Hybrid decision arbitration method and system based on automatic driving

Abstract

The invention relates to a hybrid decision arbitration method based on automatic driving. The method comprises the following steps: step 1, driving a decision to output an end-to-end trajectory set through end-to-end data; 2, outputting a regular trajectory set through regular space-time planning; step 3, constructing the end-to-end track set and the rule type track set to generate an optional track set, obtaining the optional track set, carrying out decision scene extraction, determining a plurality of decision strategies, and carrying out weighted score calculation on each decision strategy according to a scoring formula to obtain a strategy score; and 4, selecting an optimal strategy according to the score, the user conservative degree grade, a preset conservative degree M and a safety threshold, and executing the optimal strategy. According to the technical scheme, the technical problems that the end-to-end decision technology is not explained in decision, poor in long-tail scene adaptability and the like, and the rule method lacks flexibility and generalization ability are solved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a hybrid decision arbitration method and system based on autonomous driving. Background Technology

[0002] With the development of automotive intelligence, autonomous driving technology has become a research hotspot and a core driver of industrial transformation in the transportation sector. Traditional autonomous driving adopts a modular architecture, handling perception, decision-making, and control functions independently. However, this approach suffers from problems such as information loss between modules, high system complexity, and difficulty in handling complex scenarios. End-to-end decision-making technology has emerged, which directly maps raw sensor data to control commands through deep learning, reducing intermediate steps and effectively handling complex environmental information. However, it also has limitations such as uninterpretable decisions and poor adaptability to long-tail scenarios. While rule-based methods can handle known scenarios based on explicit logic, they lack flexibility and generalization ability. Summary of the Invention

[0003] In view of the above problems, the present invention provides a hybrid decision arbitration method and system based on autonomous driving to solve the technical problems of end-to-end decision technology having limitations such as uninterpretable decision-making and poor adaptability to long-tail scenarios, and rule-based methods lacking flexibility and generalization ability.

[0004] This invention provides a hybrid decision arbitration method based on autonomous driving. The method includes: Step 1, outputting an end-to-end trajectory set through end-to-end data-driven decision-making; Step 2, outputting a regular trajectory set through rule-based spatiotemporal planning; Step 3, constructing an optional trajectory set from the end-to-end trajectory set and the regular trajectory set, obtaining the optional trajectory set and refining the decision scenario to determine several decision strategies, and then calculating a weighted score for each decision strategy according to a scoring formula to obtain a strategy score; Step 4, selecting the optimal strategy based on the score, the user's conservatism level, and a preset conservatism M and safety threshold, and executing it.

[0005] Further, step 1 includes: step 11, acquiring fused perception information and vehicle lateral and longitudinal constraint information, wherein the fused perception information includes dynamic obstacle information, static obstacle information, map information, and vehicle state information; step 12, using the fused perception information to obtain dynamic obstacle codes, static obstacle codes, map codes, and vehicle state codes through an encoder; step 13, generating lateral and longitudinal queries based on the vehicle lateral and longitudinal constraint information, and combining them into lateral and longitudinal query constraints; step 14, applying lateral and longitudinal self-attention mechanisms to the dynamic obstacle codes, static obstacle codes, map codes, vehicle state codes, and lateral and longitudinal query constraints respectively, and then applying a scene interaction attention mechanism; step 15, generating an end-to-end trajectory set through the encoder.

[0006] Furthermore, step 2 includes: step 21, obtaining dynamic obstacle information, static obstacle information, map information, and vehicle information in the environment by fusing perception information, and uniformly mapping them to a spatiotemporal semantic map for modeling; step 22, generating a behavior decision tree at the action semantic level by uniformly constructing spatiotemporal semantic information; step 23, constructing multiple decision scenarios through the behavior decision tree, passing them through a vehicle trajectory generator and a surrounding vehicle trajectory generator respectively to obtain the vehicle trajectory and the surrounding vehicle trajectory, and performing a decision scenario game between the vehicle trajectory and the surrounding vehicle trajectory to obtain a passable decision trajectory; step 24, generating a rule-based trajectory set from the passable decision trajectory.

[0007] Furthermore, step 3 includes: step 31, constructing an optional trajectory set from the end-to-end trajectory set and the rule-based trajectory set, obtaining the optional trajectory set and refining the decision scenario to determine several strategies; step 32, according to the scoring formula... all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety This represents the score for the security dimension; step 33, based on the score, matches the strategy with the user's level of conservatism.

