Intelligent traffic vehicle behavior modeling method and system based on collaborative learning

Abstract

The invention provides an intelligent traffic vehicle behavior modeling method and system based on collaborative learning, and relates to the technical field of unmanned driving, and the method comprises the steps: obtaining traffic data in a target region; performing data calibration processing according to the traffic data to obtain a collaborative perception data set; a vehicle space-time trajectory function is generated according to the collaborative perception data set, and a conflict domain space-time distribution diagram is obtained; according to the vehicle space-time trajectory function and the conflict domain space-time distribution diagram, probability distribution characteristics of the vehicle driving intention and a multi-workshop competition cooperation relation matrix are extracted, and a multi-dimensional interaction characteristic vector is obtained; and behavior modeling is carried out according to the multi-dimensional interaction feature vectors, and a vehicle behavior model representing interaction behavior rules and decision logic between vehicles is obtained. According to the method, through multi-vehicle cooperative sensing data calibration and dynamic fusion of road geometric constraints, the problems of single-vehicle sensor shielding and cross-vehicle data heterogeneity are effectively solved, and the global consistency of vehicle position and motion state estimation is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned driving technology, and in particular to a collaborative learning-based intelligent traffic vehicle behavior modeling method and system. Background Art

[0002] In the field of unmanned driving and intelligent transportation, existing technologies are mostly based on single-vehicle perception systems (such as lidar, cameras) and independent decision-making frameworks, which are difficult to deal with complex interaction problems in multi-vehicle path intersection scenarios. Especially in dynamic environments such as intersections without signal lights, traditional methods have significant bottlenecks: single-vehicle perception is easily blocked and interfered by sensor noise, resulting in the accumulation of trajectory prediction errors, threatening the safety of unmanned driving systems; intelligent traffic management solutions based on fixed rules or centralized scheduling lack the ability to dynamically model the game relationship between multiple vehicles, and it is difficult to generate global optimal strategies in real time. Existing technologies attempt to achieve vehicle-road collaboration through V2X communication, such as using preset traffic priority or time slot allocation mechanisms, but such solutions still have defects: on the one hand, they are highly dependent on high-precision infrastructure and cannot adapt to dynamic changes in traffic flow; on the other hand, they lack a collaborative learning mechanism between vehicles, resulting in a low success rate of conflict resolution and reduced traffic efficiency. In addition, existing methods mostly use isolated learning models, which do not fully utilize the potential correlation of multi-vehicle interaction data. When there is a communication delay or some vehicle perception fails, the robustness of behavior prediction is significantly reduced.

[0003] Based on the above shortcomings of the prior art, there is an urgent need for an intelligent traffic vehicle behavior modeling method and system based on collaborative learning. Summary of the invention

[0004] The purpose of the present invention is to provide a method for intelligent traffic vehicle behavior modeling based on collaborative learning to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows: In a first aspect, the present application provides a method for modeling intelligent traffic vehicle behavior based on collaborative learning, comprising: Acquire traffic data in the target area, the traffic data including real-time status data of the current vehicle, road structure data and neighboring vehicle interaction data; Performing data calibration processing according to the traffic data, and eliminating abnormal position points outside lane boundaries based on road geometric constraints to obtain a collaborative perception data set; Generate a space-time trajectory function of the vehicle according to the collaborative perception data set, and obtain a space-time distribution map of the conflict domain based on the trajectory intersection time window prediction and dynamic safety radius calculation; According to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, by associating the temporal behavior pattern and quantifying the interactive game relationship, the probability distribution characteristics of the vehicle driving intention and the competition and cooperation relationship matrix of multiple workshops are extracted to obtain a multi-dimensional interactive feature vector; Behavior modeling is performed based on the multi-dimensional interaction feature vector, and a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles is obtained by constructing a hierarchical graph network model based on node dynamic feature encoding and edge relationship weight update.

[0005] In a second aspect, the present application also provides an intelligent traffic vehicle behavior modeling system based on collaborative learning, comprising: An acquisition module is used to acquire traffic data in a target area, wherein the traffic data includes real-time status data of the current vehicle, road structure data, and neighboring vehicle interaction data; A calibration module, used for performing data calibration processing according to the traffic data, and eliminating abnormal position points outside the lane boundary based on road geometric constraints to obtain a collaborative perception data set; A prediction module, used to generate a vehicle spatiotemporal trajectory function according to the collaborative perception data set, and obtain a spatiotemporal distribution map of the conflict domain based on trajectory intersection time window prediction and dynamic safety radius calculation; An extraction module is used to extract the probability distribution characteristics of vehicle driving intentions and the competition and cooperation relationship matrix of multiple workshops according to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, through the association of temporal behavior patterns and the quantification of interactive game relationships, to obtain a multi-dimensional interactive feature vector; A modeling module is used to perform behavior modeling according to the multi-dimensional interaction feature vector, and obtain a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles by constructing a hierarchical graph network model based on node dynamic feature encoding and edge relationship weight update.

[0006] The beneficial effects of the present invention are: The present invention calibrates multi-vehicle collaborative perception data and utilizes the dynamic fusion of road geometry constraints to effectively solve the problems of single-vehicle sensor occlusion and cross-vehicle data heterogeneity, and significantly improves the global consistency of vehicle position and motion state estimation; through interactive feature extraction and hierarchical graph network optimization, the driving intention probability, game relationship matrix and traffic rules are jointly encoded into node and edge attributes, realizing the joint optimization of single-vehicle strategy and multi-vehicle collaboration, and improving the success rate of conflict resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0008] Figure 1A flowchart of a method for modeling intelligent transportation vehicle behavior based on collaborative learning according to an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an intelligent transportation vehicle behavior modeling system based on collaborative learning described in an embodiment of the present invention; Figure 3 It is a schematic diagram of the structure of an intelligent transportation vehicle behavior modeling device based on collaborative learning described in an embodiment of the present invention.

