A method and system for autonomous driving game-theoretic trajectory planning
By assessing the credibility of multi-source perception data and dynamically adjusting the weights, the problem of inaccurate vehicle state judgment in autonomous driving trajectory planning is solved, enabling safe and efficient trajectory planning in complex traffic environments and improving the safety and traffic efficiency of autonomous vehicles.
Patent Information
- Application Number
- CN202611131953.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-28
- Publication Date
- 2026-11-17
AI Technical Summary
Existing autonomous driving trajectory planning methods lack a credibility assessment mechanism when fusing multi-source perception information, resulting in inaccurate vehicle state judgments and difficulty in balancing safety, traffic efficiency, and human-like driving requirements in complex interaction scenarios.
By acquiring multi-source perception data of the vehicle's surrounding environment, including data from the vehicle-mounted perception system and V2X communication, credibility assessment and dynamic weight adjustment are performed. After fusion processing, the state information of the target vehicle is determined, collision risk and behavioral intent are analyzed, and the game decision result is determined by combining historical driving behavior to generate the vehicle's driving trajectory.
It improves the accuracy of target vehicle status acquisition and the ability to predict interactive behavior in complex traffic environments, generates safer, more efficient trajectory planning results that conform to human driving habits, and enhances the safety and traffic efficiency of autonomous vehicles in congested and complex road scenarios.
Smart Images

Figure CN123104685A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of autonomous driving technology, and in particular to an autonomous driving game trajectory planning method and system. Background Technology
[0002] With the continuous increase in car ownership, urban traffic congestion has become an increasingly prominent problem. In congested scenarios, vehicle cutting in and merging into traffic occur frequently, which not only affects road traffic efficiency but also increases the risk of traffic accidents. Existing autonomous driving trajectory planning methods typically acquire surrounding vehicle status information through sensors such as cameras, millimeter-wave radar, and lidar, and combine this with V2X communication technology to obtain information such as the target vehicle's position, speed, and size, in order to improve environmental perception capabilities.
[0003] However, in existing technologies, the fusion of multi-source perception information usually lacks a credibility assessment mechanism, making it difficult to dynamically adjust data weights based on sensor status and communication quality. V2X data may suffer from issues such as latency, packet loss, and positioning drift, affecting the accuracy of vehicle status judgment and leading to conservative decision-making in complex interaction scenarios, making it difficult to balance safety, traffic efficiency, and human-like driving needs. Summary of the Invention
[0004] The purpose of this application is to overcome the above-mentioned problems and provide an autonomous driving game trajectory planning method, control system, storage medium, electronic device and computer program product.
[0005] The technical solution of this application provides an autonomous driving game trajectory planning method, including: Acquire multi-source perception data of the vehicle's surrounding environment. The multi-source perception data includes environmental information acquired by the onboard perception system and target vehicle status information acquired by V2X communication. The fused state information of the target vehicle is determined based on multi-source sensing data; Based on the fusion status information, determine the collision risk information between the vehicle and the target vehicle; Determine the future behavioral intent of the target vehicle based on fused state information; The driving behavior characteristics of the target vehicle are determined based on future behavioral intentions and the target vehicle's historical driving behavior. The game decision outcome is determined based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle. The vehicle's trajectory is generated based on the game decision-making results.
[0006] Furthermore, the fused state information of the target vehicle is determined based on multi-source sensing data, including: Determine the first level of credibility for environmental information and the second level of credibility for target vehicle status information; The fusion weights are determined based on the first and second credibility levels. The environmental information and target state information are fused according to the fusion weight to obtain fused state information.
[0007] Furthermore, determining the primary credibility of environmental information includes: Determine the confidence level of visual detection, the stability of radar tracking, and the spatiotemporal consistency of fusion; The first confidence level is calculated based on visual detection confidence, radar tracking stability, and spatiotemporal consistency of fusion.
[0008] Furthermore, determining the second level of confidence in the target vehicle status information includes: Determine the quality of the communication link, the accuracy of time synchronization, the accuracy of positioning, and the rationality of data movement; The second level of credibility is calculated based on communication link quality, time synchronization accuracy, positioning accuracy, and the rationality of data movement.
[0009] Furthermore, the fusion weights are determined based on the first and second credibility levels, including: The fusion weights corresponding to environmental information and target vehicle status information are determined based on the first and second confidence levels. When the second confidence level is lower than the preset threshold, the fusion weight corresponding to the target vehicle state information is reduced or its participation in the fusion process is stopped.
[0010] Furthermore, determining the fusion weights based on the first and second credibility levels also includes: Get weather information; The fusion weights of environmental information and target vehicle status information are dynamically adjusted based on weather information.
[0011] Furthermore, collision risk information between the vehicle and the target vehicle is determined based on the fused state information, including: Calculate the minimum distance between the vehicle and the target vehicle based on the fusion status information; The collision time between the vehicle and the target vehicle is calculated based on the minimum distance and fused state information. Collision risk information is determined based on minimum distance and collision time.
[0012] Furthermore, the future behavioral intent of the target vehicle is determined based on the fused state information, including: Acquire historical trajectory data of the target vehicle within a preset time period; Construct an intent prediction model based on fused state information; Determine future behavioral intentions based on historical trajectory data and intention prediction models.
[0013] Furthermore, the driving behavior characteristics of the target vehicle are determined based on future behavioral intentions and the target vehicle's historical driving behavior, including: Determine whether the target vehicle will change lanes based on its future behavioral intentions. If lane-changing behavior exists, the driving behavior characteristics of the target vehicle are determined by analyzing historical driving behavior based on an inverse reinforcement learning model.
[0014] Furthermore, determining the outcome of the game decision includes: Determine the interaction behavior parameters corresponding to the target vehicle based on driving behavior characteristics; Determine safety constraint parameters based on collision risk information; The game decision result corresponding to this vehicle is determined based on the interaction behavior parameters, safety constraint parameters, and the interaction relationship between the vehicle and the target vehicle.
[0015] Furthermore, acquiring multi-source perception data of the vehicle's surrounding environment also includes: Perform anomaly detection on the received V2X data; When V2X data suffers from at least one of communication delay, data loss, or location drift, the V2X data is compensated or corrected. Obtain target vehicle status information based on corrected V2X data.
[0016] The technical solution of this application also provides an autonomous driving game trajectory planning system, including: The environmental perception module is used to acquire multi-source perception data of the environment around the vehicle. The multi-source perception data includes environmental information acquired by the on-board perception system and target vehicle status information acquired by V2X communication. The state fusion module is used to determine the fused state information of the target vehicle based on multi-source perception data, and also to determine the collision risk information between the vehicle and the target vehicle based on the fused state information. The intent prediction module is used to determine the future behavioral intent of the target vehicle based on the fused state information; The driving behavior analysis module is used to determine the driving behavior characteristics of the target vehicle based on future behavioral intentions and the target vehicle's historical driving behavior. The game decision-making module is used to determine the game decision result based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle. The trajectory planning module is used to generate vehicle driving trajectories based on the game decision results.
[0017] The technical solution of this application also provides a storage medium that stores computer instructions. When the computer executes the computer instructions, it is used to execute the autonomous driving game trajectory planning method described above.
[0018] The technical solution of this application also provides an electronic device, including at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform an autonomous driving game trajectory planning method as described above.
[0019] The technical solution of this application also provides a computer program product, including a computer program / instruction, characterized in that, when the computer program / instruction is executed by a processor, it implements an autonomous driving game trajectory planning method as described in the previous claim.
[0020] The above technical solution has the following beneficial effects: This application discloses an autonomous driving game-theoretic trajectory planning method and system. It acquires multi-source perception data from an onboard perception system and a V2X communication system, fuses this data to obtain fused state information of the target vehicle, calculates collision risk information between the target and target vehicles based on this fused state information, analyzes the target vehicle's future behavioral intentions, and determines its driving behavior characteristics by combining this with its historical driving behavior. Furthermore, it integrates collision risk information, driving behavior characteristics, and inter-vehicle interaction relationships to determine the game-theoretic decision result, and generates the vehicle's driving trajectory based on this result. This improves the accuracy of target vehicle state acquisition in complex traffic environments and enhances the ability to predict vehicle actions such as cutting in and merging. By generating safer, more efficient, and human-friendly trajectory planning results through a game-theoretic decision-making mechanism, it improves the safety and traffic efficiency of autonomous vehicles in congested and complex road scenarios. Attached Figure Description
[0021] The disclosure of this application will become more readily understood with reference to the accompanying drawings. It should be understood that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this application. In the drawings: Figure 1 This is a flowchart of an autonomous driving game trajectory planning method according to an embodiment of this application; Figure 2 This is a flowchart of an autonomous driving game trajectory planning method according to one embodiment of this application; Figure 3 This is a system architecture diagram of an autonomous driving game trajectory planning system according to an embodiment of this application; Figure 4 This is a schematic diagram of the hardware structure of an electronic device in one embodiment of this application. Detailed Implementation
[0022] The specific embodiments of this application will be further described below with reference to the accompanying drawings.
[0023] It is readily understood that, based on the technical solution of this application, various structural and implementation methods can be interchanged by those skilled in the art without altering the essential spirit of this application. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this application and should not be considered as the entirety of this application or as limitations or restrictions on the technical solution of the application.
[0024] The directional terms such as up, down, left, right, front, back, front, back, top, and bottom mentioned or possibly used in this specification are defined relative to the structures shown in the accompanying drawings. These are relative concepts and may therefore vary depending on their location and usage. Therefore, these or other directional terms should not be interpreted as restrictive. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0025] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meanings of the above in this application according to the specific circumstances.
[0026] like Figure 1 As shown, a flowchart of an autonomous driving game trajectory planning method according to an embodiment of this application is presented, specifically including: Step S101: Acquire multi-source perception data of the vehicle's surrounding environment. The multi-source perception data includes environmental information acquired by the vehicle-mounted perception system and target vehicle status information acquired by V2X communication. Step S102: Determine the fusion state information of the target vehicle based on multi-source sensing data; Step S103: Determine the collision risk information between the vehicle and the target vehicle based on the fused state information; Step S104: Determine the future behavioral intent of the target vehicle based on the fused state information; Step S105: Determine the driving behavior characteristics of the target vehicle based on future behavioral intentions and the target vehicle's historical driving behavior; Step S106: Determine the game decision result based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle; Step S107: Generate the vehicle's driving trajectory based on the game decision results.
