability of each branch to the agent's current position is comprehensively considered to ultimately determine the intended road network branch. The intended road network branch represents the road network structure that the agent is most likely to traverse, providing spatial constraints for subsequent target area inference.

[0093] Road network nodes are extracted along the intended road network branches. These nodes include road segment endpoints, intersection nodes, and key intermediate nodes. Spatial clustering is then performed on the extracted nodes. A density-based clustering algorithm can be used to group nodes with similar spatial locations into the same cluster. The spatial extent covered by each node cluster constitutes a candidate target region, and the set of spatial extents of all node clusters forms the candidate target region set. This method of constructing the candidate target region set ensures a strong correlation between the target region and the road network structure, avoiding the generation of false targets in non-drivable areas.

[0094] Based on the velocity changes, a velocity change trend curve is constructed. The velocity changes for each time period in the historical time-series motion characteristics are arranged chronologically to form a discrete velocity change sequence. This sequence is smoothed to eliminate noise interference, resulting in the velocity change trend curve. Furthermore, the velocity change trend curve is extended in the time domain. Extension methods can include linear extrapolation or autoregression based on historical periodicity to generate predicted velocity values ​​for several future time steps, forming a future velocity sequence. The predicted velocity value at each time step in the future velocity sequence reflects the agent's expected travel speed at the corresponding moment, providing a dynamic basis for subsequent arrival time estimation.

[0095] Based on road network topology information, the shortest path algorithm is used to calculate the shortest path distance from the current position of each agent to each candidate target region. Let the distance from the current position of the agent to the nth... The shortest path distance between each candidate target region is By combining future velocity sequences, the predicted velocity values ​​at each time step are integrated, and the traveled distance is gradually accumulated. When the accumulated distance first reaches a certain value... When the time step is reached, the predicted arrival time of the candidate target region is obtained, thus generating an arrival time distribution. From this distribution, the candidate target region corresponding to the minimum arrival time is extracted; this is the region most likely to be reached by the agent under its current movement intention, and it is designated as the target region. This method prioritizes the shortest arrival time, aligning with the agent's tendency to choose the most efficient path within a road network.

[0096] Statistical characteristics are extracted from the future velocity series, and the standard deviation of the predicted velocity values ​​at each time step in the future velocity series is calculated, which is then used as the velocity fluctuation amplitude. The amplitude of speed fluctuations reflects the degree of uncertainty in future speed predictions. Based on the amplitude of speed fluctuations... Distance from shortest path The arrival time confidence interval is generated as follows: the greater the speed fluctuation, the higher the speed uncertainty during the journey, and the wider the corresponding arrival time confidence interval; the longer the shortest path distance, the more significant the cumulative impact of speed uncertainty on arrival time, and the confidence interval also expands accordingly. Specifically, let the predicted arrival time be... Width of the arrival time confidence interval can be and The ratio relationship is determined, that is ,in This is the confidence interval adjustment coefficient. This serves as a reference speed baseline value, used to normalize the dimensions of speed fluctuations to the time domain. The arrival time confidence interval is expressed as... This interval represents the agent's expected arrival time window. Expressed as a confidence interval, the expected arrival time window better reflects the uncertainties of the actual driving process compared to single-point-of-time estimation, providing a more robust time constraint input for subsequent conflict domain identification and traffic priority calculation.

[0097] In one optional embodiment, the trajectory envelopes of each agent are calculated, combinations of agents whose trajectory envelopes overlap in the spatiotemporal dimension are identified, and the spatiotemporal regions involved in the combinations of agents are divided into conflict domains, including:

[0098] Obtain the target region and expected arrival time window for each agent, and generate a candidate trajectory set based on the current position and target region of each agent;

[0099] The curvature distribution of each candidate trajectory in the candidate trajectory set is extracted, and the curvature distribution is converted into a spatial expansion coefficient distribution. Based on the spatial expansion coefficient distribution, the candidate trajectories are non-uniformly expanded to obtain the trajectory spatial envelope.

[0100] The expected arrival time window is discretized to obtain multiple discrete moments. At each discrete moment, the spatial position of the trajectory spatial envelope is mapped to obtain the spatial occupancy status at each discrete moment. The spatial occupancy status at each discrete moment is organized in time order to form the trajectory envelope.

[0101] Mark the trajectory envelopes of each agent in the global spatiotemporal situation map, perform time-by-time spatial overlap detection on the trajectory envelopes of each agent, and identify trajectory envelope pairs that have spatial overlap at at least one discrete time.

[0102] Extract the trajectory envelope pairs and their corresponding agent identifiers and combine them into an agent combination. Obtain the trajectory envelope of each agent in the agent combination. Extract the spatially overlapping discrete moments and spatial regions in the trajectory envelopes and combine the spatial regions with their corresponding discrete moments to form a conflict domain.

[0103] In one specific implementation, after obtaining the target area and expected arrival time window for each agent, a candidate trajectory set is generated for each agent, taking the agent's current position as the starting point and the target area as the ending point, and combining the known road topology and static obstacle distribution in the global spatiotemporal situation map. During the generation of the candidate trajectory set, road curvature constraints, lane centerline orientation, and the agent's own kinematic limitations are comprehensively considered. Several candidate trajectories that meet basic feasibility requirements are generated through sampling or graph search, ensuring that the candidate trajectory set can cover the main driving possibilities of the agent between a given start and end point, providing sufficient trajectory samples for subsequent trajectory envelope calculation.