[0009] Markings in the figure: 800, an intelligent traffic vehicle behavior modeling device based on collaborative learning; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component; 901, acquisition module; 902, calibration module; 903, prediction module; 904, extraction module; 905, modeling module. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0011] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0012] Example This embodiment provides an intelligent transportation vehicle behavior modeling method based on collaborative learning.

[0013] See also Figure 1 , the figure shows that the method includes steps S100 to S500.

[0014] Step S100, obtaining traffic data in the target area, the traffic data including real-time status data of the current vehicle, road structure data and neighboring vehicle interaction data; This step uses a multi-source heterogeneous data synchronous collection mechanism to obtain three types of data in the target area: real-time status data of the current vehicle, including position, speed, acceleration and heading angle collected by on-board sensors, and vehicle identity and communication status obtained by V2X communication; road structure data, intersection lane topology, lane line curvature and traffic rule constraints analyzed by high-precision maps; and neighboring vehicle interaction data, historical behavior sequences, intention labels and interaction event records of surrounding vehicles obtained through V2V communication. In complex scenarios such as intersections without signal lights, single-vehicle sensors are prone to perception blind spots due to occlusion or environmental interference (such as rain, fog, and backlight). Although V2X / V2V communication can supplement neighboring vehicle data, it faces heterogeneous problems such as asynchronous timestamps and inconsistent coordinate systems. This step provides global input for subsequent collaborative perception and modeling by fusing multimodal data.

[0015] Step S200: performing data calibration processing according to the traffic data, and eliminating abnormal position points outside the lane boundary based on the road geometry constraint to obtain a collaborative perception data set; This step solves the problem of spatiotemporal heterogeneity of multi-source data through a four-layer progressive calibration process: first, by combining the clock deviation observation data of multiple vehicles with the periodic clock calibration signal of V2V communication, the timestamp offset is dynamically corrected to eliminate the time asynchrony of cross-vehicle data; then the positions of multiple vehicles are mapped to the global coordinate system defined by the high-precision map, and the spatial benchmark consistency is verified in combination with the lane topology connectivity (such as the steering logic relationship); then the vehicle observation topology map is constructed based on the relative position and movement trend of the neighboring vehicles, the acceleration and heading angle of the obscured vehicle are predicted, and the sensor blind spot is filled; finally, the vehicle position is forced to the lane boundary of the high-precision map through the quadratic programming algorithm to eliminate abnormal drift caused by sensor noise or communication delay. In the scene of the intersection without signal lights, the above calibration mechanism specifically solves the problems of time asynchrony, spatial dislocation and occlusion, and outputs a collaborative perception data set that is aligned in time and space and conforms to the lane geometry rules, providing high-precision input for trajectory modeling, ensuring the accuracy of subsequent conflict domain risk quantification, and avoiding illegal trajectory generation.

[0016] Step S300: Generate a vehicle spatiotemporal trajectory function according to the collaborative perception data set, and obtain a spatiotemporal distribution map of the conflict domain based on trajectory intersection time window prediction and dynamic safety radius calculation; This step generates a continuous spatiotemporal trajectory function of vehicles based on the calibrated collaborative perception data set through dynamic trajectory modeling and conflict risk quantification mechanism, and combines the real-time motion state of vehicles (speed, acceleration) and sensor performance parameters (such as detection error, system response delay) to predict the spatiotemporal intersection of multi-vehicle trajectories at the intersection and their time window confidence intervals. It also dynamically calculates the adaptive safety radius based on the maximum deceleration of the vehicle and environmental conditions (such as road adhesion coefficient) to quantify the risk level of the conflict domain. In the scene of intersections without signal lights, the traditional fixed safety distance model cannot cope with the impact of sudden lane changes or bad weather. This step accurately identifies high-risk conflict areas and their timeliness through dynamic safety radius adjustment (such as increasing redundant fault tolerance on rainy and foggy days) and time window overlap probability analysis, and outputs a spatiotemporal distribution map of the conflict domain including location, time, and risk intensity, providing a quantitative basis for subsequent interactive game analysis, and significantly improving the reliability and real-time performance of multi-vehicle collaborative decision-making in complex scenarios.

[0017] Step S400, according to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, by quantifying the temporal behavior pattern association and the interactive game relationship, extracting the probability distribution characteristics of the vehicle driving intention and the competition and cooperation relationship matrix of multiple workshops, and obtaining a multi-dimensional interactive feature vector; This step uses driving intention reasoning and multi-vehicle game relationship modeling, based on the historical motion sequence in the vehicle spatiotemporal trajectory function and the risk level of the conflict domain spatiotemporal distribution map, and uses the long short-term memory network to analyze the temporal behavior pattern, combined with the steering logic constraints of the intersection lane topology, to output the probability distribution of driving intention; at the same time, according to the overlap of the conflict time window, relative speed and historical interaction data, a competition and cooperation relationship matrix is ​​constructed to quantify the vehicle strategy benefits, and the intention probability, game relationship and vehicle identity label are integrated into a multi-dimensional interaction feature vector through feature cascading and normalization operations. In dynamic game scenarios, traditional rule engines are difficult to deal with the uncertainty of neighboring vehicle intentions. This step uses a data-driven feature extraction mechanism to explicitly encode implicit interaction relationships into structured features, support hierarchical graph network collaborative optimization, and realize intention-interpretable and strategy-adaptive behavior modeling in multi-vehicle conflict scenarios.

[0018] Step S500 , behavior modeling is performed according to the multi-dimensional interaction feature vector, and a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles is obtained by constructing a hierarchical graph network model based on node dynamic feature coding and edge relationship weight update.