[0027] Specifically, in step S101, multi-source perception data of the vehicle's surrounding environment is acquired. This multi-source perception data includes environmental information acquired by the onboard perception system and target vehicle status information acquired through V2X communication. The onboard perception system includes one or more of LiDAR, millimeter-wave radar, cameras, and ultrasonic radar, used to acquire environmental information such as roads, lane lines, obstacles, and other vehicles around the vehicle. V2X communication is used to acquire operational status information broadcast by the target vehicle, such as its position, speed, acceleration, heading angle, and turn signal status. Furthermore, to improve the reliability of V2X data, timestamp verification can be performed on the communication data. When communication delays, data loss, or positioning drift are detected, compensation is performed using methods such as motion model prediction, historical trajectory prediction, or perception data correction to improve the data quality of subsequent fusion processing.
[0028] In step S102, the fused state information of the target vehicle is determined based on the acquired multi-source sensing data. Since the data sources of the vehicle-mounted sensing system and V2X communication are different, and they differ in sampling frequency, timestamps, and measurement accuracy, the various types of data are first synchronized in time and their coordinates are unified. Then, the target vehicle's position, speed, acceleration, heading angle, and vehicle type are combined to perform data association and fusion, obtaining the target vehicle's current fused state information. This provides a unified data foundation for subsequent risk assessment and behavior prediction.
[0029] In step S103, collision risk information between the vehicle and the target vehicle is determined based on the fused state information. Factors such as vehicle distance, relative speed, relative heading angle, target vehicle motion trend, and road topology can be combined to comprehensively assess the collision risk and obtain the corresponding collision risk level. The collision risk information reflects the degree of danger in the current traffic interaction process and serves as a basis for safety constraints in subsequent game-theoretic decisions.
[0030] In step S104, the future behavioral intention of the target vehicle is determined based on the fused state information. The historical motion trajectory sequence of the target vehicle within a preset time window can be extracted, and the trajectory sequence is input into a behavioral intention prediction model to predict the future driving behavior of the target vehicle, thus obtaining the corresponding behavioral intention. The behavioral intention includes one or more states such as normal following, changing lanes to the left, changing lanes to the right, and lane-changing hesitation, used to identify in advance the driving actions that the target vehicle may take.
[0031] In step S105, the driving behavior characteristics of the target vehicle are determined based on its future behavioral intentions and historical driving behavior. When lane-changing behavior is predicted, historical driving behavior data of the target vehicle is further acquired, and combined with behavioral information such as following distance, lateral sway, and lane-changing frequency, the driving style of the target vehicle is analyzed to determine the corresponding driving behavior characteristics. These driving behavior characteristics reflect the aggressiveness or conservatism of the target vehicle's driver and serve as an important basis for subsequent vehicle interaction and game theory.
[0032] In step S106, the game decision result is determined based on collision risk information, driving behavior characteristics, and the interaction relationship between the vehicle and the target vehicle. The interaction behavior parameters corresponding to the target vehicle can be determined based on the driving behavior characteristics, the safety constraint parameters during the vehicle interaction process can be determined based on the collision risk information, and a vehicle interaction game model can be constructed by comprehensively considering the interaction relationship between the vehicle and the target vehicle to determine the game decision result corresponding to the vehicle. The game decision result can be different driving strategies such as slowing down and yielding, maintaining the current driving state, or asserting right-of-way.
[0033] In step S107, the vehicle's driving trajectory is generated based on the game decision results. Based on the determined driving strategy, combined with vehicle kinematic constraints, road boundary constraints, and dynamic obstacle constraints, the future driving trajectory of the vehicle is planned to generate a target driving trajectory that meets the requirements of safety, comfort, and traffic efficiency. The vehicle is then controlled to travel along the target driving trajectory, realizing a closed-loop decision-making process from target vehicle behavior prediction and interactive game to vehicle trajectory planning.
[0034] This application presents an autonomous driving trajectory planning method based on game theory. It acquires multi-source perception data from an onboard perception system and a V2X communication system, fuses this data to obtain fused state information of the target vehicle, calculates collision risk information between the target and target vehicles based on this fused state information, analyzes the target vehicle's future behavioral intentions, and determines its driving behavior characteristics by combining this with its historical driving behavior. Furthermore, it integrates collision risk information, driving behavior characteristics, and inter-vehicle interaction relationships to determine the game theory decision result, and generates the vehicle's driving trajectory based on the game theory decision result. This method improves the accuracy of target vehicle state acquisition in complex traffic environments and enhances the ability to predict vehicle actions such as cutting in and merging. By generating safer, more efficient, and human-friendly trajectory planning results through a game theory decision-making mechanism, it improves the safety and traffic efficiency of autonomous vehicles in congested and complex road scenarios.
[0035] In one embodiment, determining the fused state information of the target vehicle based on multi-source sensing data includes: Determine the first level of credibility for environmental information and the second level of credibility for target vehicle status information; The fusion weights are determined based on the first and second credibility levels. The environmental information and target state information are fused according to the fusion weight to obtain fused state information.
[0036] In this embodiment, environmental information refers to information related to the road environment and target objects around the vehicle acquired through the vehicle-mounted perception system, such as the target vehicle's position, speed, direction of movement, and relative relationship with the vehicle itself. Target vehicle status information refers to operational status data actively sent by the target vehicle through V2X (Vehicle to Everything) communication, such as the vehicle's position, speed, acceleration, and heading angle. Fusion weights are used to characterize the proportion of data from different sources during the fusion process. When the reliability of a certain source of data is high, its corresponding fusion weight is increased; when the reliability of a certain source of data is low, its corresponding fusion weight is decreased.
[0037] In this embodiment, the target vehicle status information comes from both the vehicle-mounted perception system and the V2X communication system. However, different data sources may have detection errors, communication delays, or environmental interference during actual operation. Therefore, instead of simply superimposing the two types of data, the reliability of the data from different sources is first evaluated.
[0038] Specifically, the first level of confidence is determined based on environmental information. The first level of confidence reflects the accuracy of the target vehicle information currently acquired by the vehicle-mounted camera, millimeter-wave radar and other sensing devices. For example, when the vehicle is in a well-lit and clear road environment, the vehicle-mounted sensing system can stably identify the target vehicle, and the corresponding first level of confidence is high. When the vehicle is in a complex environment such as rain, fog or obstruction, the sensing results may be biased, and the corresponding first level of confidence is lower.
[0039] A second level of credibility is determined based on the target vehicle status information. This second level of credibility reflects the reliability of V2X communication data. For example, factors such as communication latency, data integrity, and positioning stability are used to determine whether V2X data is suitable for the current status.
[0040] After determining the primary and secondary credibility levels, the corresponding fusion weights are determined based on their relative relationship, ensuring that data from different sources participate in the fusion process according to their respective reliability. The influence of different data sources is dynamically adjusted based on current environmental changes: when the data acquired by the vehicle-mounted perception system is more reliable, the fusion ratio corresponding to environmental information is increased; when V2X communication data is more stable, the fusion ratio corresponding to the target vehicle's state information is increased.
[0041] Subsequently, based on the determined fusion weights, the environmental information and the target vehicle state information are fused to obtain the fused state information of the target vehicle. By fusing information from different sources, the impact of errors from a single data source on the judgment of the target vehicle's state can be reduced, making the final target vehicle state more accurate and providing a more reliable data foundation for subsequent vehicle behavior analysis and interactive decision-making.
[0042] This embodiment introduces a reliability-based dynamic fusion mechanism, enabling vehicles to select more reliable information sources based on the actual operating environment. Compared to directly using a single perception result or a single V2X data, it can reduce the impact of sensor false detections, missed detections, and communication anomalies on the judgment of the target vehicle's status, and improve the accuracy and stability of target vehicle status acquisition in complex traffic environments.
[0043] In one embodiment, determining the first credibility of the environmental information includes: Determine the confidence level of visual detection, the stability of radar tracking, and the spatiotemporal consistency of fusion; The first confidence level is calculated based on visual detection confidence, radar tracking stability, and spatiotemporal consistency of fusion.
[0044] In this embodiment, visual detection confidence refers to the reliability of the vehicle-mounted visual sensor's identification results of the target vehicle, reflecting the accuracy of the target information currently output by visual detection devices such as cameras. Radar tracking stability refers to the reliability of the millimeter-wave radar's continuous tracking results of the target vehicle, reflecting the stability of the radar's acquisition of the target vehicle's position, speed, and other state information. Fusion spatiotemporal consistency refers to the degree of matching between visual detection results and radar detection results in terms of spatial position and motion change trends. Spatial consistency is used to determine whether the target position obtained by visual detection is close to the position obtained by radar detection, while temporal consistency is used to determine whether the changes in the corresponding target's speed, direction of motion, etc., are consistent.
[0045] In this embodiment, the confidence level of visual detection is first determined. After the vehicle-mounted camera acquires images of the road environment, the target vehicle is identified using a target detection algorithm. The reliability of the visual detection results is evaluated based on information such as the category probability and the degree of target area matching in the target detection results. When the camera can clearly identify the target vehicle and the detection results match the actual shape characteristics of the vehicle, the confidence level of visual detection is high. When there is occlusion, insufficient lighting, or weather influence, the visual detection results may deviate, and the confidence level of visual detection decreases.
[0046] The formula for calculating the confidence score of visual detection is as follows: ; in, This represents the probability of the target vehicle's existence output by the target detection model, with a value range of [0,1]. The aspect ratio of the predicted bounding box output by the object detection model; The preset aspect ratio for the corresponding vehicle category.
[0047] The target detection model's output is further validated by comparing the difference between the predicted bounding box aspect ratio and the prior aspect ratio corresponding to the normal vehicle shape. When the target vehicle has no significant deformation and minimal occlusion, the predicted bounding box aspect ratio is close to the vehicle's prior aspect ratio, resulting in a high correction coefficient and maintaining high visual detection confidence. However, when the target vehicle is occluded, the detection area is abnormal, or the target shape does not conform to vehicle characteristics, the deviation between the predicted bounding box aspect ratio and the prior ratio increases, leading to a decrease in visual detection confidence.