[0008] Furthermore, step 4 includes: step 41, selecting a matching strategy X based on the user's conservatism level; step 42, determining whether the strategy score of strategy X meets the requirements. If it does, it becomes the optimal strategy and is executed; if it does not, proceed to step 43; step 43, downgrading the user's conservatism level, and then returning to step 41 to reselect a matching decision strategy X.

[0009] Furthermore, step 42 includes: step 421, determining whether the strategy score of strategy X is greater than the preset conservatism M; if yes, then strategy X is selected as the optimal strategy and executed; otherwise, proceed to step 412; step 422, determining whether the strategy score of strategy X is greater than the preset safety threshold; if yes, then strategy X is selected as the optimal strategy and executed; otherwise, proceed to step 423; step 423, from all strategies with scores greater than the safety threshold, reselect the strategy with the highest score, and proceed to step 43.

[0010] Furthermore, step 423 also includes: if the scores of all strategies are not greater than the safety threshold, then output a fault code.

[0011] Furthermore, the method of execution specifically involves outputting the driving trajectory corresponding to the optimal strategy to the intelligent driving control module, so as to control the vehicle to perform driving operations according to the trajectory.

[0012] This invention also provides a hybrid decision-making arbitration system based on autonomous driving. The system includes: a trajectory set output module, mounted on the vehicle, used to output an end-to-end trajectory set through end-to-end data-driven decision-making, output a regular trajectory set through rule-based spatiotemporal planning, and input all trajectory sets into a hybrid arbitration module; and a hybrid arbitration module, connected to the trajectory output module, used to acquire an optional trajectory set, refine decision scenarios, determine several decision strategies, and then apply a scoring formula. all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety The score represents the security dimension; the policy execution module, connected to the hybrid arbitration module, is used to select the optimal policy and execute it based on the score, the user's conservatism level, and the preset conservatism level M and security threshold.

[0013] Furthermore, the system also includes a control module connected to the strategy execution module, used to obtain the driving trajectory of the optimal strategy determined by the strategy execution module, and control the vehicle to perform driving operations according to the trajectory.

[0014] This invention provides a hybrid decision-making arbitration method and system based on autonomous driving. This technical solution leverages the advantages of an arbitration system that integrates end-to-end data-driven and rule-based methods in autonomous driving to improve the safety, reliability, and scenario adaptability of autonomous driving systems, thereby meeting the stringent requirements for autonomous driving to move from laboratory to large-scale commercial applications. This technical solution not only addresses the safety challenges caused by the uninterpretability of end-to-end decision-making in autonomous driving but also solves the problem of fusing trajectories generated by data-driven decision-making methods with those generated by rule-based decision-making methods. Attached Figure Description

[0015] Figure 1 A schematic diagram of a hybrid decision-making arbitration method based on autonomous driving provided by the present invention; Figure 2 This is an architecture diagram of a deep learning decision model based on the Transformer architecture. Figure 3 This is a flowchart of an end-to-end trajectory set output method provided by the present invention; Figure 4 This is a flowchart of a regular trajectory set output method provided by the present invention; Figure 5 This is a flowchart of a strategy method provided by the present invention; Figure 6 This is a flowchart of a method for determining a strategy and scoring provided by the present invention. Detailed Implementation

[0016] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] Example 1: This invention provides a hybrid decision-making arbitration method and system based on autonomous driving. The system includes a trajectory set output module, a hybrid arbitration module, and a policy execution module. Figure 1 As shown, the method includes the following steps.

[0018] Step 1: Output end-to-end trajectory sets through end-to-end data-driven decision-making; Step 2: Output a set of regular trajectories through rule-based spatiotemporal planning; The trajectory set output module, installed on the vehicle, is used to output an end-to-end trajectory set through end-to-end data-driven decision-making, and output a regular trajectory set through rule-based spatiotemporal planning. The end-to-end trajectory set and the regular trajectory set are used to construct an optional trajectory set. Step 3: Construct an optional trajectory set from the end-to-end trajectory set and the rule-based trajectory set. Obtain the optional trajectory set and refine the decision-making scenario to determine several strategies. Then, apply the scoring formula. all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety Indicates the score for the security dimension; The hybrid arbitration module, connected to the trajectory output module, is used to acquire a set of optional trajectories, refine decision-making scenarios, determine several decision-making strategies, and then apply the scoring formula. all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety This represents the score for the security dimension.

[0019] Step 4: Based on the user's selected level of conservatism, as well as the preset conservatism level M and safety threshold, select the optimal strategy and execute it.

[0020] The strategy execution module, connected to the hybrid arbitration module, is used to select the optimal strategy and execute it based on the score, the user's conservatism level, and the preset conservatism level M and security threshold.