[0104] When extracting the curvature distribution of each candidate trajectory in the candidate trajectory set, each candidate trajectory is sampled at equal arc lengths along the trajectory arc length direction. The local curvature value at each sampling point is calculated to obtain the curvature distribution sequence of that trajectory. A larger curvature value indicates a greater turning angle at that position, and a higher risk of lateral deviation for the agent on that path segment; a smaller curvature value indicates that the trajectory approaches a straight line at that position, and a lower risk of lateral deviation. Based on this physical meaning, the curvature distribution is converted into a spatial spread coefficient distribution: let the curvature value at a certain sampling point be... The corresponding spatial expansion factor is ,but and The relationship between them is monotonically increasing; that is, the larger the curvature, the larger the spatial expansion coefficient, resulting in a greater lateral expansion of the trajectory to both sides at that location. This reflects a wider lateral space actually occupied by the agent at the curve. Specifically, the spatial expansion coefficient... It can be represented as ,in This is the basic expansion amount, corresponding to the minimum lateral expansion required for the agent's own vehicle width when traveling in a straight line. This is the curvature sensitivity coefficient, used to adjust the degree of influence of curvature on the lateral expansion.

[0105] When performing non-uniform spatial expansion of candidate trajectories based on the spatial expansion coefficient distribution, the spatial expansion coefficient at each point along the trajectory arc length direction is considered. Extend to both sides of the trajectory normal The corresponding distances form a variable-width strip region with the candidate trajectories as its framework. Since the expansion at each point varies with curvature, the resulting strip region is narrower on straight sections and wider on curved sections, accurately reflecting the differences in lateral space occupancy of the agent on different road segments. The union of the strip regions of all candidate trajectories in the candidate trajectory set yields the trajectory space envelope of the agent, which covers the lateral space occupied by all possible routes the agent can take within the current planning cycle.

[0106] When discretizing the expected arrival time window, the starting time within the planning period is used as the reference, and the time discrete step length is determined according to the already defined time discrete step length in the global spatiotemporal situation diagram. The expected arrival time window is uniformly divided to obtain several discrete time-time sequences. At each discrete time, based on the agent's expected position interval at that time, the trajectory spatial envelope is projected along the time axis to the corresponding spatial position, obtaining the spatial occupancy state at that discrete time. The spatial occupancy state is expressed in the form of a spatiotemporal grid, that is, marking the set of all spatial grid cells covered by the trajectory spatial envelope at that discrete time. The spatial occupancy states of each discrete time are organized sequentially in time to form a complete trajectory envelope, which is essentially a three-dimensional spatiotemporal volume. Each time slice gives the set of spatial grid cells that the agent may occupy, fully describing the spatiotemporal occupancy range of the agent within the planning time domain.

[0107] After marking the trajectory envelopes of each agent in the global spatiotemporal situation map, time-by-time spatial overlap detection is performed on the trajectory envelopes of all agents. For any two agents... and Extract the set of grid cells corresponding to the spatial occupancy state at each discrete time. and Calculate the intersection of the two. If the intersection is not empty, then the agent is considered to be intelligent. With intelligent agents Spatial overlap exists at this discrete moment. By traversing all discrete moments within the planning time domain, if at least one discrete moment results in a non-empty intersection of the spatial occupancy states of two agents, then the agent pair is marked as a trajectory envelope pair with spatial overlap, and all corresponding overlapping discrete moments and overlapping grid sets are recorded. This detection process is performed pairwise on all agents in the scene to ensure that no potential spatiotemporal conflict relationships are overlooked.

[0108] The agent identifiers corresponding to the trajectory envelopes are extracted, and agent identifiers with overlapping trajectory envelopes are combined into agent combinations. For the same agent combination, the complete trajectory envelopes of each agent are obtained, and all discrete moments with spatial overlap and their corresponding overlapping spatial regions are extracted from them. The overlapping spatial region is obtained by taking the union of the intersection grid sets of the spatial occupancy states of the two agents at each overlapping discrete moment, covering the space range that the two agents may jointly occupy during the entire conflict period. The extracted overlapping spatial regions are combined with the corresponding discrete moment sets to form a spatiotemporal description of the conflict domain: the conflict domain corresponds to the region defined by the overlapping grid set in the spatial dimension, and to the time window defined by the overlapping discrete moments in the temporal dimension. Together, they determine the conflict interval in the spatiotemporal coordinate system where the agent combination needs to be coordinated and scheduled.

[0109] The conflict domain partitioning results directly reflect the potential intersection risk range of multiple agents within the planning time domain, providing accurate conflict spatiotemporal boundary information for subsequent priority calculation and spatiotemporal channel marking. By combining non-uniform spatial expansion and temporal discretization, the trajectory envelope can automatically expand the spatial occupancy estimate in high-risk areas such as curves, avoiding missed conflict detections at curves due to the use of uniform expansion. Simultaneously, it maintains a small expansion amount on straight segments to reduce unnecessary false alarms, improving the accuracy and reliability of conflict domain partitioning. For complex intersection scenarios involving multiple agents, by performing overlap detection on all pairwise combinations, multi-directional, multi-time composite conflict domains can be identified, providing a complete conflict topology structure for subsequent multi-agent collaborative path planning.

[0110] like Figure 2 The diagram shown illustrates the logic flowchart for generating the trajectory space envelope.