[0048] Subsequently, the radar tracking stability was determined. The millimeter-wave radar continuously tracks the target vehicle by acquiring information such as distance and speed. The reliability of the radar tracking results was judged based on whether the target vehicle remained continuously present, whether the target state changes were stable, and the magnitude of the state estimation error. When the target trajectory output by the radar was continuous and had small fluctuations, the radar tracking stability was high; when environmental interference caused the target to be lost or the measurement error increased, the radar tracking stability decreased.
[0049] The formula for calculating radar tracking stability is as follows: ; in, The probability of target existence output by the millimeter-wave radar, with a value range of [0,1]. This is the covariance matrix corresponding to the target state estimation in the Kalman filter, where the target state includes the position and velocity information of the target vehicle. The trace of the covariance matrix is used to reflect the degree of uncertainty in the target state estimation; This is a preset attenuation coefficient used to adjust the degree of influence of state estimation error on radar tracking stability.
[0050] When the millimeter-wave radar can continuously detect the target vehicle and the target state estimation error output by the Kalman filter is small, the probability of target presence is high, and the trace of the covariance matrix is small, resulting in high radar tracking stability and indicating more reliable radar tracking results. When the target vehicle is affected by environmental interference, increased radar measurement error, or fluctuations in tracking state, the probability of target presence decreases, and the state estimation covariance increases, leading to a decrease in radar tracking stability.
[0051] Furthermore, determining the spatiotemporal consistency of the fusion is crucial. Since visual detection and radar detection are different types of sensing methods, even if both sensors detect a target, it is necessary to determine whether the detection results correspond to the same target vehicle. Therefore, by comparing the positional relationship and velocity change relationship between the visual detection results and the radar detection results, the degree of matching between the two types of data in space and time is assessed. If the target position detected by vision is close to the position detected by radar, and their velocity change trends are consistent, it indicates a high degree of spatiotemporal consistency; otherwise, it indicates that there may be a discrepancy between the two detection results.
[0052] The formula for calculating the spatiotemporal consistency of the fusion is as follows: ; in, and These are the target position vectors output by visual detection and radar tracking, respectively, and they are in the same coordinate system after coordinate transformation. and These are the target velocity vectors output by visual detection and radar tracking, respectively. and These are the preset position fusion variance tolerance and velocity fusion variance tolerance, respectively; and This is a coefficient used to adjust the degree of attenuation due to position and velocity differences.
[0053] When the positions corresponding to the visual detection results and the radar tracking results are close, and the target velocity change trends output by both are consistent, the position difference term and the velocity difference term are both small, making the spatiotemporal consistency of the fusion close to 1. This indicates that the visual detection and radar tracking results are highly matched, and both can be used together to describe the target vehicle's state. When there is a significant positional deviation or inconsistent velocity change between the visual detection results and the radar tracking results, the corresponding positional difference or velocity difference increases, reducing the spatiotemporal consistency of the fusion, indicating that there may be a conflict between the two types of perception information.
[0054] Finally, based on the comprehensive calculation of visual detection confidence, radar tracking stability, and spatiotemporal consistency of data fusion, the first confidence level is obtained. By comprehensively considering multiple influencing factors, the overall output of the vehicle perception system is evaluated, which can make the first confidence level more comprehensively reflect the reliability of the current environmental information and provide a basis for subsequent multi-source data fusion.
[0055] The formula for calculating the first level of credibility is as follows: ; Among them, the overall credibility of the vehicle perception system is the first credibility. For visual detection confidence; For radar tracking stability; To achieve spatiotemporal consistency; , , These are the weight coefficients of the corresponding evaluation indicators, and the sum of the three is equal to 1.
[0056] In this embodiment, the weighting coefficients can be preset or dynamically adjusted based on the performance characteristics of different sensors and the actual operating environment. For example, in scenarios with good lighting conditions and high camera resolution, the weight corresponding to the visual detection confidence level can be appropriately increased; in environments such as rain, snow, or fog that affect the visual detection effect, the weight corresponding to the visual detection confidence level is reduced, while the weight corresponding to the millimeter-wave radar tracking stability is increased, so that the first confidence level can adapt to changes in perception reliability under different operating environments.
[0057] In another embodiment, determining the second credibility of the target vehicle status information includes: Determine the quality of the communication link, the accuracy of time synchronization, the accuracy of positioning, and the rationality of data movement; The second level of credibility is calculated based on communication link quality, time synchronization accuracy, positioning accuracy, and the rationality of data movement.
[0058] In this embodiment, communication link quality represents the stability of data transmission during V2X communication and can be evaluated using parameters such as communication success rate, data packet loss rate, and signal strength. Time synchronization accuracy represents the consistency between the vehicle's system time and the V2X data generation time. Positioning accuracy represents the deviation between the target vehicle location information acquired by V2X and the actual location of the target vehicle. Data motion rationality is used to determine whether the target vehicle state information acquired by V2X conforms to the actual motion patterns of the vehicle.
[0059] In this embodiment, firstly, communication link quality information during V2X communication is acquired. By analyzing parameters such as packet loss and communication stability during data transmission, the integrity and reliability of the current V2X data are determined. Subsequently, time information in the V2X data is acquired and, combined with the current time reference of the vehicle, the time synchronization accuracy of the target vehicle's status data is determined to determine whether there is a significant delay in the data. Simultaneously, based on the target vehicle's location data provided by V2X, its positioning error is analyzed to determine the reliability of the target vehicle's spatial location. Furthermore, based on the changes in the target vehicle's position, speed, and acceleration, it is determined whether the changes in the target vehicle's status conform to the vehicle's motion patterns.
[0060] Then, a comprehensive calculation is performed on communication link quality, time synchronization accuracy, positioning accuracy, and the rationality of data movement to obtain a second level of reliability. By considering multiple influencing factors, the method of judging data reliability solely based on whether V2X communication successfully receives data can be avoided, allowing the second level of reliability to more comprehensively reflect the true reliability of V2X acquisition of target vehicle status information.
[0061] Specifically, the formula for calculating the quality of a communication link is as follows: ; in, This refers to the quality of the communication link; SNR (Signal-to-Noise Ratio) is the signal-to-noise ratio of the received signal in the current V2X communication. A preset signal-to-noise ratio threshold is used to determine the range of communication quality variation; is the steepness coefficient of the Sigmoid function, used to adjust the degree of influence of signal-to-noise ratio changes on communication quality evaluation results; BER (bit error rate) is the proportion of data errors in the current communication process; This represents the maximum permissible bit error rate.
[0062] This embodiment uses the Sigmoid function to map the signal-to-noise ratio (SNR) factor to a preset range, enabling the communication link quality to smoothly adjust with changes in signal strength. When the SNR is high and the bit error rate (BER) is low in the current communication environment, it indicates that the V2X communication link is stable, and the calculated communication link quality is close to 1. When the SNR is low or the BER is high, it indicates that there is significant interference or abnormal data transmission during the communication process, resulting in a decrease in communication link quality and thus reducing the reliability of V2X data in subsequent fusion processes.
[0063] The formula for calculating time synchronization accuracy is as follows: ; in, For data timeliness, that is, time synchronization accuracy; The time difference between the current system time and the timestamp carried in the V2X message; This is a preset information time decay time constant, used to characterize the degree of effectiveness decay of V2X data over time.
[0064] The calculation formula is as follows: ; in, The current system time. This is the timestamp corresponding to when the V2X message was generated.
[0065] This embodiment evaluates the freshness of V2X data by calculating the time difference between the current time and the time the V2X message was generated. When the time difference is small, it indicates that the received target vehicle status information is closer to the current time and can more accurately reflect the current motion state of the target vehicle. Therefore, the calculated data has high timeliness. As the time difference increases, the V2X data may not be able to accurately reflect the current position, speed, and other state changes of the target vehicle, causing the data timeliness to gradually decrease.
[0066] The formula for calculating positioning accuracy is as follows: ; ; ; in, For positioning accuracy; This refers to the standard deviation of the Global Navigation Satellite System positioning carried in the V2X message, or the uncertainty of the target vehicle's position estimated based on multi-source observation data. A high-precision positioning threshold is preset. When the positioning error is less than or equal to the threshold, the target vehicle's location information is considered to have high reliability. The maximum tolerance error threshold is set. When the positioning error is greater than or equal to this threshold, the target vehicle's position information is considered unusable for subsequent state fusion. This is the attenuation coefficient, used to adjust the rate at which positioning accuracy decreases as positioning error increases.
[0067] This embodiment evaluates the reliability of V2X location data based on the magnitude of the target vehicle's positioning error. When the target vehicle's positioning error is small, meaning the deviation between the actual location of the target vehicle and the location data provided by V2X is small, the calculated positioning accuracy is high, indicating that the spatial location data of the target vehicle is more accurate. As the positioning error gradually increases, the positioning accuracy decreases according to a preset attenuation law, enabling the system to perceive the uncertainty of the V2X positioning data. When the positioning error exceeds the maximum tolerance error threshold, it is considered that the positioning data can no longer accurately reflect the actual location of the target vehicle, and therefore the positioning accuracy is reduced to a lower level.
[0068] The formula for calculating the rationality of data movement is as follows: ; in, To ensure the rationality of data movement; and These are the velocity vector and acceleration vector of the target vehicle broadcast by V2X, respectively. and These are the preset maximum reasonable speed threshold and the maximum reasonable acceleration threshold, respectively; and This is the corresponding penalty coefficient, used to adjust the degree of impact of velocity anomalies and acceleration anomalies on the rationality of data motion.
[0069] This embodiment evaluates the reasonableness of the target vehicle's state information by determining whether the speed and acceleration of the target vehicle acquired through V2X conform to the actual motion patterns of the vehicle. When both the target vehicle's speed and acceleration are within a preset reasonable range, it indicates that the target vehicle's state changes conform to the vehicle's motion patterns, the calculated data motion reasonableness is high, and it suggests that the V2X data does not have obvious anomalies. When the target vehicle's speed or acceleration exceeds the normal motion range of the vehicle, it indicates that the state data may be affected by factors such as positioning drift, communication anomalies, or data errors. Therefore, a penalty function is used to reduce the data motion reasonableness.
[0070] A maximum function is used to extract outliers exceeding a reasonable range, penalizing only those exceeding vehicle motion constraints without lowering the evaluation results for data within the normal range. Simultaneously, an exponential decay method is employed to ensure that the reasonableness of data motion decreases smoothly as the degree of anomaly increases, preventing drastic changes in credibility due to minor single fluctuations and improving the stability of the evaluation results.