[0021] This invention provides a hybrid decision-making arbitration method and system based on autonomous driving. This technical solution leverages the advantages of an arbitration system that integrates end-to-end data-driven and rule-based methods in autonomous driving to improve the safety, reliability, and scenario adaptability of autonomous driving systems, thereby meeting the stringent requirements for autonomous driving to move from laboratory to large-scale commercial applications. This technical solution not only addresses the safety challenges caused by the uninterpretability of end-to-end decision-making in autonomous driving but also solves the problem of fusing trajectories generated by data-driven decision-making methods with those generated by rule-based decision-making methods.

[0022] Example 2: This invention provides a hybrid decision-making arbitration method based on autonomous driving. The system includes a trajectory set output module, a hybrid arbitration module, and a policy execution module. Figure 1 As shown, the method includes the following steps.

[0023] Step 1: Output end-to-end trajectory sets through end-to-end data-driven decision-making; The end-to-end data-driven decision-making in step 1 uses a deep learning decision model framework based on the Transformer architecture, the main architecture of which is as follows: Figure 2 As shown, the Transformer architecture is a deep learning model for natural language processing tasks, consisting of an encoder and a decoder. The encoder is responsible for converting the input sequence into a set of vector representations, while the decoder is responsible for converting these vector representations into an output sequence. Both components utilize self-attention and multi-head attention mechanisms. The advantages of the Transformer model include strong parallel computing power, the ability to handle long text sequences, and low training and inference costs. It has become an important foundational model in the field of natural language processing and has achieved excellent results in various tasks. Figure 3 As shown, step 1 includes the following steps.

[0024] Step 11: Obtain fused perception information and vehicle lateral and longitudinal constraint information. The fused perception information includes dynamic obstacle information, static obstacle information, map information, and vehicle status information. The model input consists of fused perception information and vehicle lateral and longitudinal constraint information. The fused perception information includes dynamic obstacle information, static obstacle information, map information, and vehicle state information. The vehicle lateral and longitudinal constraint information includes the vehicle's lateral acceleration constraint information and longitudinal acceleration constraint information.

[0025] Step 12: The sensory information is fused through the encoder to obtain dynamic obstacle code, static obstacle code, map code, and vehicle status code; The dynamic obstacle information, static obstacle information, map information, and vehicle status information output by the fusion perception are used to obtain dynamic obstacle coding, static obstacle coding, map coding, and vehicle status coding through the Transformer encoder.

[0026] Step 13: Based on the vehicle's lateral and longitudinal constraint information, generate lateral and longitudinal queries and combine them into lateral and longitudinal query constraints; Step 14: Dynamic obstacle coding, static obstacle coding, map coding, vehicle status coding, and horizontal and vertical query constraints are respectively subjected to horizontal and vertical self-attention mechanisms, and then the scene interaction attention mechanism is applied. Step 15: Generate an end-to-end trajectory set using the encoder.

[0027] The end-to-end trajectory set generated by the Transformer decoder is passed to the hybrid arbitration module.

[0028] Step 2: Output a set of regular trajectories through rule-based spatiotemporal planning; like Figure 4 As shown, step 2 includes the following steps.

[0029] Step 21: By fusing perception information, dynamic obstacle information, static obstacle information, map information, and vehicle information in the environment are obtained and uniformly mapped to a spatiotemporal semantic map for modeling. Step 22: Generate a behavior decision tree at the action semantic level by uniformly constructing spatiotemporal semantic information; Step 23: Construct multiple decision scenarios through a behavior decision tree, and obtain the vehicle trajectory and the surrounding vehicle trajectory by passing through the vehicle trajectory generator and the surrounding vehicle trajectory generator respectively. Then, conduct a decision scenario game between the vehicle trajectory and the surrounding vehicle trajectory to obtain a passable decision trajectory. Step 24: Generate a set of regular trajectories from the trajectories that can be accessed.

[0030] Step 3: Construct an optional trajectory set from the end-to-end trajectory set and the rule-based trajectory set. Obtain the optional trajectory set and refine the decision-making scenario to determine several strategies. Then, apply the scoring formula. all (t)=(δ regulation *score regulation+δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety Indicates the score for the security dimension; like Figure 5 As shown, step 3 includes the following steps.