[0111] In one optional embodiment, the curvature distribution of each candidate trajectory in the candidate trajectory set is extracted, the curvature distribution is converted into a spatial expansion coefficient distribution, and the candidate trajectories are non-uniformly spatially expanded based on the spatial expansion coefficient distribution to obtain the trajectory spatial envelope, which includes:

[0112] Discretize and sample each candidate trajectory in the candidate trajectory set to obtain a trajectory point sequence, calculate the curvature value of each trajectory point in the trajectory point sequence, and arrange the curvature values ​​of each trajectory point in the trajectory order to obtain the curvature distribution;

[0113] Obtain the physical size parameters of the agent, establish the conversion relationship between curvature values ​​and space occupancy requirements, map the curvature values ​​of each trajectory point in the curvature distribution to the corresponding space occupancy requirements of the trajectory point through the conversion relationship, determine the expansion radius of each trajectory point in combination with the physical size parameters, and arrange the expansion radii of each trajectory point in the trajectory order to obtain the space expansion coefficient distribution;

[0114] Extract the position coordinates of each trajectory point of the candidate trajectory, calculate the tangent vector of the candidate trajectory at each trajectory point, determine the normal direction of each trajectory point, and expand bidirectionally along the normal direction at each trajectory point with the trajectory point as the center according to the expansion radius corresponding to the spatial expansion coefficient distribution to generate the spatial expansion boundary of each trajectory point.

[0115] Connect the spatial extension boundaries of adjacent trajectory points on the candidate trajectory to form a closed envelope boundary, and use the spatial region enclosed by the closed envelope boundary as the trajectory spatial envelope.

[0116] When calculating the spatial envelope of each candidate trajectory in the candidate trajectory set, discretization sampling is first performed on each candidate trajectory, extracting trajectory points at fixed intervals along the trajectory arc length to form a sequence of trajectory points. The selection of the sampling interval needs to balance computational accuracy and efficiency, and is usually set to the order of the grid size to ensure that the resolution of the envelope boundary matches the spatial resolution of the global spatiotemporal situation map. For each trajectory point, the tangent direction is calculated using the coordinate difference between its preceding and following trajectory points, and then the local curvature value is estimated through second-order difference. Specifically, let the... The coordinates of the trajectory points are The curvature at that point is estimated using the three-point difference formula. The curvature values ​​of all trajectory points in the trajectory point sequence Arranging the candidate trajectories in order yields their curvature distribution. The curvature distribution visually reflects the degree of curvature of the trajectory at various local locations. A larger curvature value indicates a sharper turn at that location, and a greater lateral space margin required by the agent at that position.

[0117] Obtain the physical size parameters of the intelligent agent, including vehicle length. Vehicle width and wheelbase These parameters collectively determine the actual space occupancy requirements of the agent under different driving conditions. When establishing the conversion relationship between curvature values ​​and space occupancy requirements, it is considered that when the agent is turning, due to vehicle body deflection and the difference between the inner and outer wheels, its lateral occupancy width will exceed the vehicle body width in a stationary state. Using curvature... For input, combined with wheelbase Calculate the additional lateral offset caused by the difference between the inner and outer wheels during cornering. ,in Follow It increases with the increase of [something]. With vehicle width After stacking, an additional safety margin is added. The space occupancy requirements at each trajectory point are obtained. ,Right now Based on this, Converted to the radius of expansion centered on the trajectory point ,Right now Expand the radius of each trajectory point. Arranging the samples in trajectories sequentially yields the spatial spread coefficient distribution. This distribution exhibits non-uniformity along the trajectory direction: in straight segments, the curvature... Approaching zero, expanding radius Approaching the base value In the curve section, curvature Increase, expand radius This increases the size of the envelope to ensure that it fully covers the area actually occupied by the agent when turning.

[0118] After extracting the position coordinates of each trajectory point in the candidate trajectory, the tangent vector of the candidate trajectory at each trajectory point is calculated. The tangent vector is obtained by normalizing the coordinate differences of adjacent trajectory points: For trajectory points at the endpoints, tangent estimation is performed using one-sided difference. The normal direction of each trajectory point is then determined. When, the tangent vector Rotate 90 degrees to get the result. ,in and These are the horizontal and vertical components of the tangent vector, respectively. At each trajectory point, along the normal direction... With this trajectory point Centered on the spatial expansion coefficient distribution, the corresponding expansion radius is... The trajectory point is extended bidirectionally to both sides, generating its spatial extension boundary. The coordinates of the left extension boundary point are... The coordinates of the right-side extended boundary point are: ,in and These are the horizontal and vertical components of the normal vector, respectively. By performing the above bidirectional expansion operation on all trajectory points on the candidate trajectory, a series of spatially expanded boundary point pairs distributed along the entire trajectory can be obtained.

[0119] Connect the spatial extension boundaries of adjacent trajectory points on the candidate trajectory to form a closed envelope boundary. Specifically, connect adjacent trajectory points... and The left extension boundary points are connected sequentially to form the left envelope edge curve; the right extension boundary points of adjacent trajectory points are connected sequentially to form the right envelope edge curve; at the trajectory start point, the left and right extension boundary points are connected by straight line segments, and at the trajectory end point, the left and right extension boundary points are also connected by straight line segments, thus splicing the left edge curve, the right edge curve, and the closed line segments at both ends into a complete closed envelope boundary. The spatial region enclosed by the closed envelope boundary is taken as the trajectory spatial envelope of the candidate trajectory. Due to the extension radius at each trajectory point... With curvature The change is not uniform, and the resulting trajectory space envelope is wider at curves and narrower on straight sections, which can accurately reflect the actual space occupied by the agent at different positions when it travels along the candidate trajectory.