[0071] In another embodiment, determining the fusion weight based on a first confidence level and a second confidence level includes: The fusion weights corresponding to environmental information and target vehicle status information are determined based on the first and second confidence levels. When the second confidence level is lower than the preset threshold, the fusion weight corresponding to the target vehicle state information is reduced or its participation in the fusion process is stopped.
[0072] In this embodiment, to avoid the problem that the fixed-ratio fusion method cannot adapt to complex traffic environments, the fusion weights corresponding to environmental information and target vehicle state information are dynamically determined based on the first confidence level and the second confidence level.
[0073] Specifically, the process first obtains the first level of confidence corresponding to the vehicle-mounted perception system and the second level of confidence corresponding to the V2X communication system. Then, the corresponding data fusion ratio is determined based on the relative magnitude of the two levels of confidence. When the first level of confidence is high, it indicates that the target vehicle status obtained by perception devices such as vehicle-mounted cameras and millimeter-wave radar is more reliable. Therefore, the fusion weight corresponding to the environmental information is increased, making the fusion result more dependent on the vehicle-mounted perception data. When the second level of confidence is high, it indicates that the target vehicle status information obtained by V2X is highly reliable. Therefore, the fusion weight corresponding to the target vehicle status information is increased, enabling V2X data to fully assist in vehicle status judgment.
[0074] The dynamic weights can be determined using the following formula: ; in, This is the fused target state vector; and These are the target state vectors output by the vehicle-mounted perception system and the V2X communication system, respectively, including state information such as the target vehicle's position, speed, and heading angle; The highest level of credibility; Second level of credibility; This is a preset minimum positive number, used to avoid the denominator being zero when both confidence levels are zero.
[0075] This embodiment allows data sources with higher credibility to have a greater influence during the fusion process, reducing the impact of data sources with lower credibility on the final result. When the vehicle perception system is affected by rain or fog, causing a decrease in the first credibility, but the V2X communication link is stable and the data is timely, the second credibility is relatively increased. The system automatically increases the fusion weight corresponding to the V2X data, ensuring that the target vehicle status information still maintains a high degree of accuracy.
[0076] Furthermore, when the second confidence level is lower than a preset threshold, the target vehicle status information acquired by V2X may be abnormal, such as communication anomalies, data delays, positioning deviations, or status anomalies. In this case, the fusion weight corresponding to the target vehicle status information can be reduced, or the data can be directly stopped from participating in the fusion process to avoid low-reliability V2X data affecting the final fusion result.
[0077] When the following conditions are met: ; in, As the second level of credibility, As the highest level of credibility, The preset ratio threshold can be, for example, 0.3, indicating that the reliability of V2X data only reaches a certain proportion of the reliability of vehicle perception data.
[0078] After triggering a V2X data weight reduction warning, the system reduces the fusion weight of the V2X data within the current processing cycle, treating the V2X data only as auxiliary reference information and excluding it from vehicle main control decisions. By comparing the relative reliability of the two data sources, the system avoids the problem of erroneous data affecting vehicle control due to a high self-evaluation of one data source but insufficient actual reliability.
[0079] In one embodiment, determining the fusion weight based on a first confidence level and a second confidence level further includes: Get weather information; The fusion weights of environmental information and target vehicle status information are dynamically adjusted based on weather information.
[0080] In this embodiment, weather information refers to external environmental parameters that can reflect the current operating environment of the vehicle, such as weather conditions like snow, fog, and clear skies.
[0081] This embodiment first obtains the weather conditions of the current vehicle environment, for example, by using onboard environmental sensors, a meteorological information interface, or the vehicle's own perception system to determine whether there is rain, snow, fog, or other conditions that may affect the driving environment. Then, based on the impact of different weather conditions on the reliability of the onboard perception system and the V2X communication system, the fusion weights corresponding to the first and second credibility levels are adjusted.
[0082] For example, in clear weather conditions, cameras can acquire relatively clear road images, and millimeter-wave radar can stably detect target vehicles. In this situation, the vehicle perception system is less affected by the environment and can maintain a high data fusion weight, allowing environmental information to fully participate in the assessment of the target vehicle's status. However, in adverse weather conditions such as rain, fog, or snow, the camera's target recognition accuracy may decrease due to changes in lighting, raindrop obstruction, or fog scattering. Even if the currently calculated initial confidence level is not low, there may be a further downward trend. In this case, the system reduces the fusion weight corresponding to environmental information based on weather information, while increasing the fusion weight corresponding to V2X status information with normal communication and reliable data, making the fusion result more dependent on data sources less affected by weather.
[0083] This embodiment introduces weather information to dynamically adjust the fusion weights, which allows the data fusion process to not only consider the reliability of the current data itself, but also to take into account the potential impact of external environmental factors on data reliability, thereby improving the accuracy and stability of target vehicle state estimation under complex weather conditions.
[0084] In another embodiment, before determining the collision risk information between the vehicle and the target vehicle based on the fused state information, the method further includes: Based on the vehicle size, position, and heading angle information in the target vehicle status information, construct a two-dimensional virtual occupancy rectangle or a three-dimensional envelope model corresponding to the target vehicle.
[0085] Before conducting a collision risk assessment, this embodiment first constructs a spatial occupancy model of the target vehicle based on its state information. The length, width, and height of the target vehicle are acquired via V2X communication, along with its current position coordinates and heading angle. The target vehicle's position coordinates are used to determine the positional reference of the occupancy model; the vehicle's length and width are used to determine the two-dimensional occupancy range; the vehicle's height is used to expand and form a three-dimensional spatial range; and the heading angle is used to adjust the orientation of the occupancy model to align it with the actual driving direction of the target vehicle.
[0086] When constructing a two-dimensional virtual occupancy rectangle, the system uses the target vehicle's location point as a reference, generates an initial rectangular area based on the target vehicle's length and width, and rotates the rectangular area according to the target vehicle's heading angle to ensure that the rectangular area accurately represents the actual occupancy range of the target vehicle in the current road direction. When further consideration of vehicle spatial height information is required, the system spatially expands the two-dimensional occupancy rectangle based on the vehicle height parameters to form a three-dimensional envelope model corresponding to the target vehicle.
[0087] In another embodiment, determining collision risk information between the vehicle and the target vehicle based on the fused state information includes: Calculate the minimum distance between the vehicle and the target vehicle based on the fusion status information; The collision time between the vehicle and the target vehicle is calculated based on the minimum distance and fused state information. Collision risk information is determined based on minimum distance and collision time.
[0088] In this embodiment, the collision risk information is an evaluation result describing the likelihood of a collision between the vehicle and the target vehicle during future travel. Collision time refers to the estimated time required for a collision to occur between the vehicle and the target vehicle if the vehicle continues traveling at its current state of motion.
[0089] In this embodiment, the positional relationship between the vehicle and the target vehicle is first obtained based on the fused state information, and the minimum distance between them is calculated. Based on the spatial positions, vehicle dimensions, and heading angles of the vehicle and the target vehicle, the spatial relationship between their occupied areas is determined, and the minimum distance between the occupied areas of the vehicle and the target vehicle is calculated. Calculating the distance based on the actual occupied area of the vehicle avoids errors caused by judging solely based on the distance from the vehicle's center point, thus improving the accuracy of vehicle spatial relationship determination.
[0090] The formula for calculating the minimum distance is as follows: ; in, This is the minimum distance between the vehicle and the target vehicle, used to represent the closest spatial distance between the area occupied by this vehicle and the area occupied by the target vehicle. This refers to the area occupied by this vehicle, which can be a two-dimensional occupied rectangle or a three-dimensional envelope model constructed based on the vehicle's size, position, and heading angle. The area occupied by the target vehicle can be a two-dimensional virtual occupied rectangle or a three-dimensional envelope model constructed based on the target vehicle's length, width, height, position, and heading angle information. This is the area occupied by this vehicle. Any point in space; Area occupied by the target vehicle Any point in space; This represents the Euclidean distance between a point in the space of this vehicle and a point in the space of the target vehicle. This indicates that the pair of points with the smallest distance is selected from all spatial point combinations within the area occupied by this vehicle and the area occupied by the target vehicle.
[0091] Subsequently, based on the calculated minimum distance and the vehicle motion state parameters from the fused state information, the collision time between the vehicle and the target vehicle is calculated. The vehicle speed, target vehicle speed, and changes in target vehicle motion are obtained from the fused state information to determine the relative motion trend between the two vehicles. This, combined with the current minimum distance, is used to predict the time required for them to reach a dangerous approach state. When the relative speed between the vehicle and the target vehicle is high, even if there is still a certain distance between them, they may enter the danger zone in a short time, resulting in a shorter calculated collision time. Conversely, when the relative speed is low or they are far apart, the collision time increases, indicating a lower probability of a collision.
[0092] Specifically, the formula for calculating the collision time is as follows: ; ; Among them, Time To Collision represents the estimated time required for the vehicle to reach a collision state with the target vehicle while continuing to move in its current state of motion; This represents the minimum distance between the vehicle and the target vehicle. The relative speed between the vehicle and the target vehicle represents the approach speed of the two vehicles along their respective directions of travel. The calculation formula is as follows: ; in, For the speed of this vehicle, The target vehicle speed.
[0093] At that time, it indicated that the vehicle's speed was higher than the target vehicle's speed, and there was a tendency for the two vehicles to approach each other. The collision time was then calculated based on the current minimum distance and relative speed. When the collision time is set to infinity, it means that the speed of the vehicle is not higher than that of the target vehicle, and there is no continuous approach trend between the two vehicles. Therefore, it is considered that there is no risk of collision at present.
[0094] Furthermore, to account for the impact of road curvature on the actual vehicle trajectory, the collision time is corrected based on the road curvature to obtain the corrected collision time, as shown in the following formula: ; in, Effective collision time taking into account road curvature; For road curvature; This is the curvature influence coefficient, used to adjust the degree of influence of road curvature on collision time.
[0095] When vehicles travel on curved roads, the actual trajectory of the vehicles is affected by the curvature of the road, which changes the effective approach trend between vehicles. Therefore, by appropriately shortening the effective collision time through curvature correction, the system can identify potential collision risks in advance and improve the safety warning capability under curved road conditions.
