A multi-agent based parking path planning method

Abstract

The application discloses a kind of parking path planning methods based on multi-agent, it is related to intelligent parking path planning technical field, including step one: constructing parking lot space-time road network model;Step two: constructing parking multi-agent system, and generating target vehicle cooperative state;Step three: through the generation of candidate topological path set of improved type Yen algorithm;Step four: the continuous curvature parameter solving method based on cycloid, the steering connection parameter of each candidate topological path is calculated;Step five: through vehicle sweep envelope correction mechanism to generate candidate parking path set;Step six: each candidate parking path is time parameterization and space-time traffic deduction;Step seven: each candidate parking path is space-time conflict coordination, and target parking path is obtained;Step eight: target parking path is rolling correction.The application improves the dynamic environment adaptability of parking path planning by improved type Yen algorithm.

Description

Technical Field

[0001] This invention relates to the field of intelligent parking path planning technology, and in particular to a parking path planning method based on multiple agents. Background Technology

[0002] With the continuous expansion of parking lots in large commercial complexes, transportation hubs, and residential parks, automated valet parking and intelligent parking assistance technologies have received widespread attention. Existing parking path planning methods mainly rely on static parking lot maps, the vehicle's current location, and target parking space information, using shortest path search or local obstacle avoidance to generate driving paths. However, in practical applications, the following problems commonly exist: The occupancy status of vehicles, pedestrians, and passageways in parking lots exhibits significant dynamic changes. Traditional methods, typically based on independent planning for each vehicle, struggle to reflect the movement trends of adjacent vehicles, intersection traffic sequences, and available time periods in narrow passages, easily leading to spatiotemporal conflicts, prolonged waiting times, or localized congestion. Existing methods often separate global path search at the road level from geometric path generation at the vehicle level. Paths formed by direct connections between discrete road nodes are prone to abrupt turning, curvature discontinuities, and turning radii that do not conform to vehicle kinematic constraints. This is especially problematic in parking lots with complex road boundaries, dense obstacles, and narrow passageways, where the actual driving area of ​​vehicles may exceed the permitted range of the planned path. Furthermore, some methods select parking paths solely based on path length or travel time, failing to comprehensively consider dynamic conflicts, resource occupancy, waiting times, and vehicle maneuverability. When the parking lot environment changes or vehicles experience path tracking deviations, it is difficult to effectively adjust unexecuted paths in a timely manner, impacting parking safety, traffic efficiency, and path execution stability.

[0003] Therefore, how to provide a parking path planning method based on multi-agent systems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a parking path planning method based on multi-agent systems. This invention fully utilizes multi-agent collaboration, an improved Yen algorithm, continuous curvature parameter solving of spiral curves, and vehicle sweep envelope correction technology to achieve candidate topology path search, continuous curvature path reconstruction, spatiotemporal traffic simulation, conflict coordination, path optimization, and rolling correction. It has the advantages of high path planning safety, strong vehicle kinematic adaptability, strong spatiotemporal conflict avoidance capability, and good dynamic environment adaptability.

[0005] A parking path planning method based on multiple agents according to an embodiment of the present invention includes the following steps: Step 1: Obtain parking lot environmental data and target waiting vehicle task status data, and construct a parking lot spatiotemporal road network model; Step 2: Based on the task status data of the target vehicles waiting to park and the spatiotemporal road network model of the parking lot, construct a parking multi-agent system, and generate the cooperative state of the target vehicles through the state interaction between multiple agents in the parking multi-agent system; Step 3: Input the cooperative state of the target vehicle into the improved Yen algorithm to perform multi-path search on each directed road segment in the parking lot spatiotemporal road network model and generate a candidate topology path set for the target waiting vehicle; Step 4: Using the continuous curvature parameter solution method based on the spiral curve, calculate the turning connection parameters of each candidate topology path in the candidate topology path set; Step 5: Based on the steering connection parameters, the candidate topology paths are reconstructed using a vehicle sweep envelope correction mechanism to obtain a set of candidate parking paths for the target vehicle. Step 6: Based on the cooperative state of the target vehicle, perform time parameterization and spatiotemporal traffic simulation on each candidate parking path in the candidate parking path set, and output the spatiotemporal traffic evaluation results. Step 7: Based on the spatiotemporal traffic assessment results, the parking multi-agent system coordinates spatiotemporal conflicts and evaluates the cost of path collaboration for each candidate parking path, and selects the target parking path for the target vehicle to be parked. Step 8: Send the target parking path to the target vehicle agent, and the parking multi-agent system will make rolling corrections to the target parking path based on the real-time updated target vehicle collaborative status.

[0006] Optionally, the parking lot environment data includes a parking lot road network map, road boundaries, road traffic directions, fixed obstacle areas, dynamic obstacle status, and traffic resource occupancy information; The parking lot road network map includes road nodes, road sections, intersection areas, narrow passage areas, and parking space entrance locations; The task status data includes the vehicle's current position, heading angle, speed, dimensions, wheelbase, maximum front wheel angle, minimum turning radius, target parking space, target parking position, target parking heading angle, task generation time, and task priority.

[0007] Optionally, the construction of the parking lot spatiotemporal road network model specifically includes: Based on the parking lot road network map and road traffic direction, the directed road segments between road nodes and adjacent road nodes are extracted, and the traffic areas corresponding to the directed road segments, intersection areas, narrow passage areas and parking space entrances are identified as parking lot traffic resources. The parking lot spatial road network is constructed based on the connection relationships between road nodes, directed road segments and adjacent road nodes, and the parking lot spatial road network is extended in time according to discrete time slices to construct a parking lot spatiotemporal road network model. The dynamic obstacle status and access resource occupancy information are mapped to the parking lot spatiotemporal road network model to determine the occupancy and availability status of each parking lot's access resources in each discrete time slice, and these statuses are used as the spatiotemporal state attributes of the corresponding parking lot's access resources.

[0008] Optionally, the parking multi-agent system includes a target vehicle agent, a neighboring vehicle agent, a parking lot access resource agent, and a parking lot environment coordination agent. The target vehicle cooperative status includes the vehicle's own movement status, the target parking space status, the movement status of nearby vehicles, the access resource status of adjacent parking lots, the current waiting status, and the parking task priority.