[0120] After processing all candidate trajectories in the candidate trajectory set as described above, each candidate trajectory corresponds to a trajectory spatial envelope. These trajectory spatial envelopes are further projected onto the spatial grid layer of the global spatiotemporal situation map. By determining which spatial grid cells are covered inside the closed envelope boundary, the covered grid cells are recorded as the spatial occupancy set of the candidate trajectory at the corresponding time, providing an accurate spatial occupancy description for subsequent conflict domain identification and passage priority calculation. Compared to simply uniformly expanding trajectory points with a fixed radius, the non-uniform spatial expansion method based on curvature distribution can avoid excessive occupation of spatiotemporal resources on straight segments while ensuring a safety margin, thereby reserving usable spatiotemporal channels for more agents and improving the overall passage efficiency of path collaborative planning.

[0121] In one optional embodiment, based on the expected arrival time window and task urgency of each agent within the conflict domain, passage priority values ​​are calculated and sorted. Spatiotemporal channels are then marked for each agent in the global spatiotemporal situation map according to the sorting order. A cooperative path is generated for the current agent based on the spatiotemporal channels, and the spatiotemporal grid occupied by the cooperative path is marked as occupied. The resulting cooperative path scheme includes:

[0122] Obtain the expected arrival time window and task urgency of each agent in the conflict domain, extract the time width and time center point of the expected arrival time window, calculate the time flexibility index based on the time width, calculate the time urgency index based on the time interval between the time center point and the current time, combine the time flexibility index and the time urgency index with the task urgency to generate the passage priority value of each agent, and sort the passage priority values ​​of each agent in descending order to obtain the priority ranking sequence.

[0123] Agents are selected sequentially according to priority. Candidate trajectories are extracted for the currently selected agent. The occupancy status of the spatiotemporal grid sequence corresponding to the candidate trajectory is queried in the global spatiotemporal situation map. Unoccupied spatiotemporal grids are selected as spatiotemporal channels. A passage path is determined for the current agent within the spatiotemporal channel. The spatiotemporal grid sequence corresponding to the passage path is marked as occupied, and the identity of the occupying agent and the occupation period are recorded.

[0124] Traverse all agents in the priority sorting sequence to complete the spatiotemporal channel marking, extract the spatiotemporal channel corresponding to each agent, convert the spatiotemporal channel of each agent into the corresponding agent's collaborative path, and use the set of collaborative paths of each agent as the collaborative path scheme.

[0125] In one specific implementation, after confirming the conflict domain partitioning results, it is necessary to quantitatively evaluate the passage order of each agent within the conflict domain to ensure that high-priority agents obtain spatiotemporal resources first. To this end, the expected arrival time window for each agent is extracted, its start and end times are obtained, and the difference between the two is calculated to obtain the time width. A smaller time width indicates a more stringent constraint on the arrival time of the agent, and can be used to characterize its time flexibility. The time flexibility index... Defined as the reciprocal of the time width, i.e. ,in This is to prevent the introduction of extremely small positive numbers with a denominator of zero. The larger the value, the more stringent the requirements of the agent for passage time, and the higher the response weight should be given in resource allocation.

[0126] At the same time, calculate the center point of the expected arrival time window. That is, the average of the starting and ending points of the time, then calculate. With the current moment The time interval between The shorter the time interval, the closer the agent is to the expected arrival time, and the higher the time urgency. Time Urgency Index Defined as This value increases as the time interval shortens, and can dynamically reflect the current time pressure state of the agent.

[0127] After obtaining the time flexibility index and the time urgency index, combine them with the pre-marked task urgency level. (The values ​​are normalized continuous values; the larger the value, the more urgent the task.) Calculate the passage priority value for each agent using a weighted fusion method. The calculation method is as follows ,in , , These are the weighting coefficients for the three indicators, satisfying... The weighting coefficients can be pre-configured according to the actual scenario requirements. The weights are calculated for each agent within the conflict domain. Then, the priorities are sorted in descending order to obtain the priority sorting sequence. The agents ranked higher in the sequence will be the first to obtain the right to allocate spatiotemporal channels.

[0128] Agents are selected sequentially according to priority for spatiotemporal channel allocation. For each selected agent, a candidate trajectory set is extracted. These candidate trajectories are generated from the previous trajectory envelope calculation process and contain several spatially feasible alternative paths. For each candidate trajectory, it is discretized along the time axis into a series of spatiotemporal grid coordinate sequences, with each spatiotemporal grid indexed by its spatial location. With time slice index Common identifier. The occupancy status of these spatiotemporal grids is queried one by one in the global spatiotemporal situation map. Continuous spatiotemporal grid segments with an "unoccupied" status are extracted as available spatiotemporal channels. If a candidate trajectory's corresponding spatiotemporal grid sequence contains grids already occupied by other agents, the candidate trajectory is determined to be infeasible, and the next candidate trajectory is evaluated.

[0129] After selecting available spatiotemporal channels, a travel path is determined for the current agent within each channel. The path determination comprehensively considers factors such as path length, time cost, and trajectory smoothness, selecting the path sequence with the lowest overall cost from the available spatiotemporal channels as the agent's final travel path. Once the travel path is determined, all corresponding spatiotemporal grids are marked as occupied in the global spatiotemporal situation map, and the agent's identity is recorded. and the time period occupied This is so that the agent can accurately identify the ownership of each spatiotemporal grid during subsequent queries, avoiding conflicts caused by duplicate allocation.