[0096] Finally, collision risk information is determined by combining the calculated minimum distance and collision time. The minimum distance and the corrected collision time are compared with preset safety thresholds, and the current risk level of the vehicle is determined based on the comparison results. When the distance between vehicles is large and the collision time meets safety requirements, the situation is considered safe; when the distance gradually decreases or the collision time falls below the preset threshold, a collision risk is identified.
[0097] In one embodiment, the collision risk is classified based on the minimum distance and the corrected collision time.
[0098] When the minimum distance between the vehicle and the target vehicle is greater than the safe distance threshold, and the corrected collision time is greater than the safe collision time threshold, the vehicle is determined to be in a safe state. When the minimum distance between the vehicle and the target vehicle is between the Level 1 warning distance threshold and the safe distance threshold, or when the corrected collision time is between the Level 1 warning collision time threshold and the safe collision time threshold, the vehicle is determined to be in a Level 1 warning state. When the minimum distance between the vehicle and the target vehicle is between the emergency intervention distance threshold and the Level 1 warning distance threshold, or when the corrected collision time is between the emergency intervention collision time threshold and the Level 1 warning collision time threshold, the vehicle is determined to be in a Level 2 warning state. When the minimum distance between the vehicle and the target vehicle is less than or equal to the emergency intervention distance threshold, or when the corrected collision time is less than or equal to the emergency intervention collision time threshold, the vehicle is determined to be in an emergency intervention state.
[0099] In this embodiment, the safe distance threshold, the first-level warning distance threshold, and the emergency intervention distance threshold can be set to 8m, 5m, and 2.5m, respectively; the safe collision time threshold, the first-level warning collision time threshold, and the emergency intervention collision time threshold can be set to 3s, 2s, and 1s, respectively. In practical applications, these thresholds can be adjusted according to vehicle type, road grade, vehicle speed, and traffic environment to meet the risk assessment needs of different application scenarios.
[0100] This embodiment uses both minimum distance and corrected collision time as criteria for collision risk assessment and classifies collision risks. This not only comprehensively reflects the current spatial proximity between vehicles but also indicates their future movement trends, making collision risk information more accurate. Furthermore, by adopting different response strategies based on different risk levels, it avoids false alarms or missed alarms caused by relying on a single risk assessment, improving the accuracy of collision risk identification in complex traffic scenarios and providing more reliable safety constraints for subsequent game-theoretic decision-making and trajectory planning.
[0101] In another embodiment, determining the future behavioral intent of the target vehicle based on fused state information includes: Acquire historical trajectory data of the target vehicle within a preset time period; Construct an intent prediction model based on fused state information; Determine future behavioral intentions based on historical trajectory data and intention prediction models.
[0102] In this embodiment, future behavioral intent refers to the prediction of the target vehicle's possible future driving behavior based on its current state and historical motion patterns. Examples include maintaining a following distance, changing lanes to the left, changing lanes to the right, or hesitating. Historical trajectory data refers to the motion state information of the target vehicle continuously collected within a preset time range, including data on the target vehicle's position, speed, heading angle, and lateral offset as they change over time. The intent prediction model is a model used to predict the target vehicle's future behavioral intent based on its historical motion state. Built upon fused state information, the model's input data integrates information provided by the onboard perception system and the V2X communication system, thereby improving the accuracy of target vehicle behavior prediction.
[0103] In this embodiment, in order to obtain the driving behaviors that the target vehicle may take in advance, the system determines the future behavioral intentions of the target vehicle based on fused state information. Specifically, it first obtains the historical trajectory data of the target vehicle within a preset time period, then constructs an intention prediction model based on the fused state information, and finally inputs the historical trajectory data into the intention prediction model to predict the possible future behaviors of the target vehicle.
[0104] First, historical trajectory data of the target vehicle within a preset time range is acquired. This historical trajectory data describes information such as changes in the target vehicle's position, speed, and direction of travel over a past period. For example, information such as the target vehicle's position coordinates, heading angle, longitudinal velocity, lateral velocity, and offset relative to the lane centerline can be acquired at multiple consecutive sampling times, reflecting the target vehicle's current movement trend through continuous state changes.
[0105] Subsequently, an intent prediction model is constructed based on fused state information. This is because fused state information comprehensively considers environmental information acquired by the onboard perception system and target vehicle state information acquired by V2X communication, thus providing a more complete description of the target vehicle's state for the intent prediction model. When constructing the intent prediction model, the temporal continuity of historical trajectory data can be used to perform temporal analysis of the target vehicle's motion state, thereby establishing a mapping relationship between the target vehicle's historical motion characteristics and future behavioral intentions.
[0106] Specifically, the intent prediction model in this embodiment can be constructed using a Long Short-Term Memory (LSTM) network. This is because the driving behavior of the target vehicle is usually continuous over time. For example, lane-changing behavior usually exhibits changes in speed, heading angle, and lateral offset before the actual lateral movement. Therefore, by using the LSTM model to extract temporal features from historical trajectory data, the behavioral change patterns of the target vehicle in its continuous motion state can be fully explored.
[0107] When constructing the intent prediction model, the input features of the model are first built based on the historical trajectory data of the target vehicle within a preset time period. Specifically, the position, velocity, and attitude information of the target vehicle at multiple sampling moments within the past preset time period can be obtained. For example, the motion state data of the target vehicle over the past 3 seconds is sampled at a frequency of 10Hz, resulting in a historical trajectory sequence of 30 time steps.
[0108] The input features can include spatial features, motion features, and relative features. Spatial features describe the changes in the target vehicle's spatial position, including the coordinates of the vehicle's center point and heading angle; motion features describe the changes in the target vehicle's motion state, including longitudinal velocity, lateral velocity, and yaw rate; relative features describe the relative relationship between the target vehicle and its own vehicle and the lane environment, including the lateral offset and lateral velocity of the target vehicle relative to its own vehicle and the lane centerline, to reflect the target vehicle's tendency to move into adjacent lanes.
[0109] Each time step corresponds to a feature vector containing the aforementioned spatial, motion, and relative features. Multiple consecutive time steps form an input tensor, which is then input into the intention prediction model for temporal analysis.
[0110] Furthermore, historical trajectory data can be input into the intent prediction model. The LSTM network encodes the input historical trajectory features through multiple hidden layers and uses the hidden state vector output at the last time step to represent the current motion trend of the target vehicle. Subsequently, the hidden state vector is mapped through a fully connected layer, and the Softmax activation function is used to output the probability information of the target vehicle corresponding to different driving behaviors.
[0111] In one embodiment, determining future behavioral intentions based on historical trajectory data and an intention prediction model includes: By inputting historical trajectory data into the intent prediction model, probability information of different driving behaviors corresponding to the target vehicle can be obtained; Determine the target vehicle's future behavioral intentions based on probability information.
[0112] Different driving behaviors correspond to different probability values. For example, four behavior categories can be set: normal following, changing lanes to the left, changing lanes to the right, and hesitation. The model output represents the probability that the target vehicle belongs to each behavior category. The calculation formula is as follows: ; in, This represents the probability distribution of the target vehicle corresponding to different driving behaviors; This represents the hidden state vector output at the last time step of the LSTM network. and These represent the weight parameters and bias parameters of the fully connected layer, respectively.
[0113] Furthermore, the future behavioral intention of the target vehicle is determined based on the probability values corresponding to different driving behaviors, specifically selecting the driving behavior with the highest probability as the prediction result for the target vehicle: ; in, Indicates the target vehicle's future behavioral intentions. This represents the probability value corresponding to different driving behaviors.
[0114] The maximum probability value can be used as the confidence level for judging future behavioral intentions and for subsequent vehicle interaction decisions. For example, when the maximum probability value is lower than a preset threshold, it indicates that there is a large degree of uncertainty in the target vehicle's behavior. In this case, the target vehicle's behavior tendency can be judged as hesitant to reduce the impact of misjudgment on subsequent decisions.
[0115] This embodiment combines the target vehicle's historical trajectory data to predict its future behavior trends, enabling autonomous vehicles to perceive the target vehicle's potential interactive behaviors in advance, thus providing a more accurate information basis for subsequent vehicle behavior decisions.
[0116] In another embodiment, the driving behavior characteristics of the target vehicle are determined based on future behavioral intentions and the target vehicle's historical driving behavior, including: Determine whether the target vehicle will change lanes based on its future behavioral intentions. If lane-changing behavior exists, the driving behavior characteristics of the target vehicle are determined by analyzing historical driving behavior based on an inverse reinforcement learning model.
[0117] In this embodiment, driving behavior features refer to characteristic parameters used to describe the driving habits and driving tendencies of the target vehicle. These parameters reflect the behavioral characteristics of the target vehicle driver during vehicle interaction, providing a basis for subsequent vehicle interaction decisions. Inverse reinforcement learning models refer to models that infer the objective function or behavioral preferences behind driving behavior by observing existing driving behavior data.
[0118] In this embodiment, the future behavioral intent obtained from the aforementioned steps is first used to determine whether the target vehicle currently has a clear lane-changing tendency. When the future behavioral intent is to change lanes to the left or right, it indicates that the target vehicle may engage in lateral interaction with the vehicle in the future, such as merging into the lane of the vehicle or weaving in and out of traffic. In this case, it is necessary to further analyze the driving behavior characteristics of the target vehicle in order to predict the interaction methods that the target vehicle may take subsequently.
[0119] If the target vehicle does not change lanes, it means that the target vehicle is currently mainly driving in its original lane and the interaction between it and the vehicle is low. Therefore, there is no need to perform further driving behavior feature analysis, which can reduce the computational load on the system.
[0120] When the system determines that a target vehicle is changing lanes, it analyzes the target vehicle's historical driving behavior based on an inverse reinforcement learning model. Specifically, the inverse reinforcement learning model acquires the target vehicle's historical trajectory data and, based on the vehicle's state changes and corresponding driving actions during its historical movement, learns the reward function corresponding to the target vehicle's driving behavior. ,in, This indicates the status information of the target vehicle. This represents the driving actions taken by the target vehicle. By analyzing the reward values corresponding to different states and action combinations, the potential driving preferences reflected in the target vehicle's driving behavior can be identified.
[0121] Furthermore, the baseline aggression parameters corresponding to the target vehicle are determined based on the reward function output by the inverse reinforcement learning model. The reward value corresponding to the current state-action is mapped to a preset range to obtain the behavior evaluation result based on inverse reinforcement learning. For example, whether the target vehicle frequently uses a small following distance, frequently sways laterally, or tends to quickly change lanes during its historical driving process can be used as a basis for driving behavior feature analysis.