[0009] Optionally, the improved Yen algorithm specifically includes: Based on the target vehicle's cooperative state, determine the dynamic road segment cost of each directed road segment in the parking lot's spatiotemporal road network model. The dynamic road segment cost is obtained by summing the basic traffic cost, resource occupation cost, vehicle conflict cost, waiting compensation cost, and task priority adjustment cost. The road node corresponding to the current position of the target waiting vehicle is determined as the starting road node, and the entrance position of the target parking space is determined as the target road node. The minimum cumulative cost path search is performed according to the dynamic road segment cost to generate an initial candidate topology path from the starting road node to the target road node. The initial candidate topology path is determined as the first selected topology path; From the road nodes included in the currently selected topology path, filter the road nodes that connect available branch road segments, and identify the filtered road nodes as deviation nodes; The available branch road segment is the subsequent directed road segment that the selected topology path has not passed through and whose corresponding parking lot access resources are available. For each deviation node, the road nodes and directed road segments between the starting road node and the deviation node are determined as the root path, and the subsequent directed road segments that have the same root path as the selected topology path are temporarily hidden. Based on the dynamic road segment cost of each directed road segment, a minimum cumulative cost path search is performed from the deviation node to the target road node to generate a deviation path. The root path is then concatenated with the deviation path to generate a deviation candidate topology path, which is then added to the deviation candidate path pool. Calculate the cumulative dynamic segment cost of each deviation from the candidate topology path, and calculate the path overlap between each deviation from the candidate topology path and the selected topology path using a weighted Jaccard coefficient. In the deviation candidate path pool, deviation candidate topology paths that are duplicates of the selected topology paths, as well as deviation candidate topology paths whose path overlap is greater than or equal to a preset overlap threshold, are removed. The candidate topology path with the lowest cumulative dynamic segment cost is selected from the candidate deviation path pool as the next selected topology path. The process of determining the deviation node, constructing the root path, generating the deviation path, and filtering the candidate topology path is repeated until the number of selected topology paths reaches the preset number of candidate paths. The selected topological paths are used as candidate topological paths, and they are arranged in ascending order of cumulative dynamic road segment cost to obtain the candidate topological path set for the target waiting vehicle.

[0010] Optionally, the method for solving the continuous curvature parameters based on the spiral curve specifically includes: For each candidate topology path, the road centerline of each directed road segment is extracted according to the arrangement order of road nodes and directed road segments, and the road centerlines are connected in sequence to generate the path guide line corresponding to each candidate topology path. Calculate the directional angle between the center lines of adjacent roads in the path guide line, identify road nodes with directional angles greater than a preset turning angle threshold as path turning nodes, and determine the turning direction and turning angle of each path turning node based on the direction of the center lines of adjacent roads. The ratio of the tangent of the maximum front wheel steering angle to the vehicle wheelbase is used as the upper limit of steering curvature, the reciprocal of the minimum turning radius is used as the upper limit of radius curvature, and the smaller value of the upper limit of steering curvature and the upper limit of radius curvature is determined as the maximum allowable curvature of the target parking vehicle. Taking each path turning node as the center, the turning entry point and turning exit point are determined on the center lines of the adjacent roads before and after, and the path range between the turning entry point and the turning exit point is determined as the turning connection interval. The steering angle of each path steering node is used as the target heading change. Candidate steering curvatures are scanned incrementally between zero curvature and maximum allowable curvature. The steering connection length of each candidate steering curvature is calculated, and the smallest candidate steering curvature whose steering connection length does not exceed the length of the corresponding steering connection interval is determined as the target steering curvature. Based on the spiral curve relationship where curvature changes linearly with path arc length, we construct an entry spiral curve where curvature increases from zero to the target turning curvature and an exit spiral curve where curvature decreases from the target turning curvature to zero, and calculate the spiral curve connection parameters corresponding to each path turning node. The spiral curve connection parameters include the entry spiral curve length, the exit spiral curve length, the entry curvature change rate, and the exit curvature change rate. Calculate the heading changes when entering and exiting the spiral line respectively, and subtract them from the target heading changes at the corresponding path turning nodes to obtain the remaining heading changes of the circular arc. Calculate the ratio of the absolute value of the remaining heading change of the circular arc to the absolute value of the target turning curvature to obtain the length of the circular arc connecting segment; By associating the turning entry point, turning exit point, target turning curvature, spiral line connection parameters, and arc connection segment length corresponding to each path turning node, the turning connection parameters of each candidate topology path are obtained.

[0011] Optionally, the vehicle sweep envelope correction mechanism specifically includes: For each candidate topology path, based on the corresponding path guide line and steering connection parameters, the path guide line segments in each steering connection interval are removed, and the entry spiral segment, circular arc connection segment and exit spiral segment are constructed in sequence to generate an initial continuous curvature path. Based on the vehicle size of the target waiting vehicle, calculate the vehicle sweep envelope formed by the target waiting vehicle traveling along the initial continuous curvature path, and determine the minimum clearance between the vehicle sweep envelope and the road boundary and fixed obstacle area. Adjust the lateral positions of the steering entry point, steering exit point, and arc connection segment along the left and right lateral directions of the path guide line to generate multiple candidate correction parameter combinations; Based on each candidate correction parameter combination, the initial continuous curvature path is reconstructed to generate candidate correction paths, and the minimum passage gap and cumulative lateral offset relative to the path guide line of each candidate correction path are calculated. Candidate correction paths with minimum clearance greater than or equal to preset safety clearance are selected, and the candidate correction parameter combination corresponding to the candidate correction path with the smallest cumulative lateral offset is determined as the target correction parameter combination. The initial continuous curvature path is corrected by using a combination of target correction parameters to obtain candidate parking paths, and the candidate parking paths are summarized to obtain a candidate parking path set.

[0012] Optionally, step six specifically includes: Path sampling points are set along each candidate parking path. Based on the path curvature, driving direction and the vehicle motion state in the target vehicle cooperative state, the vehicle position, vehicle heading angle, steering angle, driving speed and expected arrival time of each path sampling point are determined to generate an automatic parking time sequence trajectory. The time-series trajectory of automatic parking is mapped to the spatiotemporal road network model of the parking lot to determine the time period of traffic resource occupation for each candidate parking path; The target vehicle's time-series occupancy area is generated based on the automatic parking time-series trajectory and vehicle size. The target vehicle's time-series occupancy area is then compared with the predicted occupancy areas of neighboring vehicles in terms of time alignment. Combined with the spatiotemporal status attributes of parking lot access resources, the predicted vehicle conflict information and estimated waiting time are determined. The time period of traffic resource occupancy, predicted vehicle conflict information, and estimated waiting time corresponding to each candidate parking path are used as the spatiotemporal traffic assessment results.