[0130] After performing the aforementioned spatiotemporal channel marking operation on all agents in the priority ranking sequence, each agent corresponds to a non-overlapping travel path in the spatiotemporal dimension. After traversal, the spatiotemporal channels corresponding to each agent are extracted, and the spatial coordinate sequences and time slice sequences within these channels are reconstructed into continuous cooperative paths. These cooperative paths are expressed as a series of timestamped spatial coordinate points, fully describing the agent's trajectory on the time axis. The cooperative paths of all agents are then aggregated to form a cooperative path scheme. This scheme ensures the spatiotemporal non-intersection of paths between agents within the conflict domain at the global spatiotemporal situation map level.

[0131] Furthermore, for agents identified as being within non-conflict domains, their trajectory envelopes do not overlap with those of other agents in the spatiotemporal dimension, thus eliminating the need for priority ranking and spatiotemporal channel competition. Independent path planning is directly performed for these agents in the global spatiotemporal situation map. The map queries the distribution of occupied grid cells to avoid spatiotemporal resources already occupied by collaborative path schemes. Within the remaining available spatiotemporal grid cells, a spatiotemporal search algorithm generates an independent path from the agent's current position to the target area. Independent paths are also expressed as timestamped spatial coordinate sequences, and the occupancy status of the corresponding spatiotemporal grid cells is marked in the global spatiotemporal situation map to ensure consistency between independent path schemes and collaborative path schemes at the overall spatiotemporal resource level, preventing new conflicts. After generating independent paths for all agents in non-conflict domains, these independent paths are aggregated to form an independent path scheme, which, together with the collaborative path scheme, constitutes the complete initial path set for this planning cycle, providing input for subsequent conflict verification and path distribution.

[0132] In one optional embodiment, verifying the cooperative path scheme and the independent path scheme, and when a connection conflict is detected at the boundary of the conflict domain, adjusting the passage priority and passage order of the agents to obtain the complete path and distribute it to the corresponding agents includes:

[0133] Extract the conflict domain exit time of each agent in the collaborative path scheme and the non-conflict domain entry time of the corresponding agent in the independent path scheme, calculate the time interval difference to determine the spatiotemporal connection error, and determine whether the spatiotemporal connection error exceeds the preset safe spatiotemporal interval threshold.

[0134] When the time limit is exceeded, identify the corresponding agent pair, obtain the expected arrival time window margin and task urgency weight of the agent pair, and recalculate the passage priority adjustment amount.

[0135] The adjustment strategy type is determined based on the traffic priority adjustment amount. When the adjustment strategy type is time adjustment, the non-conflict domain entry time of low-priority agents is adjusted within the expected arrival time window margin, and independent path segments and cooperative path segments are regenerated.

[0136] When the adjustment strategy type is sequential adjustment, the passage order of the agent pairs is swapped and the cooperative path segments and independent path segments are regenerated;

[0137] The adjusted collaborative path segments and independent path segments are spliced ​​together in spatiotemporal order to form a complete path. The continuity and uniqueness of the spatiotemporal grid occupancy in the complete path are verified. After the verification is passed, the complete path is sent to the corresponding agent.

[0138] In one specific implementation, after generating the cooperative path scheme and the independent path scheme, it is necessary to systematically verify the spatiotemporal connection state of the two schemes at the boundary of the conflict domain. Specifically, the exit time of each agent leaving the conflict domain in the cooperative path scheme is extracted, and the entry time of the corresponding agent entering the non-conflict domain in the independent path scheme is extracted. The time interval difference between the two is calculated as a measure of the spatiotemporal connection error. If there is a time jump or overlap between the end time of a certain agent's cooperative path segment and the start time of the independent path segment, it indicates that there is a connection conflict between the two paths at the boundary of the conflict domain. Let the agent... The conflict domain exit time in the cooperative path scheme is In the independent path scheme, the entry time of the non-conflict domain is Then the spatiotemporal connection error Defined as .when When the value is negative, it indicates that the two paths overlap on the timeline, resulting in a time conflict; when... Exceeding the preset safe time interval threshold This indicates that the time interval between two path segments is too large, which may cause the agent to wait or become discontinuous near the boundary of the conflict domain. As long as Exceeding This means that the intelligent agent is determined to have a connection conflict and needs to enter the subsequent adjustment process.

[0139] After identifying agents with conflicting connections, the agents are paired to form agent pairs. The expected arrival time window margin and task urgency weight of each member in each pair are then obtained. The expected arrival time window margin reflects the agent's adjustable space regarding the target arrival time; a larger margin indicates more flexible scheduling, allowing the agent to move its path execution time forward or backward within a certain range without affecting task completion. The task urgency weight is derived from the task urgency values ​​assigned to each agent during the upstream priority calculation phase and is reused in the adjustment phase to ensure consistency between the adjustment decision and the overall priority system. Based on these two parameters, the passage priority adjustment amount is recalculated. Its calculation method comprehensively considers the agent's current time window margin. Weighted by task urgency Specifically expressed as ,in To prevent the introduction of extremely small positive numbers with a denominator of zero, and the already used... They have the same meaning but participate in computation independently at this stage. This is achieved by comparing the two members in the agent pair. The numerical values ​​determine their relative priorities, providing a basis for selecting subsequent adjustment strategy types.