[0122] In one embodiment, to improve the accuracy of driving behavior feature evaluation, a comprehensive analysis can be performed combining the target vehicle's implicit lane-changing behavior and historical driving state changes. When the lateral distance between the target vehicle and the lane line is detected to be less than a preset threshold and the turn signal is not activated, it is determined that the target vehicle has implicit lane-changing behavior, and the corresponding behavior evaluation result is improved.
[0123] In one embodiment, the target vehicle's overall aggression factor α is calculated based on the baseline aggression parameters obtained from inverse reinforcement learning, the implicit lane change event triggering parameters, and the historical behavior correction parameters: ; in, This represents the baseline aggression parameters obtained based on the inverse reinforcement learning reward function mapping; This indicates a hidden lane change event trigger; This represents a correction term derived from historical driving behavior; , , These are the corresponding weighting coefficients.
[0124] The baseline aggression parameter reflects the overall driving style of the target vehicle, the implicit lane change event triggering item reflects whether the target vehicle currently has a tendency to suddenly change lanes, and the historical behavior correction item is used to adjust the evaluation results based on the target vehicle's long-term driving habits.
[0125] This embodiment uses an inverse reinforcement learning model to analyze and map the historical driving behavior of a target vehicle into behavioral parameters that reflect its driving characteristics. For example, the type of driving behavior of the target vehicle can be determined based on the comprehensive aggression factor α. When α is greater than a preset threshold, the target vehicle is determined to have aggressive driving behavior; when α is in the middle range, the target vehicle is determined to have standard driving behavior; and when α is low, the target vehicle is determined to have conservative driving behavior.
[0126] Aggressive driving behavior is typically characterized by a smaller following distance, a higher frequency of lateral movement, and a stronger tendency to change lanes; conservative driving behavior is typically characterized by maintaining a larger safe distance and fewer proactive lateral changes.
[0127] By filtering target vehicles based on future behavioral intentions, it is possible to avoid performing complex behavioral modeling on all target vehicles. Instead, driving behavior analysis is performed only on lane-changing target vehicles that may engage in vehicle interaction, thereby improving system processing efficiency.
[0128] Furthermore, by combining the target vehicle's future behavioral intentions and historical driving behavior characteristics, the system can simultaneously consider the target vehicle's current upcoming behavior and historical driving habits, thereby improving the accuracy of predicting the target vehicle's response mode during subsequent vehicle interaction decision-making processes.
[0129] Furthermore, when making decisions in vehicle interaction games, the comprehensive aggression factor α can be used as an important parameter for predicting the target vehicle's behavioral response. Specifically, the target vehicle's behavioral tendency during the game is judged based on the target vehicle's aggression factor α. When α is high, it indicates that the target vehicle is more inclined to take interactive behaviors such as proactive lane changing and forced merging, thus increasing the probability of the target vehicle taking aggressive response actions; when α is low, it indicates that the target vehicle is more inclined to maintain its current driving state or actively avoid the situation, thus reducing the possibility of it taking aggressive behavior.
[0130] By further incorporating driving behavior characteristics into the game-theoretic decision-making process, the system can consider not only the current motion state of the target vehicle but also its long-term driving style differences when predicting the target vehicle's response behavior. For example, for a target vehicle with a high aggression factor, even if the current collision risk is low, the system can still predict in advance that it may force its way in and choose a more reasonable interaction strategy during the game-theoretic decision-making process; for a target vehicle with a more conservative driving behavior, the avoidance intensity can be appropriately reduced to improve vehicle traffic efficiency.
[0131] Specifically, the formula for calculating the baseline aggression parameter is as follows: ; in, This represents the baseline aggression parameter obtained based on the inverse reinforcement learning reward function mapping, used to reflect the degree of aggression of the target vehicle's current driving behavior relative to human driving behavior data; This represents the reward value output by the inverse reinforcement learning model, where s represents the current driving state of the target vehicle, and a represents the driving action taken by the target vehicle. This reward value is used to characterize the evaluation result of the current state-action combination in human driving behavior data; and The adjustment parameter is used to adjust the degree of change of the reward value mapping to a preset range, where β is used to control the degree of influence of the reward value on the change of the aggression parameter, and γ is used to adjust the offset of the mapping function; This represents the Sigmoid activation function, which maps the reward value output by inverse reinforcement learning to the range [0,1], so that the obtained baseline aggression parameters meet the unified evaluation criteria.
[0132] Furthermore, the formula for calculating the trigger term of the invisible lane change event is as follows: And the turn signal was not turned on. And in other cases ; in, This indicates a hidden lane change event trigger, used to reflect the impact on the aggressiveness of driving behavior when a target vehicle engages in potential lane change behavior without explicit driving signal prompts, such as turning on the turn signal; This indicates the lateral distance between the target vehicle and the current lane boundary, used to describe the degree to which the target vehicle is laterally close to the adjacent lane; This indicates a preset lateral distance threshold. When the lateral distance between the target vehicle and the lane line is less than this threshold, it means that the target vehicle has approached the lane boundary and has a tendency to move into the adjacent lane.
[0133] When the target vehicle meets the conditions of having a lateral distance less than the threshold and not activating its turn signal, it indicates that the target vehicle may be engaging in a hidden lane change without prior notification to other vehicles. In this case, the hidden lane change event trigger is set to a preset value to improve the evaluation result of the target vehicle's aggressive driving behavior. When the above conditions are not met, it is considered that there is no obvious hidden lane change event, and the hidden lane change event trigger is set to 0.
[0134] The formula for calculating the historical behavior correction term is as follows: ; in, This represents the historical behavior correction term, which comprehensively reflects the following habits and lateral movement characteristics of the target vehicle during long-term driving, and corrects the behavior evaluation results obtained based on inverse reinforcement learning. This represents the behavior correction parameter based on following distance, used to reflect the following distance characteristics between the target vehicle and the vehicle in front; This represents the behavior correction parameter based on the lateral sway frequency, used to reflect the degree of change in the lateral motion of the target vehicle.
[0135] The formula for calculating the following distance is as follows: ; in, This indicates the historical following distance of the target vehicle, used to describe the time distance between the target vehicle and the vehicle in front. This represents the attenuation constant, used to adjust the degree of influence of following distance on the behavior correction results.
[0136] When the following distance between the target vehicle and the vehicle in front is small, it indicates that the target vehicle is close to the vehicle in front and has a strong tendency to follow. The calculated distance under these conditions is... A larger following distance indicates more aggressive driving behavior from the target vehicle; a larger following distance also indicates that the target vehicle is maintaining a greater safe distance. reduce.
[0137] Furthermore, the lateral oscillation frequency term is normalized using the Sigmoid function, and its calculation formula is as follows: ; in, The value represents the number of times the target vehicle's lateral displacement crosses zero per unit time, i.e., the vehicle's lateral sway frequency, which is used to describe the lateral stability of the target vehicle; k represents the scaling factor, which is used to adjust the degree of influence of the lateral sway frequency on the evaluation results.
[0138] This embodiment uses normalization to map the lateral sway frequency to the range [0,1]. When the target vehicle's lateral sway frequency is high, it indicates that the vehicle's lateral movement changes more frequently, and the driving behavior is more proactive or aggressive. Increase; when the lateral oscillation frequency is low, it indicates that the target vehicle's driving state is relatively stable, and the corresponding correction parameter is reduced.
[0139] In another embodiment, determining the game decision outcome includes: Determine the interaction behavior parameters corresponding to the target vehicle based on driving behavior characteristics; Determine safety constraint parameters based on collision risk information; The game decision result corresponding to this vehicle is determined based on the interaction behavior parameters, safety constraint parameters, and the interaction relationship between the vehicle and the target vehicle.
[0140] In this embodiment, the interaction behavior parameters refer to parameters used to describe the behavioral tendencies that the target vehicle may take during vehicle interaction; the safety constraint parameters refer to the constraints used to limit the safety of the vehicle's decision-making results, which are determined based on collision risk information; the game decision result refers to the optimal driving strategy of the vehicle obtained by analyzing the possible driving behaviors of both parties and selecting the optimal driving strategy of the vehicle, taking into account the mutual influence between the vehicle and the target vehicle.
[0141] In this embodiment, interaction behavior parameters are first determined based on the driving behavior characteristics of the target vehicle. These driving behavior characteristics are then used to determine the target vehicle's behavioral tendencies during vehicle interactions. For example, if the target vehicle corresponds to a higher driving aggression parameter, it indicates that the target vehicle is more inclined to proactively change lanes, shorten following distances, or take aggressive merging actions; in this case, the influence of the target vehicle's proactive interaction behavior is increased. Conversely, if the target vehicle's driving behavior is more conservative, the influence of its proactive interaction behavior on the vehicle's decision-making is reduced.
[0142] Subsequently, safety constraint parameters are determined based on collision risk information, which reflects the degree of risk of current and future collisions between the vehicle and the target vehicle. When the collision risk is low, the system can allow for more diverse interaction strategies, such as maintaining the current speed or making slight yielding maneuvers; when the collision risk is high, safety constraint requirements need to be increased, making the vehicle's decisions more inclined towards safe behaviors such as deceleration and avoidance.
[0143] Further game theory analysis is conducted by comprehensively considering interactive behavior parameters, safety constraint parameters, and the interaction relationship between the vehicle and the target vehicle. During vehicle interaction, the driving actions of the current vehicle will affect the behavior of the target vehicle, and the behavior of the target vehicle will in turn affect the driving state of the current vehicle. Therefore, decisions cannot be made based solely on the state of a single vehicle. This embodiment analyzes the possible responses of the target vehicle after the current vehicle takes different actions, and combines the behavioral gains of both parties with safety constraints to determine a more reasonable game theory decision in the current scenario.
[0144] The formula for calculating the interaction behavior parameters is as follows: ; in, This represents the probability that the target vehicle will choose the jjjth response action when the vehicle takes the iiith action. It is used to describe the possible behavioral responses of the target vehicle to different driving strategies of the vehicle. This represents the rationality coefficient of the target vehicle's decision-making, used to adjust the degree to which the target vehicle's payoff value influences the probability of its behavioral choice. When When the value is larger, it indicates that the target vehicle is more likely to choose a response action with higher returns; when A smaller value indicates greater randomness in the target vehicle's behavior; This represents the payoff value of the target vehicle when it takes the jjj-th response action in the current game process, and is used to evaluate the overall payoff that the target vehicle can obtain after taking different response actions.