[0031] Step 31: Construct an optional trajectory set from the end-to-end trajectory set and the rule-based trajectory set, obtain the optional trajectory set and refine the decision scenario to determine several strategies; The system acquires a set of selectable trajectories and extracts decision-making scenarios to determine decision-making strategies, including strategy 1, strategy 2, and strategy n, totaling N available strategies. For example, the current set of selectable trajectories contains several trajectories, which can be decomposed into N available strategies based on the actual scenario, such as slowing down to follow, accelerating to overtake and change lanes to the left, slowing down to yield and change lanes to the left, accelerating to overtake and change lanes to the right, slowing down to yield and change lanes to the right, large vehicles avoiding and detouring, and slowing down to wait for pedestrians.

[0032] Step 32, according to the scoring formula (score) all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety Indicates the score for the security dimension; Among them, the regulatory dimension score regulation Fraction calculation method: , α type The current lane line type coefficient for trajectory interaction includes dashed lines, single solid lines, double yellow lines, etc.; ω typeT Tl represents the time coefficient for crossing the lane markings of the current lane type. traj Indicates the expected time for the current trajectory to cross the line, Tl type Indicates the penalty time for crossing the lane line for the current lane type; ω typeD D represents the lane marking distance coefficient for the current lane marking type. invade β represents the expected distance of the current trajectory from the line, where W represents the vehicle width; type This represents the score coefficient (legal penalty intensity) for different lane line types under different user modes (conservative, normal, aggressive, efficiency-first, etc.). For example, in the efficiency-first mode, the β of a single solid line... type With a coefficient of 0, in the conservative mode, β type The coefficient is 1. Efficiency score. efficiency Fraction calculation method: , Where ω1, ω2, and ω3 are the weights of different efficiency parameters; T traj T represents the expected travel time of the current trajectory. min T represents the minimum time among all available trajectories. max L represents the longest time among all available trajectories. traj L represents the length of the current trajectory. shortest V represents the shortest trajectory length among all available trajectories; avg V represents the average velocity of the current trajectory. limit This indicates the current speed limit for the road. Comfort score. comfort Fraction calculation method: , Where λ1, λ2, λ3, λ4, and λ5 are the weights of different comfort parameters; J x It represents lateral jerk, which is the rate of change of lateral acceleration. J represents the maximum lateral jerk in this trajectory. x,max The comfort threshold representing lateral jerk is typically set to 1.0 m / s². 3 J y It represents longitudinal jerk, which is the rate of change of longitudinal acceleration. J represents the maximum longitudinal jerk in this trajectory. y,maxThe comfort threshold representing longitudinal jerk is typically set to 1.5 m / s². 3 A x Indicates lateral acceleration. A represents the maximum lateral acceleration in this trajectory. x,max The comfort threshold representing lateral acceleration is typically set to 0.3g; A y Represents longitudinal acceleration. A represents the maximum longitudinal acceleration in this trajectory. y,max The comfort threshold representing longitudinal acceleration can generally be set to 0.3g; Represents the rate of change of curvature. This represents the maximum rate of change of curvature in the trajectory. The comfort threshold representing the rate of change of curvature can generally be set to 0.02 rad / m / s. (4) Safety dimension score safety The method for calculating scores: score safety =collision*drivable*direction, where collision represents the result of collision detection between the current trajectory and dynamic and static obstacles, which is 1 if there is no collision and 0 otherwise; drivable represents the result of topology judgment of the current trajectory within the drivable area, which is 1 if within the drivable area and 0 otherwise; direction represents the judgment of whether the current trajectory deviates from the current navigation route, which is 1 if the deviation angle is less than 90 degrees and 0 otherwise.

[0033] Step 33: Match the strategy to the user's level of conservatism based on the rating.

[0034] Because the strategy score includes a conservatism attribute, each strategy has a corresponding quantitative score related to its conservatism.

[0035] Step 4: Based on the rating, the user's conservatism level, and the preset conservatism level M and safety threshold, select the optimal strategy and execute it.

[0036] This invention provides a hybrid decision-making arbitration method and system based on autonomous driving. This technical solution leverages the advantages of an arbitration system that integrates end-to-end data-driven and rule-based methods in autonomous driving to improve the safety, reliability, and scenario adaptability of autonomous driving systems, thereby meeting the stringent requirements for autonomous driving to move from laboratory to large-scale commercial applications. This technical solution not only addresses the safety challenges caused by the uninterpretability of end-to-end decision-making in autonomous driving but also solves the problem of fusing trajectories generated by data-driven decision-making methods with those generated by rule-based decision-making methods.

[0037] Example 3: This invention provides a hybrid decision-making arbitration method based on autonomous driving. The system includes a trajectory set output module, a hybrid arbitration module, a policy execution module, and a control module. Figure 1 As shown, the method includes the following steps.