[0140] Based on the calculation results of the priority adjustment amount, it is determined whether to adopt a time adjustment strategy or a sequence adjustment strategy. When the priority difference between the two members in an agent pair is small, and the expected arrival time window margin of the lower-priority agent is sufficient to cover the connection error, the time adjustment strategy is selected. Under the time adjustment strategy, within the expected arrival time window margin of the lower-priority agent, its non-conflict domain entry time is adjusted. Shift backward to ensure the adjusted entry time meets the requirements. Simultaneously, it ensures that the adjusted entry time does not exceed the upper bound of the agent's expected arrival time window. Based on the adjusted entry time, the agent's independent path segment is regenerated, achieving temporal alignment by adjusting the travel speed profile while maintaining the spatial trajectory. If the end time of the cooperative path segment of a low-priority agent also needs to be adjusted accordingly, the cooperative path segment is regenerated synchronously, ensuring that the interval between the exit time of the cooperative path segment and the entry time of the independent path segment strictly falls within the safe time-space interval threshold.

[0141] When the priority difference between two members in an agent pair is significant, or the time window margin of the lower-priority agent is insufficient to eliminate the connection error through simple time shifting, a sequence adjustment strategy is adopted. The core operation of the sequence adjustment strategy is to swap the passage order of the agent pair, that is, to adjust the agent that was originally ranked later to pass through the conflict domain first, and to change the agent that was originally ranked earlier to pass later. After swapping the passage order, it is necessary to regenerate the cooperative path segments and independent path segments for each of the two agents. During regeneration, the spatiotemporal channels are remarked in the global spatiotemporal situation map according to the new passage order. First, the spatiotemporal channel is allocated to the agent with the higher priority after adjustment and a cooperative path segment is generated. Then, the corresponding cooperative path segment and independent path segment are generated for the other agent in the remaining available spatiotemporal resources. During the regeneration process, the occupancy marking of the spatiotemporal grid is updated synchronously to avoid new spatiotemporal conflicts between newly generated path segments and existing path segments.

[0142] After time or sequence adjustment, the adjusted cooperative path segments and independent path segments are spliced ​​together in spatiotemporal order to form the complete path for each agent. The splicing is performed with the time axis as the primary order, ensuring that the final spatiotemporal state of the cooperative path segment and the initial spatiotemporal state of the independent path segment remain continuous in position and velocity direction, without spatial jumps or abrupt changes in velocity direction. After splicing, two checks are performed on the complete path: continuity check and uniqueness check of spatiotemporal grid occupancy. Continuity check requires that the spatiotemporal grid sequence traversed by the complete path has no gaps on the time axis, i.e., there is no situation where the spatial grid corresponding to an agent is empty at any given moment. Uniqueness check requires that the same spatiotemporal grid is occupied by only one agent at the same moment, and there is no situation where two or more agents simultaneously occupy the same spatiotemporal grid. If either continuity or uniqueness check fails, the process returns to the adjustment strategy selection stage, re-evaluates the adjustment direction, and generates new path segments until both checks pass.

[0143] After successful verification, the complete path for each agent is issued to the corresponding agent in the form of structured instructions. The issued instructions include the spatiotemporal grid sequence of the complete path, the target position coordinates at each moment, the desired speed, and the start and end times of path execution. Upon receiving the complete path, the agent executes the driving task according to the spatiotemporal constraints specified in the path, thereby achieving spatiotemporally conflict-free collaborative passage between multiple agents in a vehicle-road cooperative environment. The entire verification and adjustment process ensures a smooth transition between the collaborative path scheme and the independent path scheme at the boundary of the conflict domain, eliminating the boundary inconsistency problem introduced by the independent generation of the two types of schemes, and guaranteeing the global consistency and executability of the final complete path in the spatiotemporal dimension.

[0144] A second aspect of the present invention provides a multi-agent path cooperative planning system in a vehicle-road cooperative environment, comprising:

[0145] The situation construction unit is used to acquire the current state information of multiple intelligent agents, environmental perception information collected by roadside perception devices, and road network topology information, and construct a global spatiotemporal situation map in the form of a spatiotemporal grid.

[0146] The conflict identification unit is used to extract the temporal motion features of multiple agents, infer the target region and expected arrival time window of each agent, calculate the trajectory envelope of each agent, identify the agent combination whose trajectory envelope overlaps in the spatiotemporal dimension, and divide the spatiotemporal region involved in the agent combination into a conflict domain.

[0147] The path generation unit is used to calculate and sort the passage priority values ​​based on the expected arrival time window and task urgency of each agent in the conflict domain. It then marks the spatiotemporal channels for each agent in the global spatiotemporal situation map according to the sorting order, generates a cooperative path for the current agent based on the spatiotemporal channels, marks the spatiotemporal grid occupied by the cooperative path as occupied, and obtains the cooperative path scheme. For agents in non-conflict domains, it generates independent paths based on the global spatiotemporal situation map and obtains the independent path scheme.

[0148] The verification and adjustment unit is used to verify the cooperative path scheme and the independent path scheme. When a connection conflict is detected at the boundary of the conflict domain, the unit adjusts the passage priority and passage order of the agents to obtain the complete path and sends it to the corresponding agents.