[0145] Among them, the revenue value of the target vehicle's corresponding response action. The calculation formula is as follows: ; in, This indicates the expected following distance that the target vehicle can achieve after taking the current response action, reflecting the spatial gain of the target vehicle after taking this action; This represents the target following distance that the target vehicle expects to maintain, and is used as a benchmark for evaluating spatial benefits. The closer the actual spatial distance obtained by the target vehicle after taking a certain response action is to or exceeds the expected distance, the higher the corresponding spatial benefit. This indicates the estimated time required for the target vehicle to complete the lane change, reflecting the time cost required for the target vehicle to complete the interactive behavior. This represents the time decay constant, used to adjust the degree of influence of lane change time on the benefit evaluation results. The shorter the lane change time, the higher the time benefit. This represents the target vehicle aggression factor determined based on driving behavior characteristics, used to describe the degree to which the target vehicle tends to actively interact or aggressively change lanes. When α is high, it indicates that the target vehicle is more likely to take proactive or even forceful interactive behaviors; This represents the risk suppression coefficient determined based on collision risk information, reflecting the limiting effect of the current safety status on the behavioral benefits of the target vehicle. When the collision risk is high, This reduces the expected benefits of the target vehicle engaging in high-risk interactive behaviors; The indicator function takes a value of 1 when the target vehicle selects the "forced lane cutting" response action, and a value of 0 otherwise, to further reflect the behavioral tendency influence of the forced lane cutting behavior; , , These represent the weighting coefficients corresponding to spatial gains, temporal gains, and behavioral tendency gains, respectively, and the sum of the three equals 1.
[0146] This embodiment comprehensively considers the spatial and temporal gains of the target vehicle when it takes different response actions, as well as its own driving behavior tendencies, to predict the probability of the target vehicle's possible response after the vehicle takes different actions. Compared to judging solely based on the target vehicle's current position or speed, it can dynamically predict the interaction process between vehicles by combining the target vehicle's driving style and current safety status, providing a basis for subsequent game-theoretic decisions.
[0147] In another embodiment, acquiring multi-source perception data of the vehicle's surrounding environment further includes: Perform anomaly detection on the received V2X data; When V2X data suffers from at least one of communication delay, data loss, or location drift, the V2X data is compensated or corrected. Obtain target vehicle status information based on corrected V2X data.
[0148] In this embodiment, anomaly detection refers to the process of assessing the reliability of received V2X data to identify unexpected situations during data transmission. Compensation or correction refers to restoring or adjusting missing, delayed, or biased data based on existing data and the target vehicle's motion patterns to make the corrected data closer to the actual state of the target vehicle.
[0149] In this embodiment, the target vehicle status data transmitted via V2X communication is first received, and anomaly detection is performed on the data. Since V2X data may be affected by factors such as communication distance, network load, and signal interference during actual transmission, the received data may not fully reflect the current status of the target vehicle. Therefore, it is necessary to first determine whether there are any anomalies in the current V2X data.
[0150] Specifically, it can detect and handle abnormal situations such as V2X communication delay, data loss, and location drift.
[0151] During anomaly detection, the existence of communication delays can be determined based on the time information of V2X data. For example, by comparing the timestamp in the V2X data with the current system time, if the time difference between the two exceeds a preset threshold, it indicates that the data may no longer accurately reflect the current state of the target vehicle, and a communication delay is determined to exist.
[0152] When a communication delay is detected in V2X data, time compensation can be performed on the target vehicle's state based on its historical motion state and the currently received vehicle motion parameters. For example, based on the delay time and the target vehicle's current speed, the target vehicle's position and motion state can be extrapolated to the current moment, making the obtained target vehicle state closer to its actual state at the current moment.
[0153] When V2X data loss is detected, the missing status can be predicted and supplemented based on the target vehicle's existing historical trajectory data. For example, based on information such as the target vehicle's position change trend and speed change over a period of time, the current position status of the target vehicle can be predicted, thereby ensuring the continuity of the target vehicle's status information.
[0154] Specifically, when some V2X data is not successfully received, the system can use the target vehicle's historical trajectory sequence to fit short-term motion trends. For example, it can fit the historical trajectory using the least squares method, or use a long short-term memory network to learn the historical state change patterns, and predict the target vehicle's position and motion state at the current moment based on the existing trajectory. This achieves state compensation in the event of data loss and avoids interruption of subsequent vehicle state judgment due to data loss.
[0155] When location drift is detected in V2X data, the location data provided by V2X can be corrected. For example, the location of the target vehicle acquired by V2X can be compared with the location detected by the vehicle perception system, and the deviation between the two can be used to determine whether there is a location anomaly. When the deviation exceeds a preset range, the impact of the abnormal location data can be reduced, or the target vehicle location can be corrected by combining the vehicle perception results.
[0156] The location of the target vehicle can be broadcast via V2X. The target vehicle position detected by the vehicle-mounted perception system The formula for calculating the positional deviation between the two is as follows: ; in, This represents the position residual vector between V2X data and vehicle-mounted perception data; This represents the target vehicle position vector broadcast by V2X; This represents the target vehicle position vector detected by the vehicle-mounted perception system.
[0157] Furthermore, the position residual vector is decomposed into longitudinal residuals along the vehicle's direction of travel. And the lateral residual perpendicular to the vehicle's direction of travel ,in, Used to represent the longitudinal positional deviation of the target vehicle, in meters (m). Used to represent the lateral positional deviation of the target vehicle, measured in meters (m).
[0158] The formula for calculating the weighted composite residual is as follows: ; Where E represents the weighted composite residual, which is used to reflect the overall degree of deviation between V2X location data and vehicle-sensed location data; and represent the weighting coefficients for the longitudinal and lateral directions, respectively. Since lateral position deviation of a vehicle directly affects lane change judgment and collision risk assessment, a lateral weight can be set. Greater than the vertical direction weight This is to improve the system's sensitivity to lateral positioning anomalies.
[0159] When the calculated weighted composite residual E exceeds a preset threshold, it indicates a significant deviation between the V2X broadcast location data and the vehicle-mounted sensing results, potentially indicating a positioning drift anomaly. In this case, the impact of the V2X location data in subsequent processing can be reduced, or the location data acquired by the vehicle-mounted sensing system can be prioritized to correct the target vehicle's state, thereby preventing abnormal positioning information from affecting the accuracy of the target vehicle's state assessment.
[0160] After anomaly detection and corresponding compensation correction, the target vehicle's state information is obtained based on the corrected V2X data. Because the corrected V2X data has eliminated some communication anomalies, the obtained target vehicle state information can more accurately describe the target vehicle's current motion state, providing a more reliable data foundation for subsequent multi-source data fusion.
[0161] like Figure 2 As shown, a flowchart of an autonomous driving game trajectory planning method in one embodiment of this application is illustrated, specifically including: Step S201: Acquire multi-source perception data of the vehicle's surrounding environment. The multi-source perception data includes environmental information acquired by the vehicle-mounted perception system and target vehicle status information acquired by V2X communication. Step S202: Determine the visual detection confidence, radar tracking stability, and spatiotemporal consistency of the fusion. Step S203: Calculate the first confidence level based on the visual detection confidence level, radar tracking stability, and spatiotemporal consistency of the fusion. Step S204: Determine the communication link quality, time synchronization accuracy, positioning accuracy, and the rationality of data movement; Step S205: Calculate the second level of confidence based on communication link quality, time synchronization accuracy, positioning accuracy, and data movement rationality; Step S206: Determine the fusion weights corresponding to the environmental information and the target vehicle state information based on the first confidence level and the second confidence level; Step S207: Perform fusion processing on environmental information and target state information according to fusion weights to obtain fused state information; Step S208: Calculate the minimum distance between the vehicle and the target vehicle based on the fused state information; Step S209: Calculate the collision time between the vehicle and the target vehicle based on the minimum distance and fused state information; Step S210: Determine collision risk information based on minimum distance and collision time; Step S211: Obtain historical trajectory data of the target vehicle within a preset time period; Step S212: Construct an intent prediction model based on fused state information; Step S213: Determine future behavioral intentions based on historical trajectory data and the intention prediction model; Step S214: Determine whether the target vehicle will change lanes based on its future behavioral intentions; Step S215: If lane-changing behavior exists, analyze historical driving behavior based on the inverse reinforcement learning model to determine the driving behavior characteristics of the target vehicle. Step S216: Determine the interaction behavior parameters corresponding to the target vehicle based on the driving behavior characteristics; Step S217: Determine safety constraint parameters based on collision risk information; Step S218: Determine the game decision result corresponding to this vehicle based on the interaction behavior parameters, safety constraint parameters, and the interaction relationship between the vehicle and the target vehicle; Step S219: Generate the vehicle's driving trajectory based on the game decision results.
[0162] like Figure 3 As shown, embodiments of this application also provide an autonomous driving game trajectory planning system, including: The environmental perception module 301 is used to acquire multi-source perception data of the environment around the vehicle. The multi-source perception data includes environmental information acquired by the on-board perception system and target vehicle status information acquired by V2X communication. The state fusion module 302 is used to determine the fused state information of the target vehicle based on multi-source perception data, and also to determine the collision risk information between the vehicle and the target vehicle based on the fused state information. Intent prediction module 303 is used to determine the future behavioral intent of the target vehicle based on fused state information; The driving behavior analysis module 304 is used to determine the driving behavior characteristics of the target vehicle based on future behavioral intentions and the target vehicle's historical driving behavior. The game decision module 305 is used to determine the game decision result based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle. The trajectory planning module 306 is used to generate vehicle driving trajectories based on the game decision results.
[0163] This embodiment of an autonomous driving game-theoretic trajectory planning system includes an environment perception module 301, a state fusion module 302, an intent prediction module 303, a driving behavior analysis module 304, a game-theoretic decision-making module 305, and a trajectory planning module 306. The environment perception module 301 acquires multi-source perception data of the vehicle's surrounding environment. The onboard perception system can acquire information about the vehicle's surrounding environment through sensors such as cameras and millimeter-wave radar, while the V2X communication system can acquire target vehicle state information sent by other vehicles. By simultaneously acquiring onboard perception information and V2X communication information, target vehicle-related information can be obtained from both the vehicle's own perception and vehicle-road cooperation perspectives, improving the completeness of environmental information acquisition.