[0019] This step maps the multi-dimensional interaction feature vector into a spatiotemporal graph structure through a hierarchical graph network architecture and a multi-agent collaborative optimization mechanism, where nodes represent the dynamic state of vehicles and edge weights quantify the interaction intensity between vehicles. Based on the node dynamic feature encoding and the iterative update of edge relationship weights, a multi-agent reinforcement learning framework is used to jointly optimize the single-vehicle strategy and multi-vehicle collaborative relationship, with the conflict resolution success rate and traffic efficiency as the global reward function, while pruning the illegal strategy action space through traffic rules constraints. This model breaks through the limitations of traditional centralized scheduling or independent decision-making, and directly supports the highly robust collaborative decision-making of autonomous driving systems in complex scenarios through the divide-and-conquer optimization of hierarchical graph networks (nodes focus on single-vehicle behavior generation, and edges focus on multi-vehicle interaction and coordination).

[0020] Further, step S200 includes step S210 to step S240.

[0021] Step S210: Perform time synchronization processing according to the traffic data, correct the timestamp offset between vehicles by combining the clock deviation observation data of multiple vehicles, and use the periodic clock calibration signal of the communication between vehicles as the time reference to obtain a multi-vehicle state data set; This step aims to solve the time asynchrony problem of multi-vehicle data by establishing a unified time reference through periodic clock calibration signals between vehicles (such as time synchronization messages embedded in V2V communication), combining the clock deviation observation data of multiple vehicles, and dynamically correcting the timestamp offset of each vehicle's sensor data. The specific process includes: periodically broadcasting calibration signals so that all vehicles align their local time with the clock of a reference node (such as the main vehicle or roadside unit); calculating the timestamp compensation based on the joint observation value of the clock deviation of multiple vehicles; and resampling the original state data on the time axis to eliminate the time misalignment of cross-vehicle data.

[0022] Step S220, performing unified spatial coordinate processing according to the multi-vehicle state data set, mapping the multi-vehicle position data to the global map coordinate system, and obtaining a vehicle position data set; This step uses the global coordinate system defined by the high-precision map to uniformly convert the position data collected by each vehicle based on its own local coordinate system (such as the center of the vehicle body as the origin) into the global coordinates under the map coordinate system. In practice, the local coordinate system deviations of multiple vehicles may occur due to different driving directions. Direct use of the original position data will cause trajectory intersection prediction errors. Through coordinate transformation and lane topology logic verification, ensure that all vehicle position data are aligned under a unified spatial reference, and eliminate relative position calculation errors caused by coordinate system differences. Step S230, performing occlusion compensation processing according to the vehicle position data set, constructing a vehicle observation topology map and predicting the acceleration and heading angle of the occluded vehicle based on the relative position and movement trend of neighboring vehicles, to obtain a vehicle status data set; This step aims to solve the problem of missing vehicle status caused by sensor occlusion. By constructing a vehicle observation topology graph (nodes are vehicles, and edges represent the relative position and motion trend relationship between vehicles), the acceleration and heading angle of the blocked vehicle are inferred using the real-time position, speed and historical behavior sequence of neighboring vehicles. The dynamic state of the blocked vehicle is predicted through motion trend extrapolation and interaction history association, ensuring the integrity of multi-vehicle data.

[0023] Step S240: Perform road geometry constraint correction processing according to the vehicle state data set, use a quadratic programming algorithm to constrain the vehicle position to the lane boundary defined by the map, and obtain a collaborative perception data set.

[0024] This step addresses the problem of abnormal vehicle position drift caused by sensor noise or communication delays. The lane boundaries defined by the high-precision map are used as hard constraints to construct a quadratic programming optimization goal: Under the premise of minimizing the deviation between the original position and the corrected position of the vehicle, the vehicle position is forced to be within the lane boundary. For curvature lanes (such as roundabouts), the lane boundary constraints are converted into lateral displacement limits in the Frenet coordinate system, and the correction range is dynamically adjusted to adapt to lane geometry changes. Traditional rigid boundary corrections are prone to trajectory distortion (such as vehicles in curvature lanes being incorrectly corrected to adjacent lanes), and this step ensures that the corrected vehicle position complies with physical rules and maintains motion continuity through dynamic coordinate system conversion and smoothing optimization.

[0025] Further, step S300 includes step S310 to step S340.

[0026] Step S310: Generate a vehicle spatiotemporal trajectory function based on the collaborative sensing data set and in combination with the vehicle's historical position and real-time motion state; Specifically, data preprocessing is first performed. In the data preprocessing stage, single-point data errors are eliminated by spatial weighted averaging. The weight calculation uses the Gaussian kernel function, and the formula is: ; ; ; in, Indicates The original coordinates of the cooperative vehicles; Indicates the serial number of the collaborative sensing vehicle; represents the number of collaborative sensing vehicles; , represents the coordinates of the vehicle; represents the spatial correlation attenuation coefficient; Indicates the communication radius; Represents the estimated vehicle position after spatiotemporal fusion; Indicates The spatial credibility weight of the cooperative vehicles; Represents the Hadamard product.

[0027] Then a prediction model is established to integrate historical trends and real-time data. The formula is: ; ; in, express The predicted space-time coordinates of the moment; represents the time-varying fusion acceleration; represents the position vector calibrated by collaborative data; represents the fused velocity vector; represents the historical average acceleration; Indicates real-time acceleration measurement value; Indicates the weight of historical data; Indicates the time sequence number.

[0028] The final space-time trajectory function is expressed as: ;in, represents the space-time trajectory function; represents the prediction time window; represents the confidence interval; Indicates the time synchronization reference point.