[0013] Optionally, step seven specifically includes: Based on the predicted vehicle conflict information of each candidate parking path, the neighboring vehicle intelligent agent determines the neighboring vehicles involved in the conflict, the time interval of the conflict, and the suggested passage order, and generates cooperative response information of neighboring vehicles. The parking lot access resource intelligent agent determines the resource availability time window, candidate waiting positions, and maximum allowable waiting time based on the access resource occupancy time of each candidate parking path and the spatiotemporal status attributes of the parking lot access resources, and generates access resource coordination information. The parking environment coordination agent adjusts the estimated arrival time and waiting time of each candidate parking path based on the cooperative response information of neighboring vehicles, traffic resource coordination information, estimated waiting time and parking task priority, and generates a path traffic coordination scheme. Based on the route coordination scheme, the estimated arrival time, waiting time and predicted vehicle conflict information after coordination of each candidate parking route are re-determined. Remove candidate parking paths from the candidate parking path set that still have predicted vehicle conflicts after coordination, or candidate parking paths whose waiting time after coordination exceeds the maximum allowable waiting time; Calculate the path coordination execution cost of the remaining candidate parking paths in the candidate parking path set, and determine the candidate parking path with the minimum path coordination execution cost as the target parking path; The cost of path coordination is calculated by weighting the cost of coordinated travel time, waiting time, traffic resource occupation, and vehicle operation.

[0014] Optionally, step eight specifically includes: The target vehicle agent receives the target parking path, generates vehicle control commands based on the vehicle position, heading angle, steering angle and driving speed corresponding to the target parking path, and controls the target vehicle to drive along the target parking path. The system acquires the actual movement status of the target waiting vehicle in real time, updates the predicted occupied area of ​​the neighboring vehicle through the neighboring vehicle intelligent agent, and updates the spatiotemporal status attributes of the parking lot access resource through the parking lot access resource intelligent agent, thereby generating the target vehicle cooperative status updated in real time. Based on the actual motion state of the target waiting vehicle and the target parking path, the path tracking deviation is calculated, and based on the real-time updated target vehicle coordination state, predictive vehicle conflict detection and access resource availability judgment are performed on the unexecuted part of the target parking path. When the path tracking deviation exceeds the preset tracking deviation threshold, a predicted vehicle conflict occurs in the part of the path that has not yet been executed, or the corresponding parking lot access resources are unavailable, the current position of the target waiting vehicle is determined as the starting point for rolling correction. The path planning, time parameterization, and spatiotemporal conflict coordination are carried out again for the part of the path that has not yet been executed to generate a corrected target parking path. The target vehicle agent updates the vehicle control commands according to the corrected target parking path until the target vehicle reaches the target parking position and meets the target parking heading angle.

[0015] The beneficial effects of this invention are: This invention constructs a spatiotemporal road network model for parking lots and utilizes target vehicle agents, neighboring vehicle agents, parking lot access resource agents, and parking lot environment coordination agents for state interaction. This forms a target vehicle collaborative state that reflects vehicle status, surrounding traffic status, and access resource status, improving the completeness and real-time performance of path planning information. Furthermore, an improved Yen algorithm comprehensively considers road traffic conditions, resource occupancy, vehicle conflict risk, waiting time, and task priority to search and filter multiple differentiated candidate topologies, improving the rationality of global path selection. Steering connection parameters are calculated using a continuous curvature parameter solution method based on spirals, and the path is corrected using vehicle sweep envelopes, ensuring that the generated candidate parking paths satisfy vehicle kinematic constraints and parking lot spatial constraints, reducing curvature abrupt changes, vehicle boundary crossings, and collision risks. By performing time parameterization and spatiotemporal traffic extrapolation on the candidate parking paths, access resource occupancy is determined, vehicle conflicts are predicted, and estimated waiting times are estimated. Finally, a multi-agent parking system is used to complete conflict coordination and path optimization, improving the safety and efficiency of the parking process. Finally, the target parking path is continuously revised based on the real-time updated cooperative status of the target vehicles, enabling the planning results to adapt to vehicle tracking deviations and changes in the parking lot environment. Therefore, this invention can improve the dynamic adaptability, conflict avoidance capability, and execution stability of parking paths. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a parking path planning method based on multi-agent proposed in this invention; Figure 2This is a flowchart of the improved Yen algorithm in a multi-agent-based parking path planning method proposed in this invention; Figure 3 This is a flowchart of the continuous curvature parameter solution method based on the spiral curve in the parking path planning method based on multi-agent proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A parking path planning method based on multi-agent systems includes the following steps: Step 1: Obtain parking lot environmental data and target waiting vehicle task status data, and construct a parking lot spatiotemporal road network model; Step 2: Based on the task status data of the target vehicles waiting to park and the spatiotemporal road network model of the parking lot, construct a parking multi-agent system, and generate the cooperative state of the target vehicles through the state interaction between multiple agents in the parking multi-agent system; Step 3: Input the cooperative state of the target vehicle into the improved Yen algorithm to perform multi-path search on each directed road segment in the parking lot spatiotemporal road network model and generate a candidate topology path set for the target waiting vehicle; Step 4: Using the continuous curvature parameter solution method based on the spiral curve, calculate the turning connection parameters of each candidate topology path in the candidate topology path set; Step 5: Based on the steering connection parameters, the candidate topology paths are reconstructed using a vehicle sweep envelope correction mechanism to obtain a set of candidate parking paths for the target vehicle. Step 6: Based on the cooperative state of the target vehicle, perform time parameterization and spatiotemporal traffic simulation on each candidate parking path in the candidate parking path set, and output the spatiotemporal traffic evaluation results. Step 7: Based on the spatiotemporal traffic assessment results, the parking multi-agent system coordinates spatiotemporal conflicts and evaluates the cost of path collaboration for each candidate parking path, and selects the target parking path for the target vehicle to be parked. Step 8: Send the target parking path to the target vehicle agent, and the parking multi-agent system will make rolling corrections to the target parking path based on the real-time updated target vehicle collaborative status.

[0019] In this embodiment, the parking lot environment data includes the parking lot road network map, road boundaries, road traffic direction, fixed obstacle area, dynamic obstacle status, and traffic resource occupancy information; The parking lot road network map includes road nodes, road sections, intersection areas, narrow passage areas, and parking space entrance locations; The task status data includes the vehicle's current position, heading angle, speed, dimensions, wheelbase, maximum front wheel angle, minimum turning radius, target parking space, target parking position, target parking heading angle, task generation time, and task priority.