[0149] A third aspect of the present invention provides an electronic device, comprising:

[0150] processor;

[0151] Memory used to store processor-executable instructions;

[0152] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0153] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0154] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-agent path cooperative planning method in a vehicle-road cooperative environment, characterized in that, include: The system acquires current state information of multiple intelligent agents, environmental perception information collected by roadside sensing devices, and road network topology information, and constructs a global spatiotemporal situation map in the form of a spatiotemporal grid. Extract the temporal motion features of multiple agents, infer the target region and expected arrival time window of each agent, calculate the trajectory envelope of each agent, identify the agent combination whose trajectory envelope overlaps in the spatiotemporal dimension, and divide the spatiotemporal region involved in the agent combination into a conflict domain. Based on the expected arrival time window and task urgency of each agent in the conflict domain, the passage priority values ​​are calculated and sorted. In order of sorting, the spatiotemporal channels of each agent are marked in the global spatiotemporal situation map. Based on the spatiotemporal channels, a cooperative path is generated for the current agent. The spatiotemporal grid occupied by the cooperative path is marked as occupied to obtain the cooperative path scheme. For agents in non-conflict domains, an independent path is generated based on the global spatiotemporal situation map to obtain the independent path scheme. The cooperative path scheme and the independent path scheme are verified. When a connection conflict is detected at the boundary of the conflict domain, the passage priority and passage order of the agents are adjusted to obtain the complete path and send it to the corresponding agents.

2. The method according to claim 1, characterized in that, Constructing a global spatiotemporal situation map in the form of a spatiotemporal grid includes: Obtain road centerline data and road boundary data from road network topology information, perform spatial sampling on the road centerline data to obtain a sampling point sequence, and determine the grid size for spatial discretization based on the sampling point sequence and road boundary data; The road network is spatially divided according to the grid size, generating a set of spatial grid units, and a spatial coordinate identifier is assigned to each spatial grid unit; The system acquires dynamic and static obstacle location information from environmental perception information, matches the obstacle location information with spatial coordinate identifiers, determines the spatial grid cell occupied by the obstacle, and marks it as an obstacle grid. Set the time interval step length, and generate a time slice sequence starting from the current moment according to the time interval step length; A Cartesian product operation is performed on the set of spatial grid cells and the sequence of time slices to generate a unique spatiotemporal grid identifier for each spatiotemporal combination, thus constructing a three-dimensional spatiotemporal grid structure. The spatial grid cells marked as obstacle grids are expanded on the time slice sequence to generate a set of spatiotemporal grids occupied by obstacles, and marked as impassable in the three-dimensional spatiotemporal grid structure to obtain a global spatiotemporal situation map.

3. The method according to claim 1, characterized in that, Extracting the temporal motion features of multiple agents and inferring the target region and expected arrival time window for each agent includes: The historical trajectory data of multiple intelligent agents is acquired and segmented according to a preset time interval. The position change and velocity change within each time period are extracted to obtain temporal motion features. Construct a motion trend vector field, map the positional changes of continuous time periods in the temporal motion features to the motion trend vector field, and generate a motion trend direction distribution map; Obtain road network topology information, spatially overlay the motion trend direction distribution map with the road network topology information, identify road network branches that are consistent with the dominant direction in the motion trend direction distribution map, and determine the intended road network branch; Road network nodes are extracted along the intended road network branches and spatial clustering is performed to obtain node clusters. The spatial range covered by the node clusters is used as a set of candidate target regions. Based on the velocity change, a velocity change trend curve is constructed, and the future velocity sequence is obtained by extending it in the time domain. Based on road network topology information, calculate the shortest path distance from the current position of each agent to each candidate target area, combine the future velocity sequence to generate the arrival time distribution, and extract the candidate target area corresponding to the minimum arrival time as the target area. The statistical characteristics of future velocity sequences are extracted to obtain the velocity fluctuation amplitude. Based on the velocity fluctuation amplitude and the shortest path distance, the arrival time confidence interval is generated, and the expected arrival time window is determined.

4. The method according to claim 1, characterized in that, Calculate the trajectory envelope of each agent, identify agent combinations whose trajectory envelopes overlap in the spatiotemporal dimensions, and divide the spatiotemporal regions involved in the agent combinations into conflict domains, including: Obtain the target region and expected arrival time window for each agent, and generate a candidate trajectory set based on the current position and target region of each agent; The curvature distribution of each candidate trajectory in the candidate trajectory set is extracted, and the curvature distribution is converted into a spatial expansion coefficient distribution. Based on the spatial expansion coefficient distribution, the candidate trajectories are non-uniformly expanded to obtain the trajectory spatial envelope. The expected arrival time window is discretized to obtain multiple discrete moments. At each discrete moment, the spatial position of the trajectory spatial envelope is mapped to obtain the spatial occupancy status at each discrete moment. The spatial occupancy status at each discrete moment is organized in time order to form the trajectory envelope. Mark the trajectory envelopes of each agent in the global spatiotemporal situation map, perform time-by-time spatial overlap detection on the trajectory envelopes of each agent, and identify trajectory envelope pairs that have spatial overlap at at least one discrete time. Extract the trajectory envelope pairs and their corresponding agent identifiers and combine them into an agent combination. Obtain the trajectory envelope of each agent in the agent combination. Extract the spatially overlapping discrete moments and spatial regions in the trajectory envelopes and combine the spatial regions with their corresponding discrete moments to form a conflict domain.