[0029] Step S320, performing trajectory intersection prediction processing according to the vehicle space-time trajectory function, and obtaining a trajectory intersection space-time distribution table by calculating the space-time intersection coordinates of multiple vehicle trajectories and the confidence interval of the intersection time window; It can be understood that this step uses the spatiotemporal overlay analysis method to identify the intersection area of ​​different vehicle trajectories in time and space dimensions by integrating the predicted driving routes of each vehicle. First, based on the motion trend prediction model of each vehicle, the sequence of position points that the vehicle may pass through in the future time period is calculated, and then the intersection positions and corresponding time nodes between different vehicle paths are found through the cross-validation algorithm. For each predicted intersection point, combined with the positioning accuracy of each vehicle, communication delay and other influencing factors, the time fluctuation range and spatial position deviation interval of the intersection event are calculated, and finally these prediction results are integrated into a structured data table. This table records in detail the most likely time of occurrence, location coordinates and corresponding credibility evaluation indicators of each potential conflict point.

[0030] Step S330: performing conflict time window prediction processing according to the temporal and spatial distribution table of track intersections, and obtaining a high conflict risk time window by calculating the estimated time distribution of vehicles arriving at the intersections and the overlap probability of the conflict time windows; First, a Gaussian distribution model is established for the estimated arrival time of each vehicle, where the mean is calculated by the trajectory function, and the standard deviation integrates sensor noise (such as GPS error) and vehicle dynamic fluctuations (such as acceleration and deceleration randomness) to quantify time uncertainty. Subsequently, the overlapping probability of the arrival time distribution of multiple vehicles is calculated: for two vehicles, the risk of conflict is determined by integrating the joint probability density function of their time difference within the safety threshold; for multi-vehicle scenarios, the Monte Carlo sampling method is used to simulate a large number of possible arrival time combinations, and the frequency of the intersection of all vehicle time windows is counted to efficiently estimate the probability of conflict. At the same time, the environmental visibility factor is introduced to dynamically adjust the safety time threshold, such as extending the basic threshold in rainy and foggy weather, to enhance the adaptability of the model to complex conditions. Finally, the high-risk time window and its probability level are output to provide a quantitative basis for subsequent behavior modeling, ensuring that the autonomous driving system prioritizes urgent conflicts and optimizes traffic efficiency while ensuring safety. This step solves the shortcomings of the traditional fixed threshold method in real-time and adaptability by integrating probability theory and dynamic environmental parameters, and significantly improves the accuracy of conflict prediction in complex traffic scenarios.

[0031] Step S340: dynamically calculate and process the safety radius according to the high conflict risk time window, dynamically adjust the safety radius threshold based on the vehicle's maximum deceleration, system response delay, and sensor detection error parameters, and obtain a conflict domain spatiotemporal distribution diagram.

[0032] It should be noted that this step calculates the basic safety distance based on the maximum deceleration of the vehicle, the system reaction delay (including the time taken for sensor data processing and decision-making) and the real-time vehicle speed, covering the space margin required for the vehicle to stop completely from sensing; secondly, sensor detection error parameters (such as LiDAR point cloud density attenuation, camera resolution reduction under low light) and environmental visibility factors (such as rain and fog weather coefficient) are introduced to dynamically expand the safety radius threshold to ensure that the redundant fault tolerance is increased under adverse conditions. On this basis, the safety radius is associated with the conflict time window, and the multi-vehicle safety area is superimposed along the spatiotemporal distribution of the vehicle trajectory. The overlapping area is identified as a high-conflict domain, and multi-dimensional risk quantification is performed based on the overlapping area, the urgency of the time window (such as the remaining time ratio) and the environmental risk level (such as the slippery road coefficient), forming a spatiotemporal distribution map of the conflict domain including location, time and risk intensity.

[0033] Further, step S400 includes step S410 to step S440.

[0034] Step S410: performing driving intention classification processing according to the vehicle spatiotemporal trajectory function, taking the acceleration change rate of the vehicle historical trajectory, the heading angle deviation characteristics, and the steering logic constraints of the intersection lane topology as input values ​​of the long short-term memory network, and outputting the probability distribution characteristics of the driving intention; Step S420: Modeling the interaction relationship according to the spatial-temporal distribution diagram of the conflict domain, and constructing a competition-cooperation relationship matrix by calculating the spatial-temporal proximity of the conflict domains between vehicles; Step S430, calculating the game strategy benefits according to the competition and cooperation relationship matrix, quantifying the benefits of each vehicle taking the acceleration, deceleration and yielding strategies in the conflict domain, and generating a strategy benefit distribution table; Step S440: Perform feature fusion processing according to the probability distribution characteristics and the strategy return distribution table to obtain a multi-dimensional interactive feature vector.

[0035] It can be understood that first, in step S410, based on the acceleration change rate and heading angle deviation characteristics of the vehicle's historical trajectory, combined with the steering logic constraints of the intersection lane topology (such as the left turn lane prohibits straight driving), the long short-term memory network is used to analyze the temporal behavior pattern, and the probability distribution of driving intentions (such as straight driving, left turn, slowing down and giving way) is output. At the same time, illegal intentions are forced to be excluded through lane rules to ensure that the prediction results comply with traffic regulations. Then in step S420, according to the conflict domain spatiotemporal distribution map, the spatiotemporal proximity between vehicles (such as the time difference and relative distance to the conflict domain) is quantified, and the competition and cooperation relationship matrix is ​​constructed in combination with historical interaction data (such as the frequency of giving way between vehicles and the habit of following vehicles) to characterize the game tendency of vehicles in the conflict domain (such as aggressive competition or conservative giving way). Further in step S430, based on the non-cooperative game model, the benefit value of each vehicle taking strategies such as acceleration, deceleration, and giving way is calculated, and the remaining time window of the conflict domain is introduced as a benefit attenuation factor to encourage vehicles to make decisions as soon as possible, while excluding physically infeasible strategies (such as aggressive actions exceeding the maximum acceleration), and generating a strategy benefit distribution table. Finally, through feature cascading and normalization operations, the intention probability distribution, game relationship matrix and vehicle identity tags (such as emergency vehicles and truck priorities) are integrated into a multi-dimensional interactive feature vector, and feature dimensionality reduction is performed based on the spatial and temporal range of the conflict domain to eliminate redundant information. This process breaks through the static limitations of traditional rule engines, and through data-driven dynamic feature extraction and game quantification, it realizes multi-vehicle collaborative modeling with interpretable intentions and adaptive strategies in complex scenarios, providing high-information-density input for subsequent hierarchical graph networks, significantly improving the success rate of intersection conflict resolution and traffic efficiency.