[0020] In this embodiment, the construction of the parking lot spatiotemporal road network model specifically includes: Based on the parking lot road network map and road traffic direction, the directed road segments between road nodes and adjacent road nodes are extracted, and the traffic areas corresponding to the directed road segments, intersection areas, narrow passage areas and parking space entrances are identified as parking lot traffic resources. The parking lot spatial road network is constructed based on the connection relationships between road nodes, directed road segments and adjacent road nodes, and the parking lot spatial road network is extended in time according to discrete time slices to construct a parking lot spatiotemporal road network model. The dynamic obstacle status and access resource occupancy information are mapped to the parking lot spatiotemporal road network model to determine the occupancy and availability status of each parking lot's access resources in each discrete time slice, and these statuses are used as the spatiotemporal state attributes of the corresponding parking lot's access resources.

[0021] In this embodiment, the parking multi-agent system includes a target vehicle agent, a neighboring vehicle agent, a parking lot access resource agent, and a parking lot environment coordination agent. The target vehicle cooperative status includes the vehicle's own movement status, the target parking space status, the movement status of nearby vehicles, the access resource status of adjacent parking lots, the current waiting status, and the parking task priority.

[0022] In this embodiment, the improved Yen algorithm specifically includes: Based on the target vehicle's cooperative state, determine the dynamic road segment cost of each directed road segment in the parking lot's spatiotemporal road network model. The dynamic road segment cost is obtained by summing the basic traffic cost, resource occupation cost, vehicle conflict cost, waiting compensation cost, and task priority adjustment cost. The basic passage cost is the passage cost required for a target waiting vehicle to pass through the corresponding directed road segment, determined based on the length of the directed road segment, the allowed speed, and the turning radius. The resource occupancy cost is the cost incurred when the corresponding parking lot's access resources are occupied or unavailable during the expected passage time, determined based on the number of discrete time slices corresponding to the occupied and unavailable states. The vehicle conflict cost is the cost incurred when a target waiting vehicle encounters a spatiotemporal conflict with a neighboring vehicle while passing through the corresponding directed road segment, determined based on the conflict duration, the predicted minimum distance, and the number of conflicting vehicles. The waiting compensation cost is the cost of adjusting the passage cost of the directed road segment based on the current waiting state of the target waiting vehicle; the longer the current waiting time, the smaller the waiting compensation cost. The task priority adjustment cost is the cost of adjusting the passage cost of the directed road segment based on the parking task priority; the higher the parking task priority, the smaller the task priority adjustment cost. The road node corresponding to the current position of the target waiting vehicle is determined as the starting road node, and the entrance position of the target parking space is determined as the target road node. The minimum cumulative cost path search is performed according to the dynamic road segment cost to generate an initial candidate topology path from the starting road node to the target road node. The initial candidate topology path is determined as the first selected topology path; From the road nodes included in the currently selected topology path, filter the road nodes that connect available branch road segments, and identify the filtered road nodes as deviation nodes; The available branch road segment is the subsequent directed road segment that the selected topology path has not passed through and whose corresponding parking lot access resources are available. For each deviation node, the road nodes and directed road segments between the starting road node and the deviation node are determined as the root path, and the subsequent directed road segments that have the same root path as the selected topology path are temporarily hidden. Based on the dynamic road segment cost of each directed road segment, a minimum cumulative cost path search is performed from the deviation node to the target road node to generate a deviation path. The root path is then concatenated with the deviation path to generate a deviation candidate topology path, which is then added to the deviation candidate path pool. Calculate the cumulative dynamic segment cost of each deviation from the candidate topology path, and calculate the path overlap between each deviation from the candidate topology path and the selected topology path using a weighted Jaccard coefficient. Specifically, the weighted Jaccard coefficient is as follows: for any deviation candidate topology path and any selected topology path, determine the directed road segments commonly included by both and all non-repeating directed road segments included by both. The ratio of the sum of the dynamic road segment costs of each commonly included directed road segment to the sum of the dynamic road segment costs of all non-repeating directed road segments is determined as the path overlap coefficient between the deviation candidate topology path and the selected topology path. The maximum value of the path overlap coefficient is taken as the degree of path overlap of the deviation candidate topology path. In the deviation candidate path pool, deviation candidate topology paths that are duplicates of the selected topology paths, as well as deviation candidate topology paths whose path overlap is greater than or equal to a preset overlap threshold, are removed; whereby the preset overlap threshold is set to 0.70. The candidate topology path with the lowest cumulative dynamic segment cost is selected from the candidate deviation path pool as the next selected topology path. The process of determining the deviation node, constructing the root path, generating the deviation path, and filtering the candidate topology paths is repeated until the number of selected topology paths reaches the preset number of candidate paths. The preset number of candidate paths is set to 3. The selected topological paths are used as candidate topological paths, and they are arranged in ascending order of cumulative dynamic road segment cost to obtain the candidate topological path set for the target waiting vehicle.

[0023] In this invention, the improved Yen algorithm does not change the basic framework of "initial shortest path search - deviation node determination - root path retention - deviation path generation - candidate path pool sorting" compared with the standard Yen algorithm. The improvements are mainly focused on three aspects: road segment cost construction, candidate branch constraints, and path difference screening.

[0024] First, the standard Yen algorithm typically generates the top K acyclic shortest paths in a static weighted graph, using road length or fixed travel time as edge weights. This implementation introduces a parking lot spatiotemporal road network model and the target vehicle's cooperative state into the path search. Based on basic travel costs, resource occupancy costs, vehicle conflict costs, waiting compensation costs, and task priority adjustment costs, the dynamic segment cost of each directed road segment is dynamically determined. Therefore, the path search no longer only considers the shortest distance, but also simultaneously considers the availability of traffic resources, the possibility of conflicts between neighboring vehicles, the waiting time of the target vehicle, and the parking task priority, enabling candidate topologies to adapt to changes in vehicle movement and traffic resource occupancy within the parking lot.

[0025] Secondly, the standard Yen algorithm typically constructs deviation paths sequentially for each node in the selected path, while this implementation only identifies road nodes connecting available branch road segments as deviation nodes, and excludes subsequent directed road segments where the corresponding parking lot access resources are unavailable. This improvement reduces the search for unclimbable branches and invalid deviation nodes, narrows the candidate path search space, and lowers the probability of generating unexecutable paths.