5. The method according to claim 4, characterized in that, The curvature distribution of each candidate trajectory in the candidate trajectory set is extracted and transformed into a spatial expansion coefficient distribution. Based on the spatial expansion coefficient distribution, the candidate trajectories are non-uniformly expanded to obtain the trajectory spatial envelope, which includes: Discretize and sample each candidate trajectory in the candidate trajectory set to obtain a trajectory point sequence, calculate the curvature value of each trajectory point in the trajectory point sequence, and arrange the curvature values ​​of each trajectory point in the trajectory order to obtain the curvature distribution; Obtain the physical size parameters of the agent, establish the conversion relationship between curvature values ​​and space occupancy requirements, map the curvature values ​​of each trajectory point in the curvature distribution to the corresponding space occupancy requirements of the trajectory point through the conversion relationship, determine the expansion radius of each trajectory point in combination with the physical size parameters, and arrange the expansion radii of each trajectory point in the trajectory order to obtain the space expansion coefficient distribution; Extract the position coordinates of each trajectory point of the candidate trajectory, calculate the tangent vector of the candidate trajectory at each trajectory point, determine the normal direction of each trajectory point, and expand bidirectionally along the normal direction at each trajectory point with the trajectory point as the center according to the expansion radius corresponding to the spatial expansion coefficient distribution to generate the spatial expansion boundary of each trajectory point. Connect the spatial extension boundaries of adjacent trajectory points on the candidate trajectory to form a closed envelope boundary, and use the spatial region enclosed by the closed envelope boundary as the trajectory spatial envelope.

6. The method according to claim 1, characterized in that, Based on the expected arrival time windows and task urgency of each agent within the conflict domain, passage priority values ​​are calculated and sorted. Spatiotemporal channels are then marked for each agent in the global spatiotemporal situation map according to the sorting order. Cooperative paths are generated for the current agent based on these spatiotemporal channels, and the spatiotemporal grids occupied by the cooperative paths are marked as occupied. The resulting cooperative path schemes include: Obtain the expected arrival time window and task urgency of each agent in the conflict domain, extract the time width and time center point of the expected arrival time window, calculate the time flexibility index based on the time width, calculate the time urgency index based on the time interval between the time center point and the current time, combine the time flexibility index and the time urgency index with the task urgency to generate the passage priority value of each agent, and sort the passage priority values ​​of each agent in descending order to obtain the priority ranking sequence. Agents are selected sequentially according to priority. Candidate trajectories are extracted for the currently selected agent. The occupancy status of the spatiotemporal grid sequence corresponding to the candidate trajectory is queried in the global spatiotemporal situation map. Unoccupied spatiotemporal grids are selected as spatiotemporal channels. A passage path is determined for the current agent within the spatiotemporal channel. The spatiotemporal grid sequence corresponding to the passage path is marked as occupied, and the identity of the occupying agent and the occupation period are recorded. Traverse all agents in the priority sorting sequence to complete the spatiotemporal channel marking, extract the spatiotemporal channel corresponding to each agent, convert the spatiotemporal channel of each agent into the corresponding agent's collaborative path, and use the set of collaborative paths of each agent as the collaborative path scheme.

7. The method according to claim 1, characterized in that, Verify the cooperative path scheme and the independent path scheme. When a connection conflict is detected at the boundary of the conflict domain, adjust the passage priority and passage order of the agents to obtain the complete path and send it to the corresponding agents, including: Extract the conflict domain exit time of each agent in the collaborative path scheme and the non-conflict domain entry time of the corresponding agent in the independent path scheme, calculate the time interval difference to determine the spatiotemporal connection error, and determine whether the spatiotemporal connection error exceeds the preset safe spatiotemporal interval threshold. When the time limit is exceeded, identify the corresponding agent pair, obtain the expected arrival time window margin and task urgency weight of the agent pair, and recalculate the passage priority adjustment amount. The adjustment strategy type is determined based on the priority adjustment amount. When the adjustment strategy type is time adjustment, the non-conflict domain entry time of low-priority agents is adjusted within the expected arrival time window margin, and independent path segments and cooperative path segments are regenerated. When the adjustment strategy type is sequence adjustment, the passage order of agent pairs is swapped and cooperative path segments and independent path segments are regenerated. The adjusted collaborative path segments and independent path segments are spliced ​​together in spatiotemporal order to form a complete path. The continuity and uniqueness of the spatiotemporal grid occupancy in the complete path are verified. After the verification is passed, the complete path is sent to the corresponding agent.

8. A multi-agent path cooperative planning system in a vehicle-road cooperative environment, used to implement the method as described in any one of claims 1-7, characterized in that, include: The situation construction unit is used to acquire the current state information of multiple intelligent agents, environmental perception information collected by roadside perception devices, and road network topology information, and construct a global spatiotemporal situation map in the form of a spatiotemporal grid. The conflict identification unit is used to extract the temporal motion features of multiple agents, infer the target region and expected arrival time window of each agent, calculate the trajectory envelope of each agent, identify the agent combination whose trajectory envelope overlaps in the spatiotemporal dimension, and divide the spatiotemporal region involved in the agent combination into a conflict domain. The path generation unit is used to calculate and sort the passage priority values ​​based on the expected arrival time window and task urgency of each agent in the conflict domain. It then marks the spatiotemporal channels for each agent in the global spatiotemporal situation map according to the sorting order, generates a cooperative path for the current agent based on the spatiotemporal channels, marks the spatiotemporal grid occupied by the cooperative path as occupied, and obtains the cooperative path scheme. For agents in non-conflict domains, it generates independent paths based on the global spatiotemporal situation map and obtains the independent path scheme. The verification and adjustment unit is used to verify the cooperative path scheme and the independent path scheme. When a connection conflict is detected at the boundary of the conflict domain, the unit adjusts the passage priority and passage order of the agents to obtain the complete path and sends it to the corresponding agents.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images