[0036] Further, step S500 includes step S510 to step S540.

[0037] Step S510: constructing a spatiotemporal graph structure according to the multi-dimensional interactive feature vector, converting the vehicle identity label, intention probability distribution and conflict domain association into node attributes and edge connection relationships of the spatiotemporal graph structure, and combining the connectivity constraints of the intersection lane topology to obtain a hierarchical spatiotemporal graph structure; It can be understood that this step maps the multi-dimensional interaction features into a hierarchical spatiotemporal graph structure to intuitively represent the dynamic interaction relationship between vehicles. Nodes are defined as vehicle entities, whose attributes include real-time status (speed, acceleration), intention probability (such as 70% left turn, 30% straight drive) and identity labels (such as emergency vehicles, ordinary vehicles); edges represent the interaction relationship between vehicles, and the weights are determined by the spatiotemporal proximity of the conflict domain (time difference, distance difference) and the historical cooperation tendency (such as frequent yielding). Secondly, this step introduces lane topology connectivity constraints, such as prohibiting lane changes on solid lines or specific turning restrictions, and automatically filters illegal connections (such as potential conflicting edges across solid lines) to ensure that the graph structure strictly complies with traffic rules. The hierarchical design (such as the bottom single-vehicle status and the top multi-vehicle interaction) supports divide-and-conquer optimization, reduces computational complexity, and provides a scalable topological framework for subsequent feature aggregation and strategy optimization.

[0038] Step S520: Aggregate node features based on the hierarchical spatiotemporal graph structure, and perform bias correction on the attention weight in combination with the vehicle priority label to obtain the encoded feature vector of the node; Preferably, this step is based on a hierarchical spatiotemporal graph and uses a gated graph attention network (GGAT) to aggregate neighborhood information. Each node uses a multi-head attention mechanism to weightedly fuse its own features with neighboring vehicle features, and the weight calculation introduces a vehicle priority bias: for example, the ambulance node is given a higher weight coefficient in the attention calculation, forcing the model to give priority to its dynamic intention. At the same time, lane topology information (such as the turning authority of the lane where the vehicle is located) is used to constrain the feature propagation path to avoid the spread of illegal information (such as a vehicle in the straight lane receiving information about a neighboring vehicle turning left). This step outputs a node encoding feature vector, which not only contains the state of a single vehicle, but also implies the influence of neighborhood interactions, providing a node representation with high information density for the dynamic update of edge weights.

[0039] Step S530, iteratively updating the edge relationship weights according to the encoded feature vectors, and modifying the weights based on the real-time conflict domain risk changes and the yielding habits between vehicles to obtain a dynamic edge relationship matrix; Specifically, in this step, the edge weights are dynamically adjusted according to the real-time scenario, including: conflict time attenuation, for example, when the remaining time of the conflict domain is reduced from 10 seconds to 2 seconds, the corresponding edge weight increases exponentially, and the urgency-driven model prioritizes high-risk interactions; historical interaction habit correction, if the historical yielding probability of vehicle A to vehicle B reaches 80%, the competition weight is reduced in the current interaction to enhance the tendency of cooperation; sensor confidence compensation, for vehicles with increased LiDAR detection errors in low visibility, the weight reliability of their associated edges is reduced. Through the online iteration mechanism, the edge relationship matrix responds to environmental changes in real time to ensure that the interaction intensity is strictly matched with the risk level, avoiding the decision lag problem of the static model.

[0040] Step S540: Based on the encoded feature vector and the dynamic edge relationship matrix, a vehicle behavior model is constructed through collaborative strategy optimization and strategy space pruning of traffic rule constraints.

[0041] It should be noted that this step is based on node encoding features and dynamic edge matrices, and is jointly trained using the multi-agent proximal policy optimization (MAPPO) framework. The global reward function is designed as the weighted sum of the conflict resolution success rate and traffic efficiency, which encourages the model to optimize the overall traffic flow under the premise of safety; the policy space pruning directly embeds traffic rules, such as prohibiting lane changes in solid line areas and stopping and waiting under mandatory yield signs. During the training process, the node policy generator (controls single-vehicle behavior) and the edge relationship optimizer (coordinates multi-vehicle interactions) are updated alternately to achieve hierarchical optimization. The final model output is a deployable policy network that supports real-time reasoning at the millisecond level, generating compliant, collaborative, and dynamically adaptive vehicle behaviors in complex scenarios such as intersections without signal lights, such as triggering multi-vehicle collaborative deceleration in high-risk conflict domains and maintaining efficient traffic in low-risk scenarios.

[0042] Embodiment 2: like Figure 2 As shown, this embodiment provides an intelligent traffic vehicle behavior modeling system based on collaborative learning, and the system includes: The acquisition module 901 is used to acquire traffic data in the target area, where the traffic data includes real-time status data of the current vehicle, road structure data, and neighboring vehicle interaction data; A calibration module 902 is used to perform data calibration processing according to traffic data, and eliminate abnormal location points outside lane boundaries based on road geometric constraints to obtain a collaborative sensing data set; Prediction module 903, used to generate a vehicle spatiotemporal trajectory function according to the collaborative perception data set, and obtain a spatiotemporal distribution map of the conflict domain based on trajectory intersection time window prediction and dynamic safety radius calculation; Extraction module 904 is used to extract the probability distribution characteristics of vehicle driving intentions and the competition and cooperation relationship matrix of multiple vehicles according to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, through the association of temporal behavior patterns and the quantification of interactive game relationships, to obtain a multi-dimensional interactive feature vector; The modeling module 905 is used to perform behavior modeling based on the multi-dimensional interaction feature vector, and obtain a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles by constructing a hierarchical graph network model based on node dynamic feature encoding and edge relationship weight update.