[0026] Furthermore, the standard Yen algorithm primarily selects candidate paths based on cumulative path cost, which may result in significant overlap between the generated paths. This implementation calculates the degree of path overlap between the deviating candidate topology paths and each selected topology path using a weighted Jaccard coefficient, and eliminates paths with an overlap greater than or equal to a preset overlap threshold. Since the weighted Jaccard coefficient uses the dynamic segment cost of each directed road segment as weight, it can effectively identify the reuse of high-conflict, high-occupancy, or high-traffic-cost road segments, thereby improving the differentiation and candidate value of the topology paths.

[0027] Furthermore, this implementation method appropriately favors vehicles with long waiting times and high-priority parking tasks by using waiting compensation costs and task priority adjustment costs, thus avoiding the problem of some vehicles waiting for a long time when selecting based solely on path length.

[0028] Therefore, the improved Yen algorithm can improve the adaptability of candidate topology paths to dynamic parking environments, reduce resource occupation conflicts and vehicle spatiotemporal conflicts, and reduce the number of invalid branch searches while retaining the multi-path search capabilities of the standard Yen algorithm. At the same time, it can improve the diversity, executability and traffic coordination capabilities of candidate paths, and provide a higher quality set of candidate topology paths for subsequent continuous curvature path reconstruction and multi-agent spatiotemporal conflict coordination.

[0029] In this embodiment, the method for solving the continuous curvature parameters based on the spiral curve specifically includes: For each candidate topology path, the road centerline of each directed road segment is extracted according to the arrangement order of road nodes and directed road segments, and the road centerlines are connected in sequence to generate the path guide line corresponding to each candidate topology path. Calculate the directional angle between the centerlines of adjacent roads in the path guide line, identify road nodes with directional angles greater than a preset turning angle threshold as path turning nodes, and determine the turning direction and turning angle of each path turning node according to the direction of the centerlines of adjacent roads; wherein, the preset turning angle threshold is set to 5°; The ratio of the tangent of the maximum front wheel steering angle to the vehicle wheelbase is used as the upper limit of steering curvature, the reciprocal of the minimum turning radius is used as the upper limit of radius curvature, and the smaller value of the upper limit of steering curvature and the upper limit of radius curvature is determined as the maximum allowable curvature of the target parking vehicle. Taking each path turning node as the center, the turning entry point and turning exit point are determined on the center lines of the adjacent roads before and after, and the path range between the turning entry point and the turning exit point is determined as the turning connection interval. The steering angle of each path steering node is used as the target heading change. Candidate steering curvatures are scanned incrementally between zero curvature and maximum allowable curvature. The steering connection length of each candidate steering curvature is calculated, and the smallest candidate steering curvature whose steering connection length does not exceed the length of the corresponding steering connection interval is determined as the target steering curvature. The scanning step size for the candidate steering curvature is set to 0.002m⁻¹; Based on the spiral curve relationship where curvature changes linearly with path arc length, we construct an entry spiral curve where curvature increases from zero to the target turning curvature and an exit spiral curve where curvature decreases from the target turning curvature to zero, and calculate the spiral curve connection parameters corresponding to each path turning node. The spiral curve connection parameters include the entry spiral curve length, the exit spiral curve length, the entry curvature change rate, and the exit curvature change rate. Calculate the heading changes when entering and exiting the spiral line respectively, and subtract them from the target heading changes at the corresponding path turning nodes to obtain the remaining heading changes of the circular arc. Calculate the ratio of the absolute value of the remaining heading change of the circular arc to the absolute value of the target turning curvature to obtain the length of the circular arc connecting segment; By associating the turning entry point, turning exit point, target turning curvature, spiral line connection parameters, and arc connection segment length corresponding to each path turning node, the turning connection parameters of each candidate topology path are obtained.

[0030] In this embodiment, the vehicle sweep envelope correction mechanism specifically includes: For each candidate topology path, based on the corresponding path guide line and steering connection parameters, the path guide line segments in each steering connection interval are removed, and the entry spiral segment, circular arc connection segment and exit spiral segment are constructed in sequence to generate an initial continuous curvature path. Based on the vehicle size of the target waiting vehicle, calculate the vehicle sweep envelope formed by the target waiting vehicle traveling along the initial continuous curvature path, and determine the minimum clearance between the vehicle sweep envelope and the road boundary and fixed obstacle area. Adjust the lateral positions of the steering entry point, steering exit point, and arc connection segment along the left and right lateral directions of the path guide line to generate multiple candidate correction parameter combinations; Based on each candidate correction parameter combination, the initial continuous curvature path is reconstructed to generate candidate correction paths, and the minimum passage gap and cumulative lateral offset relative to the path guide line of each candidate correction path are calculated. Candidate correction paths with a minimum clearance greater than or equal to a preset safety clearance are selected, and the candidate correction parameter combination corresponding to the candidate correction path with the smallest cumulative lateral offset is determined as the target correction parameter combination; wherein, the preset safety clearance is set to 0.35m; The initial continuous curvature path is corrected by using a combination of target correction parameters to obtain candidate parking paths, and the candidate parking paths are summarized to obtain a candidate parking path set.

[0031] In this embodiment, step six specifically includes: Path sampling points are set along each candidate parking path. Based on the path curvature, driving direction and the vehicle motion state in the target vehicle cooperative state, the vehicle position, vehicle heading angle, steering angle, driving speed and expected arrival time of each path sampling point are determined to generate an automatic parking time sequence trajectory. The time-series trajectory of automatic parking is mapped to the spatiotemporal road network model of the parking lot to determine the time period of traffic resource occupation for each candidate parking path; The target vehicle's time-series occupancy area is generated based on the automatic parking time-series trajectory and vehicle size. The target vehicle's time-series occupancy area is then compared with the predicted occupancy areas of neighboring vehicles in terms of time alignment. Combined with the spatiotemporal status attributes of parking lot access resources, the predicted vehicle conflict information and estimated waiting time are determined. The time period of traffic resource occupancy, predicted vehicle conflict information, and estimated waiting time corresponding to each candidate parking path are used as the spatiotemporal traffic assessment results.