[0164] The state fusion module 302 is used to fuse the multi-source perception data acquired by the environment perception module 301 to obtain the fused state information of the target vehicle, and further determine the collision risk information between the vehicle and the target vehicle based on the fused state information. During the fusion process, the corresponding fusion weights can be dynamically adjusted according to the reliability of different data sources, so that data with higher reliability has a greater impact on the fusion result, thereby reducing the impact of anomalies in a single data source on the judgment of the target vehicle's state. At the same time, based on the fused target vehicle's position, speed, and motion trend information, the distance between vehicles and the collision time are analyzed to determine the current collision risk state.
[0165] The intent prediction module 303 is used to determine the future behavioral intent of the target vehicle based on fused state information. This module analyzes the target vehicle's historical trajectory and current state to predict possible future driving behaviors, such as maintaining following distance, changing lanes, or other interactive behaviors. By predicting the target vehicle's future behavior in advance, autonomous vehicles can respond before the target vehicle actually performs its actions, improving the proactiveness of the vehicle interaction process.
[0166] The driving behavior analysis module 304 is used to determine the driving behavior characteristics of the target vehicle based on future behavioral intentions and the target vehicle's historical driving behavior. This module extracts the target vehicle's driving habits and behavioral tendencies, such as aggressive, standard, or conservative driving behaviors, by analyzing the target vehicle's historical driving trajectory. By incorporating the target vehicle's driving behavior characteristics, the system can consider not only the target vehicle's current state but also its long-term driving characteristics, improving the accuracy of predicting the target vehicle's subsequent response behavior.
[0167] The game theory decision-making module 305 determines the game theory decision outcome based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle. This module establishes an interaction model between vehicles, comprehensively considering the possible response behaviors of the target vehicle and the safety and benefits of different driving strategies to determine the driving decision applicable to the current traffic scenario. Through game theory analysis, autonomous vehicles can avoid making decisions solely based on their own state, enabling them to dynamically adjust according to the behavioral characteristics of other traffic participants.
[0168] The trajectory planning module 306 generates the vehicle's driving trajectory based on the game decision results output by the game decision module 305. This module converts the decision results into the actual driving path executed by the vehicle, enabling the vehicle to perform operations such as deceleration, avoidance, maintaining driving, or interaction according to the expected strategy.
[0169] The autonomous driving game-theoretic trajectory planning system in this embodiment can integrate multi-source information provided by vehicle perception and V2X communication to improve the accuracy of target vehicle state acquisition; enhance the ability to predict complex traffic interaction behaviors through future behavioral intention prediction and driving behavior feature analysis; and further combine collision risk assessment and game-theoretic decision-making mechanisms to enable autonomous vehicles to select more reasonable driving strategies according to the characteristics of different traffic participants, thereby improving the safety, traffic efficiency and rationality of autonomous vehicles in complex road environments.
[0170] Embodiments of this application also provide a storage medium that stores computer instructions, which, when executed by a computer, are used to perform an autonomous driving game trajectory planning method as described in any of the preceding embodiments.
[0171] Figure 4 An electronic device according to this application is shown, comprising: At least one processor 401; and, Memory 402 is communicatively connected to at least one processor 401; wherein, The memory 402 stores instructions that can be executed by at least one processor 401, which enables the at least one processor 401 to perform all the steps of an autonomous driving game trajectory planning method in any of the foregoing method embodiments.
[0172] Figure 4 Taking a processor 401 as an example: The electronic device may also include an input device 403 and an output device 404.
[0173] The processor 401, memory 402, input device 403 and output device 404 can be connected by a bus or other means. The figure shows an example of connection by bus.
[0174] Memory 402, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to an autonomous driving game trajectory planning method in this embodiment of the application. Figure 1 and Figure 2 The method flow is shown. The processor 401 executes various functional applications and data processing by running non-volatile software programs, instructions, and modules stored in the memory 402, thereby implementing an autonomous driving game trajectory planning method in the above embodiment.
[0175] Memory 402 may include a program storage area and a data storage area. The program storage area may store an operating system and an application program required for at least one function. The data storage area may store data created using an autonomous driving game trajectory planning method. Furthermore, memory 402 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 402 may optionally include memory remotely located relative to processor 401, and these remote memories may be connected via a network to the apparatus performing an autonomous driving game trajectory planning method. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0176] The input device 403 can receive user clicks and generate signal inputs related to user settings and function control for the autonomous driving game trajectory planning method. The output device 404 may include a display device such as a screen.
[0177] One or more modules are stored in memory 402. When run by one or more processors 401, an autonomous driving game trajectory planning method in any of the above method embodiments is executed.
[0178] Embodiments of this application also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement an autonomous driving game trajectory planning method as described in any of the preceding embodiments.
[0179] As needed, the above technical solutions can be combined to achieve the best technical effect.
[0180] The above are merely the principles and preferred embodiments of this application. It should be noted that, for those skilled in the art, implementation methods obtained by appropriately combining the technical solutions disclosed in different embodiments are also included within the technical scope of this invention. Based on the principles of this application, several other modifications can also be made, which should also be considered within the protection scope of this application.
Claims
1. A method for autonomous driving game-theoretic trajectory planning, characterized in that, include: Acquire multi-source perception data of the vehicle's surrounding environment. The multi-source perception data includes environmental information acquired by the onboard perception system and target vehicle status information acquired by V2X communication. The fused state information of the target vehicle is determined based on multi-source sensing data; Based on the fusion status information, determine the collision risk information between the vehicle and the target vehicle; Determine the future behavioral intent of the target vehicle based on fused state information; The driving behavior characteristics of the target vehicle are determined based on future behavioral intentions and the target vehicle's historical driving behavior. The game decision outcome is determined based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle. The vehicle's trajectory is generated based on the game decision results.
2. The method according to claim 1, characterized in that, The fused state information of the target vehicle is determined based on multi-source sensing data, including: Determine the first level of credibility for environmental information and the second level of credibility for target vehicle status information; The fusion weights are determined based on the first and second credibility levels. The environmental information and target state information are fused according to the fusion weights to obtain fused state information.
3. The method according to claim 2, characterized in that, Determining the primary credibility of environmental information includes: Determine the confidence level of visual detection, the stability of radar tracking, and the spatiotemporal consistency of fusion; The first confidence level is calculated based on visual detection confidence, radar tracking stability, and spatiotemporal consistency of fusion.
4. The method according to claim 2, characterized in that, Determining the second level of confidence in the target vehicle's status information includes: Determine the quality of the communication link, the accuracy of time synchronization, the accuracy of positioning, and the rationality of data movement; The second level of credibility is calculated based on communication link quality, time synchronization accuracy, positioning accuracy, and the rationality of data movement.
5. The method according to claim 2, characterized in that, The fusion weights are determined based on the first and second confidence levels, including: The fusion weights corresponding to environmental information and target vehicle status information are determined based on the first and second confidence levels. When the second confidence level is lower than the preset threshold, the fusion weight corresponding to the target vehicle state information is reduced or its participation in the fusion process is stopped.
6. The method according to claim 5, characterized in that, Determining the fusion weight based on the first and second confidence levels also includes: Get weather information; The fusion weights of environmental information and target vehicle status information are dynamically adjusted based on weather information.
7. The method according to claim 1, characterized in that, Based on the fusion status information, collision risk information between the vehicle and the target vehicle is determined, including: Calculate the minimum distance between the vehicle and the target vehicle based on the fusion status information; The collision time between the vehicle and the target vehicle is calculated based on the minimum distance and fused state information. Collision risk information is determined based on minimum distance and collision time.
8. The method according to claim 1, characterized in that, Determining the future behavioral intent of the target vehicle based on fused state information includes: Acquire historical trajectory data of the target vehicle within a preset time period; Construct an intent prediction model based on fused state information; Determine future behavioral intentions based on historical trajectory data and intention prediction models.
9. The method according to claim 1, characterized in that, The driving behavior characteristics of the target vehicle are determined based on future behavioral intentions and the target vehicle's historical driving behavior, including: Determine whether the target vehicle is attempting a lane change based on its future behavioral intentions. If lane-changing behavior exists, the driving behavior characteristics of the target vehicle are determined by analyzing historical driving behavior based on an inverse reinforcement learning model.
10. The method according to claim 1, characterized in that, Determining the outcome of a game decision includes: Determine the interaction behavior parameters corresponding to the target vehicle based on driving behavior characteristics; Determine safety constraint parameters based on collision risk information; The game decision result corresponding to this vehicle is determined based on the interaction behavior parameters, safety constraint parameters, and the interaction relationship between the vehicle and the target vehicle.
11. The method according to claim 1, characterized in that, Acquiring multi-source perception data of the vehicle's surrounding environment also includes: Perform anomaly detection on the received V2X data; When V2X data suffers from at least one of communication delay, data loss, or location drift, the V2X data is compensated or corrected. Obtain target vehicle status information based on corrected V2X data.
12. An autonomous driving game-theoretic trajectory planning system, characterized in that, include: The environmental perception module is used to acquire multi-source perception data of the environment around the vehicle. The multi-source perception data includes environmental information acquired by the on-board perception system and target vehicle status information acquired by V2X communication. The state fusion module is used to determine the fused state information of the target vehicle based on multi-source perception data, and also to determine the collision risk information between the vehicle and the target vehicle based on the fused state information. The intent prediction module is used to determine the future behavioral intent of the target vehicle based on the fused state information; The driving behavior analysis module is used to determine the driving behavior characteristics of the target vehicle based on future behavioral intentions and the target vehicle's historical driving behavior. The game decision-making module is used to determine the game decision result based on collision risk information, driving behavior characteristics, and the interaction between the vehicle and the target vehicle. The trajectory planning module is used to generate vehicle driving trajectories based on the game decision results.
13. A storage medium, characterized in that, The storage medium stores computer instructions, which, when executed by the computer, are used to perform an autonomous driving game trajectory planning method as described in any one of claims 1-11.
14. An electronic device, characterized in that, Includes at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform an autonomous driving game trajectory planning method as described in any one of claims 1-11.
15. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements an autonomous driving game trajectory planning method as described in any one of claims 1-11.