[0043] In some embodiments disclosed in the present application, the calibration module 902 includes: A first calibration unit is used to perform time synchronization processing according to traffic data, correct the timestamp offset between vehicles by combining the clock deviation observation data of multiple vehicles, and use the periodic clock calibration signal of the communication between vehicles as the time reference to obtain a multi-vehicle state data set; A second calibration unit is used to perform unified spatial coordinate processing according to the multi-vehicle state data set, map the multi-vehicle position data to the global map coordinate system, and obtain a vehicle position data set; The third calibration unit is used to perform occlusion compensation processing according to the vehicle position data set, and obtain the vehicle state data set by constructing a vehicle observation topology map and predicting the acceleration and heading angle of the occluded vehicle based on the relative position and movement trend of the neighboring vehicles; The fourth calibration unit is used to perform road geometry constraint correction processing according to the vehicle state data set, and use a quadratic programming algorithm to constrain the vehicle position to the lane boundary defined by the map to obtain a collaborative perception data set.

[0044] In some embodiments disclosed in the present application, the prediction module 903 includes: A first prediction unit is used to generate a vehicle spatiotemporal trajectory function based on the collaborative perception data set and in combination with the vehicle's historical position and real-time motion state; The second prediction unit is used to perform trajectory intersection prediction processing according to the vehicle space-time trajectory function, and obtain a trajectory intersection space-time distribution table by calculating the space-time intersection coordinates of multiple vehicle trajectories and the confidence interval of the intersection time window; The third prediction unit is used to perform conflict time window prediction processing according to the spatiotemporal distribution table of trajectory intersections, and obtain a high conflict risk time window by calculating the estimated time distribution of vehicles arriving at the intersection and the overlap probability of the conflict time window; The fourth prediction unit is used to perform dynamic safety radius calculation and processing according to the high conflict risk time window, dynamically adjust the safety radius threshold based on the vehicle's maximum deceleration, system response delay and sensor detection error parameters, and obtain a conflict domain spatiotemporal distribution map.

[0045] In some embodiments disclosed in the present application, the extraction module 904 includes: The first extraction unit is used to perform driving intention classification processing according to the vehicle spatiotemporal trajectory function, and outputs the probability distribution characteristics of the driving intention by taking the acceleration change rate and heading angle deviation characteristics of the vehicle's historical trajectory and the steering logic constraints of the intersection lane topology as the input values ​​of the long short-term memory network; The second extraction unit is used to model the interaction relationship according to the time-space distribution diagram of the conflict domain, and to construct a competition-cooperation relationship matrix by calculating the time-space proximity of the conflict domains between vehicles; The third extraction unit is used to calculate the game strategy benefits according to the competition and cooperation relationship matrix, quantify the benefits of each vehicle taking the acceleration, deceleration and yielding strategies in the conflict domain, and generate a strategy benefit distribution table; The fourth extraction unit is used to perform feature fusion processing according to the probability distribution characteristics and the strategy return distribution table to obtain a multi-dimensional interactive feature vector.

[0046] In some embodiments disclosed in the present application, the modeling module 905 includes: The first modeling unit is used to construct a spatiotemporal graph structure according to the multi-dimensional interactive feature vector, convert the vehicle identity label, intention probability distribution and conflict domain association relationship into the node attributes and edge connection relationship of the spatiotemporal graph structure, and combine the connectivity constraints of the intersection lane topology to obtain a hierarchical spatiotemporal graph structure; The second modeling unit is used to aggregate node features based on the hierarchical spatiotemporal graph structure, and to perform bias correction on the attention weights in combination with the vehicle priority labels to obtain the encoded feature vectors of the nodes; The third modeling unit is used to iteratively update the edge relationship weight according to the encoded feature vector, and to modify the weight based on the real-time conflict domain risk change and the yielding habit between vehicles to obtain a dynamic edge relationship matrix; The fourth modeling unit is used to construct a vehicle behavior model based on the encoded feature vector and the dynamic edge relationship matrix through collaborative strategy optimization and strategy space pruning processing constrained by traffic rules.

[0047] Embodiment 3: Corresponding to the above method embodiment, this embodiment also provides an intelligent traffic vehicle behavior modeling device based on collaborative learning. The intelligent traffic vehicle behavior modeling device based on collaborative learning described below and the intelligent traffic vehicle behavior modeling method based on collaborative learning described above can refer to each other.

[0048] Figure 3 is a block diagram of an intelligent traffic vehicle behavior modeling device 800 based on collaborative learning according to an exemplary embodiment. Figure 3 As shown, the intelligent traffic vehicle behavior modeling device 800 based on collaborative learning may include: a processor 801 and a memory 802. The intelligent traffic vehicle behavior modeling device 800 based on collaborative learning may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0049] The processor 801 is used to control the overall operation of the intelligent traffic vehicle behavior modeling device 800 based on collaborative learning to complete all or part of the steps in the above-mentioned intelligent traffic vehicle behavior modeling method based on collaborative learning. The memory 802 is used to store various types of data to support the operation of the intelligent traffic vehicle behavior modeling device 800 based on collaborative learning, and these data may include, for example, instructions for any application or method operating on the intelligent traffic vehicle behavior modeling device 800 based on collaborative learning, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the intelligent traffic vehicle behavior modeling device 800 based on collaborative learning and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0050] In an exemplary embodiment, an intelligent traffic vehicle behavior modeling device 800 based on collaborative learning can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components, and is used to execute the above-mentioned intelligent traffic vehicle behavior modeling method based on collaborative learning.