[0032] In this embodiment, step seven specifically includes: Based on the predicted vehicle conflict information of each candidate parking path, the neighboring vehicle intelligent agent determines the neighboring vehicles involved in the conflict, the time interval of the conflict, and the suggested passage order, and generates cooperative response information of neighboring vehicles. The parking lot access resource intelligent agent determines the resource availability time window, candidate waiting positions, and maximum allowable waiting time based on the access resource occupancy time of each candidate parking path and the spatiotemporal status attributes of the parking lot access resources, and generates access resource coordination information. The parking environment coordination agent adjusts the estimated arrival time and waiting time of each candidate parking path based on the cooperative response information of neighboring vehicles, traffic resource coordination information, estimated waiting time and parking task priority, and generates a path traffic coordination scheme. Based on the route coordination scheme, the estimated arrival time, waiting time and predicted vehicle conflict information after coordination of each candidate parking route are re-determined. Remove candidate parking paths from the candidate parking path set that still have predicted vehicle conflicts after coordination, or candidate parking paths whose waiting time after coordination exceeds the maximum allowable waiting time; Calculate the path coordination execution cost of the remaining candidate parking paths in the candidate parking path set, and determine the candidate parking path with the minimum path coordination execution cost as the target parking path; The cost of path coordination is calculated by weighting the cost of coordinated travel time, waiting time, traffic resource occupation, and vehicle operation.

[0033] The weights of the coordinated travel time cost, waiting time cost, traffic resource occupation cost, and vehicle operation cost are set to 0.35, 0.30, 0.20, and 0.15, respectively. Among them, the coordinated travel time cost is determined based on the estimated travel time after coordination for the candidate parking path; the waiting time cost is determined based on the waiting time configured in the path traffic coordination scheme; the traffic resource occupation cost is determined based on the parking lot traffic resource occupation time for the candidate parking path; and the vehicle manipulation cost is determined based on the cumulative curvature change and the number of driving direction switching for the candidate parking path.

[0034] In this embodiment, step eight specifically includes: The target vehicle agent receives the target parking path, generates vehicle control commands based on the vehicle position, heading angle, steering angle and driving speed corresponding to the target parking path, and controls the target vehicle to drive along the target parking path. The system acquires the actual movement status of the target waiting vehicle in real time, updates the predicted occupied area of ​​the neighboring vehicle through the neighboring vehicle intelligent agent, and updates the spatiotemporal status attributes of the parking lot access resource through the parking lot access resource intelligent agent, thereby generating the target vehicle cooperative status updated in real time. Based on the actual motion state of the target waiting vehicle and the target parking path, the path tracking deviation is calculated, and based on the real-time updated target vehicle coordination state, predictive vehicle conflict detection and access resource availability judgment are performed on the unexecuted part of the target parking path. When the path tracking deviation exceeds the preset tracking deviation threshold, a predicted vehicle conflict occurs in the part of the path that has not yet been executed, or the corresponding parking lot access resources are unavailable, the current position of the target waiting vehicle is determined as the starting point for rolling correction. The path planning, time parameterization, and spatiotemporal conflict coordination are carried out again for the part of the path that has not yet been executed to generate a corrected target parking path. The preset tracking deviation thresholds include a lateral position deviation threshold and a heading angle deviation threshold. The lateral deviation threshold is set to 0.2m, and the heading angle deviation threshold is set to 5°. The target vehicle agent updates the vehicle control commands according to the corrected target parking path until the target vehicle reaches the target parking position and meets the target parking heading angle.

[0035] Example 1: To verify the feasibility of this invention in practice, it was applied to an automated parking path planning scenario in an underground parking garage of a commercial complex. The effective test area of ​​this parking garage was approximately 12,600 square meters, including 368 parking spaces, 42 road nodes, 57 oriented road sections, 6 intersection areas, and 4 narrow passage areas, with road widths ranging from 3.2 meters to 6.5 meters. The test vehicle had a length of 4.78 meters, a width of 1.86 meters, a wheelbase of 2.82 meters, a maximum front wheel turning angle of 35°, and a minimum turning radius of 5.3 meters. During the test, 8 to 15 adjacent vehicles were simultaneously operating in the parking garage, and scenarios such as temporary parking, intersection rushing, narrow passage occupation, and target vehicle tracking deviation were randomly set to simulate the continuous changes in road conditions, mutual interference of vehicle traffic, and limited local space in a real parking garage.

[0036] In the implementation scenario of this invention, the parking lot road network map, road boundaries, fixed obstacle areas, dynamic obstacle states, and traffic resource occupancy information are first acquired, and a parking lot spatiotemporal road network model is constructed according to a discrete time slice of 0.5 seconds. The target vehicle agent, neighboring vehicle agents, parking lot traffic resource agents, and parking lot environment coordination agents interact to generate the target vehicle's collaborative state. The improved Yen algorithm calculates the dynamic road segment cost based on road length, resource occupancy, predicted vehicle conflicts, current waiting state, and parking task priority, searching for three candidate topological paths with a path overlap threshold set to 0.70. For each candidate topological path, a continuous curvature parameter solution method based on spirals is used to determine the turning entry point, turning exit point, target turning curvature, and spiral connection parameters. Then, the turning connection position is adjusted through a vehicle sweep envelope correction mechanism, with a preset safety gap set to 0.35m. Path sampling points are set at 0.2m intervals for the formed candidate parking paths, and time parameterization and spatiotemporal traffic extrapolation are performed. The parking multi-agent system generates a path coordination scheme based on vehicle conflict information, resource availability time windows, and estimated waiting time, and ultimately selects the candidate parking path with the lowest cost of path coordination. When the actual movement state of the vehicles changes, the unexecuted parts of the path are continuously corrected.

[0037] In the test, 300 parking tasks were selected and compared with existing solutions that use Dijkstra's shortest path search, cubic spline smoothing, and static obstacle detection. The analysis results are shown in Table 1.

[0038] Table 1. Comparison of test results between the method of the present invention and existing parking path planning schemes.

[0039] As shown in Table 1, the path planning success rate of the method of this invention reaches 98.7%, which is 7.4 percentage points higher than the existing scheme, and the collision-free parking completion rate is increased to 98.0%, indicating that this invention can effectively improve the success rate of parking path generation and execution safety. The average parking completion time of this invention is reduced from 96.4s to 78.6s, the average waiting time is reduced from 21.7s to 10.3s, and the average path length is reduced from 164.8m to 158.2m, indicating that multi-agent collaboration and spatiotemporal traffic coordination can reduce unnecessary waiting and detours.