[0051] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned intelligent traffic vehicle behavior modeling method based on collaborative learning are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by a processor 801 of an intelligent traffic vehicle behavior modeling device 800 based on collaborative learning to complete the above-mentioned intelligent traffic vehicle behavior modeling method based on collaborative learning.

[0052] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for modeling intelligent traffic vehicle behavior based on collaborative learning, characterized in that: include: Acquire traffic data in the target area, the traffic data including real-time status data of the current vehicle, road structure data and neighboring vehicle interaction data; Performing data calibration processing according to the traffic data, and eliminating abnormal position points outside lane boundaries based on road geometric constraints to obtain a collaborative perception data set; Generate a space-time trajectory function of the vehicle according to the collaborative perception data set, and obtain a space-time distribution map of the conflict domain based on the trajectory intersection time window prediction and dynamic safety radius calculation; According to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, by quantifying the temporal behavior pattern association and the interactive game relationship, the probability distribution characteristics of vehicle driving intentions and the competition and cooperation relationship matrix of multiple workshops are extracted to obtain a multi-dimensional interactive feature vector; Behavior modeling is performed based on the multi-dimensional interaction feature vector, and a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles is obtained by constructing a hierarchical graph network model based on node dynamic feature encoding and edge relationship weight update.

2. The intelligent traffic vehicle behavior modeling method based on collaborative learning according to claim 1 is characterized in that: Data calibration is performed according to the traffic data, and abnormal position points outside the lane boundary are eliminated based on road geometric constraints to obtain a collaborative perception data set, including: Performing time synchronization processing according to the traffic data, correcting the timestamp offset between vehicles by combining the clock deviation observation data of multiple vehicles, and using the periodic clock calibration signal of the communication between vehicles as the time reference, to obtain a multi-vehicle state data set; Performing unified spatial coordinate processing according to the multi-vehicle state data set, mapping the multi-vehicle position data to a global map coordinate system, and obtaining a vehicle position data set; Performing occlusion compensation processing according to the vehicle position data set, obtaining a vehicle state data set by constructing a vehicle observation topology map and predicting the acceleration and heading angle of the occluded vehicle based on the relative positions and movement trends of neighboring vehicles; Road geometry constraint correction processing is performed according to the vehicle state data set, and a quadratic programming algorithm is used to constrain the vehicle position to within the lane boundary defined by the map to obtain a collaborative perception data set.

3. The intelligent traffic vehicle behavior modeling method based on collaborative learning according to claim 1 is characterized in that: A vehicle spatiotemporal trajectory function is generated according to the collaborative perception data set, and a collision domain spatiotemporal distribution diagram is obtained based on trajectory intersection time window prediction and dynamic safety radius calculation, including: Generate a vehicle spatiotemporal trajectory function based on the collaborative perception data set and combining the vehicle's historical position and real-time motion state; Performing trajectory intersection prediction processing according to the vehicle space-time trajectory function, and obtaining a trajectory intersection space-time distribution table by calculating the space-time intersection coordinates of multiple vehicle trajectories and the confidence interval of the intersection time window; Performing conflict time window prediction processing according to the trajectory intersection time-space distribution table, and obtaining a high conflict risk time window by calculating the estimated time distribution of vehicles arriving at the intersection and the probability of conflict time window overlap; A dynamic safety radius calculation process is performed according to the high conflict risk time window, and a safety radius threshold is dynamically adjusted based on the vehicle's maximum deceleration, system reaction delay, and sensor detection error parameters to obtain a conflict domain spatiotemporal distribution map.

4. The intelligent traffic vehicle behavior modeling method based on collaborative learning according to claim 1 is characterized in that: According to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, the probability distribution characteristics of vehicle driving intentions and the multi-vehicle competition and cooperation relationship matrix are extracted through the association of temporal behavior patterns and the quantification of interactive game relationships, and a multi-dimensional interactive feature vector is obtained, including: The driving intention classification process is performed according to the vehicle spatiotemporal trajectory function, and the probability distribution characteristics of the driving intention are output by using the acceleration change rate, the heading angle deviation characteristics of the vehicle historical trajectory, and the steering logic constraints of the intersection lane topology as the input values ​​of the long short-term memory network; Modeling the interaction relationship according to the spatial-temporal distribution diagram of the conflict domain, and constructing a competition-cooperation relationship matrix by calculating the spatial-temporal proximity of the conflict domains between vehicles; Calculate the game strategy benefits according to the competition and cooperation relationship matrix, quantify the benefits of each vehicle taking acceleration, deceleration and yielding strategies in the conflict domain, and generate a strategy benefit distribution table; Feature fusion processing is performed according to the probability distribution characteristics and the strategy return distribution table to obtain a multi-dimensional interactive feature vector.

5. The intelligent traffic vehicle behavior modeling method based on collaborative learning according to claim 1 is characterized in that: Behavior modeling is performed based on the multi-dimensional interaction feature vector, and a layered graph network model based on node dynamic feature encoding and edge relationship weight update is constructed to obtain a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles, including: The spatiotemporal graph structure is constructed according to the multi-dimensional interactive feature vector, the vehicle identity labels, the intention probability distribution and the conflict domain association relationship are converted into the node attributes and edge connection relationship of the spatiotemporal graph structure, and the connectivity constraint of the intersection lane topology is combined to obtain a hierarchical spatiotemporal graph structure; Aggregating node features based on the hierarchical spatiotemporal graph structure, and performing bias correction on attention weights in combination with vehicle priority labels to obtain a node encoding feature vector; Iteratively update the edge relationship weights according to the encoded feature vectors, and modify the weights based on the real-time conflict domain risk changes and the yielding habits between vehicles to obtain a dynamic edge relationship matrix; According to the encoded feature vector and the dynamic edge relationship matrix, a vehicle behavior model is constructed through collaborative strategy optimization and strategy space pruning processing constrained by traffic rules.

6. An intelligent traffic vehicle behavior modeling system based on collaborative learning, characterized in that: include: An acquisition module is used to acquire traffic data in a target area, wherein the traffic data includes real-time status data of the current vehicle, road structure data, and neighboring vehicle interaction data; A calibration module, used for performing data calibration processing according to the traffic data, and eliminating abnormal position points outside the lane boundary based on road geometric constraints to obtain a collaborative perception data set; A prediction module, used to generate a vehicle spatiotemporal trajectory function according to the collaborative perception data set, and obtain a spatiotemporal distribution map of the conflict domain based on trajectory intersection time window prediction and dynamic safety radius calculation; An extraction module is used to extract the probability distribution characteristics of vehicle driving intentions and the competition and cooperation relationship matrix of multiple workshops according to the vehicle spatiotemporal trajectory function and the conflict domain spatiotemporal distribution diagram, through the association of temporal behavior patterns and the quantification of interactive game relationships, to obtain a multi-dimensional interactive feature vector; A modeling module is used to perform behavior modeling according to the multi-dimensional interaction feature vector, and obtain a vehicle behavior model that characterizes the interaction behavior rules and decision logic between vehicles by constructing a hierarchical graph network model based on node dynamic feature encoding and edge relationship weight update.

7. The intelligent traffic vehicle behavior modeling system based on collaborative learning according to claim 6 is characterized in that: The calibration module comprises: A first calibration unit is used to perform time synchronization processing according to the traffic data, correct the timestamp offset between vehicles by combining the clock deviation observation data of multiple vehicles, and use the periodic clock calibration signal communicated between vehicles as a time reference to obtain a multi-vehicle state data set; A second calibration unit is used to perform spatial coordinate uniform processing according to the multi-vehicle state data set, map the multi-vehicle position data to a global map coordinate system, and obtain a vehicle position data set; A third calibration unit is used to perform occlusion compensation processing according to the vehicle position data set, and obtain a vehicle state data set by constructing a vehicle observation topology map and predicting the acceleration and heading angle of the occluded vehicle based on the relative positions and movement trends of neighboring vehicles; The fourth calibration unit is used to perform road geometry constraint correction processing according to the vehicle state data set, and use a quadratic programming algorithm to constrain the vehicle position to within the lane boundary defined by the map to obtain a collaborative perception data set.

8. The intelligent traffic vehicle behavior modeling system based on collaborative learning according to claim 6 is characterized in that: The prediction module comprises: A first prediction unit, configured to generate a vehicle spatiotemporal trajectory function based on the collaborative sensing data set and in combination with the vehicle's historical position and real-time motion state; A second prediction unit is used to perform trajectory intersection prediction processing according to the vehicle space-time trajectory function, and obtain a trajectory intersection space-time distribution table by calculating the space-time intersection coordinates of multiple vehicle trajectories and the confidence interval of the intersection time window; A third prediction unit is used to perform conflict time window prediction processing according to the track intersection time-space distribution table, and obtain a high conflict risk time window by calculating the estimated time distribution of the vehicle arriving at the intersection and the conflict time window overlap probability; The fourth prediction unit is used to perform dynamic safety radius calculation processing according to the high conflict risk time window, dynamically adjust the safety radius threshold based on the vehicle's maximum deceleration, system response delay and sensor detection error parameters, and obtain a conflict domain spatiotemporal distribution map.

9. The intelligent traffic vehicle behavior modeling system based on collaborative learning according to claim 6, characterized in that: The extraction module comprises: A first extraction unit is used to perform driving intention classification processing according to the vehicle spatiotemporal trajectory function, and output the probability distribution characteristics of the driving intention by taking the acceleration change rate and heading angle deviation characteristics of the vehicle historical trajectory and the steering logic constraint of the intersection lane topology as the input value of the long short-term memory network; A second extraction unit is used to perform interactive relationship modeling according to the conflict domain spatiotemporal distribution diagram, and to construct a competition and cooperation relationship matrix by calculating the spatiotemporal proximity of the conflict domains between vehicles; A third extraction unit is used to calculate the game strategy benefits according to the competition and cooperation relationship matrix, quantify the benefits of each vehicle taking acceleration, deceleration and yielding strategies in the conflict domain, and generate a strategy benefit distribution table; The fourth extraction unit is used to perform feature fusion processing according to the probability distribution characteristics and the strategy return distribution table to obtain a multi-dimensional interactive feature vector.

10. The intelligent traffic vehicle behavior modeling system based on collaborative learning according to claim 6, characterized in that: The modeling module includes: A first modeling unit is used to construct a spatiotemporal graph structure according to the multi-dimensional interactive feature vector, convert the vehicle identity label, intention probability distribution and conflict domain association relationship into the node attribute and edge connection relationship of the spatiotemporal graph structure, and combine the connectivity constraint of the intersection lane topology to obtain a hierarchical spatiotemporal graph structure; A second modeling unit is used to aggregate node features based on the hierarchical spatiotemporal graph structure, and perform bias correction on attention weights in combination with vehicle priority labels to obtain a node encoding feature vector; A third modeling unit is used to iteratively update the edge relationship weight according to the encoded feature vector, and to modify the weight based on the real-time conflict domain risk change and the yielding habit between vehicles to obtain a dynamic edge relationship matrix; The fourth modeling unit is used to construct a vehicle behavior model based on the encoded feature vector and the dynamic edge relationship matrix through collaborative strategy optimization and strategy space pruning processing constrained by traffic rules.

Images