[0040] Furthermore, the minimum clearance increased from 0.18m to 0.41m, the maximum curvature abrupt change decreased from 0.083m⁻¹ to 0.012m⁻¹, and the average lateral tracking error decreased from 0.21m to 0.09m, indicating that the continuous curvature parameter solution method based on the spiral curve and the vehicle sweep envelope correction mechanism can improve path smoothness, spatial safety, and vehicle tracking accuracy. Meanwhile, the spatiotemporal conflict rate decreased from 14.3% to 3.0%, and the dynamic event replanning success rate increased from 82.5% to 96.8%, demonstrating that the invention can adapt to the dynamic changes in the motion state of neighboring vehicles and the state of traffic resources. Although the average planning time increased from 186ms to 238ms, it still meets the real-time planning requirements for low-speed parking scenarios in parking lots.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A parking path planning method based on multi-agent systems, characterized in that, Includes the following steps: Step 1: Obtain parking lot environmental data and target waiting vehicle task status data, and construct a parking lot spatiotemporal road network model; Step 2: Based on the task status data of the target vehicles waiting to park and the spatiotemporal road network model of the parking lot, construct a parking multi-agent system, and generate the cooperative state of the target vehicles through the state interaction between multiple agents in the parking multi-agent system; Step 3: Input the cooperative state of the target vehicle into the improved Yen algorithm to perform multi-path search on each directed road segment in the parking lot spatiotemporal road network model and generate a candidate topology path set for the target waiting vehicle; Step 4: Using the continuous curvature parameter solution method based on the spiral curve, calculate the turning connection parameters of each candidate topology path in the candidate topology path set; Step 5: Based on the steering connection parameters, the candidate topology paths are reconstructed using a vehicle sweep envelope correction mechanism to obtain a set of candidate parking paths for the target vehicle. Step 6: Based on the cooperative state of the target vehicle, perform time parameterization and spatiotemporal traffic simulation on each candidate parking path in the candidate parking path set, and output the spatiotemporal traffic evaluation results. Step 7: Based on the spatiotemporal traffic assessment results, the parking multi-agent system coordinates spatiotemporal conflicts and evaluates the cost of path collaboration for each candidate parking path, and selects the target parking path for the target vehicle to be parked. Step 8: Send the target parking path to the target vehicle agent, and the parking multi-agent system will make rolling corrections to the target parking path based on the real-time updated target vehicle collaborative status.

2. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, The parking lot environment data includes parking lot road network map, road boundaries, road traffic direction, fixed obstacle area, dynamic obstacle status, and traffic resource occupancy information; The parking lot road network map includes road nodes, road sections, intersection areas, narrow passage areas, and parking space entrance locations; The task status data includes the vehicle's current position, heading angle, speed, dimensions, wheelbase, maximum front wheel angle, minimum turning radius, target parking space, target parking position, target parking heading angle, task generation time, and task priority.

3. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, The construction of the parking lot spatiotemporal road network model specifically includes: Based on the parking lot road network map and road traffic direction, the directed road segments between road nodes and adjacent road nodes are extracted, and the traffic areas corresponding to the directed road segments, intersection areas, narrow passage areas and parking space entrances are identified as parking lot traffic resources. The parking lot spatial road network is constructed based on the connection relationships between road nodes, directed road segments and adjacent road nodes, and the parking lot spatial road network is extended in time according to discrete time slices to construct a parking lot spatiotemporal road network model. The dynamic obstacle status and access resource occupancy information are mapped to the parking lot spatiotemporal road network model to determine the occupancy and availability status of each parking lot's access resources in each discrete time slice, and these statuses are used as the spatiotemporal state attributes of the corresponding parking lot's access resources.

4. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, The parking multi-agent system includes a target vehicle agent, a neighboring vehicle agent, a parking lot access resource agent, and a parking lot environment coordination agent. The target vehicle cooperative status includes the vehicle's own movement status, the target parking space status, the movement status of nearby vehicles, the access resource status of adjacent parking lots, the current waiting status, and the parking task priority.

5. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, The improved Yen algorithm specifically includes: Based on the target vehicle's cooperative state, determine the dynamic road segment cost of each directed road segment in the parking lot's spatiotemporal road network model. The dynamic road segment cost is obtained by summing the basic traffic cost, resource occupation cost, vehicle conflict cost, waiting compensation cost, and task priority adjustment cost. The road node corresponding to the current position of the target waiting vehicle is determined as the starting road node, and the entrance position of the target parking space is determined as the target road node. The minimum cumulative cost path search is performed according to the dynamic road segment cost to generate an initial candidate topology path from the starting road node to the target road node. The initial candidate topology path is determined as the first selected topology path; From the road nodes included in the currently selected topology path, filter the road nodes that connect available branch road segments, and identify the filtered road nodes as deviation nodes; The available branch road segment is the subsequent directed road segment that the selected topology path has not passed through and whose corresponding parking lot access resources are available. For each deviation node, the road nodes and directed road segments between the starting road node and the deviation node are determined as the root path, and the subsequent directed road segments that have the same root path as the selected topology path are temporarily hidden. Based on the dynamic road segment cost of each directed road segment, a minimum cumulative cost path search is performed from the deviation node to the target road node to generate a deviation path. The root path is then concatenated with the deviation path to generate a deviation candidate topology path, which is then added to the deviation candidate path pool. Calculate the cumulative dynamic segment cost of each deviation from the candidate topology path, and calculate the path overlap between each deviation from the candidate topology path and the selected topology path using a weighted Jaccard coefficient. In the deviation candidate path pool, deviation candidate topology paths that are duplicates of the selected topology paths, as well as deviation candidate topology paths whose path overlap is greater than or equal to a preset overlap threshold, are removed. The candidate topology path with the lowest cumulative dynamic segment cost is selected from the candidate deviation path pool as the next selected topology path. The process of determining the deviation node, constructing the root path, generating the deviation path, and filtering the candidate topology path is repeated until the number of selected topology paths reaches the preset number of candidate paths. The selected topological paths are used as candidate topological paths, and they are arranged in ascending order of cumulative dynamic road segment cost to obtain the candidate topological path set for the target waiting vehicle.

6. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, The method for solving the continuous curvature parameters based on the spiral curve specifically includes: For each candidate topology path, the road centerline of each directed road segment is extracted according to the arrangement order of road nodes and directed road segments, and the road centerlines are connected in sequence to generate the path guide line corresponding to each candidate topology path. Calculate the directional angle between the center lines of adjacent roads in the path guide line, identify road nodes with directional angles greater than a preset turning angle threshold as path turning nodes, and determine the turning direction and turning angle of each path turning node based on the direction of the center lines of adjacent roads. The ratio of the tangent of the maximum front wheel steering angle to the vehicle wheelbase is used as the upper limit of steering curvature, the reciprocal of the minimum turning radius is used as the upper limit of radius curvature, and the smaller value of the upper limit of steering curvature and the upper limit of radius curvature is determined as the maximum allowable curvature of the target parking vehicle. Taking each path turning node as the center, the turning entry point and turning exit point are determined on the center lines of the adjacent roads before and after, and the path range between the turning entry point and the turning exit point is determined as the turning connection interval. The steering angle of each path steering node is used as the target heading change. Candidate steering curvatures are scanned incrementally between zero curvature and maximum allowable curvature. The steering connection length of each candidate steering curvature is calculated, and the smallest candidate steering curvature whose steering connection length does not exceed the length of the corresponding steering connection interval is determined as the target steering curvature. Based on the spiral curve relationship where curvature changes linearly with path arc length, we construct an entry spiral curve where curvature increases from zero to the target turning curvature and an exit spiral curve where curvature decreases from the target turning curvature to zero, and calculate the spiral curve connection parameters corresponding to each path turning node. The spiral curve connection parameters include the entry spiral curve length, the exit spiral curve length, the entry curvature change rate, and the exit curvature change rate. Calculate the heading changes when entering and exiting the spiral line respectively, and subtract them from the target heading changes at the corresponding path turning nodes to obtain the remaining heading changes of the circular arc. Calculate the ratio of the absolute value of the remaining heading change of the circular arc to the absolute value of the target turning curvature to obtain the length of the circular arc connection segment; By associating the turning entry point, turning exit point, target turning curvature, spiral line connection parameters, and arc connection segment length corresponding to each path turning node, the turning connection parameters of each candidate topology path are obtained.

7. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, The vehicle sweep envelope correction mechanism specifically includes: For each candidate topology path, based on the corresponding path guide line and steering connection parameters, the path guide line segments in each steering connection interval are removed, and the entry spiral segment, circular arc connection segment and exit spiral segment are constructed in sequence to generate an initial continuous curvature path. Based on the vehicle size of the target waiting vehicle, calculate the vehicle sweep envelope formed by the target waiting vehicle traveling along the initial continuous curvature path, and determine the minimum clearance between the vehicle sweep envelope and the road boundary and fixed obstacle area. Adjust the lateral positions of the steering entry point, steering exit point, and arc connection segment along the left and right lateral directions of the path guide line to generate multiple candidate correction parameter combinations; Based on each candidate correction parameter combination, the initial continuous curvature path is reconstructed to generate candidate correction paths, and the minimum passage gap and cumulative lateral offset relative to the path guide line of each candidate correction path are calculated. Candidate correction paths with minimum clearance greater than or equal to preset safety clearance are selected, and the candidate correction parameter combination corresponding to the candidate correction path with the smallest cumulative lateral offset is determined as the target correction parameter combination. The initial continuous curvature path is corrected by using a combination of target correction parameters to obtain candidate parking paths, and the candidate parking paths are summarized to obtain a candidate parking path set.

8. The parking path planning method based on multi-agents according to claim 1, characterized in that, Step six specifically includes: Path sampling points are set along each candidate parking path. Based on the path curvature, driving direction and the vehicle motion state in the target vehicle cooperative state, the vehicle position, vehicle heading angle, steering angle, driving speed and expected arrival time of each path sampling point are determined to generate an automatic parking time sequence trajectory. The time-series trajectory of automatic parking is mapped to the spatiotemporal road network model of the parking lot to determine the time period of traffic resource occupation for each candidate parking path; The target vehicle's time-series occupancy area is generated based on the automatic parking time-series trajectory and vehicle size. The target vehicle's time-series occupancy area is then compared with the predicted occupancy areas of neighboring vehicles in terms of time alignment. Combined with the spatiotemporal status attributes of parking lot access resources, the predicted vehicle conflict information and estimated waiting time are determined. The time period of traffic resource occupancy, predicted vehicle conflict information, and estimated waiting time corresponding to each candidate parking path are used as the spatiotemporal traffic assessment results.

9. The parking path planning method based on multi-agent systems according to claim 1, characterized in that, Step seven specifically includes: Based on the predicted vehicle conflict information of each candidate parking path, the neighboring vehicle intelligent agent determines the neighboring vehicles involved in the conflict, the time interval of the conflict, and the suggested passage order, and generates cooperative response information of neighboring vehicles. The parking lot access resource intelligent agent determines the resource availability time window, candidate waiting positions, and maximum allowable waiting time based on the access resource occupancy time of each candidate parking path and the spatiotemporal status attributes of the parking lot access resources, and generates access resource coordination information. The parking environment coordination agent adjusts the estimated arrival time and waiting time of each candidate parking path based on the cooperative response information of neighboring vehicles, traffic resource coordination information, estimated waiting time and parking task priority, and generates a path traffic coordination scheme. Based on the route coordination scheme, the estimated arrival time, waiting time and predicted vehicle conflict information after coordination of each candidate parking route are re-determined. Remove candidate parking paths from the candidate parking path set that still have predicted vehicle conflicts after coordination, or candidate parking paths whose waiting time after coordination exceeds the maximum allowable waiting time; Calculate the path coordination execution cost of the remaining candidate parking paths in the candidate parking path set, and determine the candidate parking path with the minimum path coordination execution cost as the target parking path; The cost of path coordination is calculated by weighting the cost of coordinated travel time, waiting time, traffic resource occupation, and vehicle operation.

10. A parking path planning method based on multi-agent systems according to claim 1, characterized in that, Step eight specifically includes: The target vehicle agent receives the target parking path, generates vehicle control commands based on the vehicle position, heading angle, steering angle and driving speed corresponding to the target parking path, and controls the target vehicle to drive along the target parking path. The system acquires the actual movement status of the target waiting vehicle in real time, updates the predicted occupied area of ​​the neighboring vehicle through the neighboring vehicle intelligent agent, and updates the spatiotemporal status attributes of the parking lot access resource through the parking lot access resource intelligent agent, thereby generating the target vehicle cooperative status updated in real time. Based on the actual motion state of the target waiting vehicle and the target parking path, the path tracking deviation is calculated, and based on the real-time updated target vehicle coordination state, predictive vehicle conflict detection and access resource availability judgment are performed on the unexecuted part of the target parking path. When the path tracking deviation exceeds the preset tracking deviation threshold, a predicted vehicle conflict occurs in the part of the path that has not yet been executed, or the corresponding parking lot access resources are unavailable, the current position of the target waiting vehicle is determined as the starting point for rolling correction. The path planning, time parameterization, and spatiotemporal conflict coordination are carried out again for the part of the path that has not yet been executed to generate a corrected target parking path. The target vehicle agent updates the vehicle control commands according to the corrected target parking path until the target vehicle reaches the target parking position and meets the target parking heading angle.

Images