[0038] Step 1: Output the end-to-end trajectory set through end-to-end data-driven decision-making and input it into the multi-level conservatism arbitration model; Step 2: Output a set of regular trajectories through regular spatiotemporal planning and input them into a multi-level conservatism arbitration model; Step 3: The multi-level conservatism arbitration model extracts decision-making scenarios based on the trajectory set, determines several conservative strategies and their scores, and provides a basis for selecting the optimal strategy later. Step 4: Select the optimal strategy based on the user's conservatism preference, conservatism level M, and safety threshold, and execute it.

[0039] like Figure 6 As shown, step 4 includes: Step 41: Select the matching decision strategy X based on the level of conservatism. The user's level of conservatism mentioned here refers to the user's own choice of conservatism level. For example, a user can choose between conservative mode, normal mode, and efficiency mode. Each mode will have a corresponding level of conservatism, with conservative being the highest level and efficiency being the lowest.

[0040] Step 42: Determine whether the score of strategy X meets the strategy requirements. If it does, it becomes the optimal strategy and is executed. If it does not, proceed to step 43. like Figure 6 As shown, step 42 includes: Step 421: Determine whether the score of strategy X is greater than the preset conservatism M. If yes, then strategy X is the optimal strategy and is executed. If not, proceed to step 422. The level of conservatism M is determined by the intelligent driving mode selected by the user. For example, the level of conservatism M is different for Comfort mode, Efficiency mode, and Aggressive mode.

[0041] Step 422: Determine whether the score of strategy X is greater than the safety threshold. If yes, then strategy X is the optimal strategy and is executed. If not, proceed to step 423. The safety threshold is the lowest tolerable value for trajectory scoring.

[0042] Step 423: From all strategies with scores greater than the safety threshold, select the strategy with the highest score and proceed to step 43. If none of the scores are greater than the safety threshold, output a fault code.

[0043] The conservatism level M and safety threshold here are empirical values ​​based on operational experience, which can be trained using takeover data from the operational process. For example, in non-efficiency mode, the regulatory weight for pressing the solid line becomes a 0 / 1 judgment; a trajectory pressing the solid line receives a score of 0. In efficiency mode, the regulatory weight for pressing the solid line becomes a percentage, such as a penalty of 0.8 points, but it doesn't directly assign a score of 0 to the trajectory. The execution method is as follows: the driving trajectory corresponding to the optimal strategy is output to the intelligent driving control module to control the vehicle to drive according to that trajectory. If the scores of all decision strategies do not meet the safety requirements, it indicates that no safe decision exists, and a fault code is output to the intelligent driving system.

[0044] Step 43: Downgrade the user's conservatism level, then proceed to step 41: Reselect the matching decision strategy X.

[0045] This invention provides a hybrid decision-making arbitration method and system based on autonomous driving. This technical solution leverages the advantages of an arbitration system that integrates end-to-end data-driven and rule-based methods in autonomous driving to improve the safety, reliability, and scenario adaptability of autonomous driving systems, thereby meeting the stringent requirements for autonomous driving to move from laboratory to large-scale commercial applications. This technical solution not only addresses the safety challenges caused by the uninterpretability of end-to-end decision-making in autonomous driving but also solves the problem of fusing trajectories generated by data-driven decision-making methods with those generated by rule-based decision-making methods.

[0046] In summary, this invention provides a hybrid decision-making arbitration method and system based on autonomous driving. This technical solution integrates end-to-end data-driven decision-making and rule-based spatiotemporal planning methods through a multi-level conservative arbitration model, achieving complementary advantages and constructing an optimal decision trajectory. Based on the Transformer architecture, a deep learning decision-making model framework realizes closed-loop optimization of perception-decision-control, significantly improving the intelligence ceiling of the autonomous driving system. The proposed rule-based spatiotemporal planning method enhances the interaction flexibility and response speed in dynamic scenarios.

[0047] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A hybrid decision-making arbitration method based on autonomous driving, characterized in that, The method includes: Step 1: Output end-to-end trajectory sets through end-to-end data-driven decision-making; Step 2: Output a set of regular trajectories through rule-based spatiotemporal planning; Step 3: Construct an optional trajectory set from the end-to-end trajectory set and the rule-based trajectory set. Obtain the optional trajectory set and refine the decision-making scenario to determine several strategies. Then, apply the scoring formula. all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each strategy, a weighted score is calculated to obtain the strategy's rating, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety Indicates the score for the security dimension; Step 4: Based on the rating, the user's conservatism level, and the preset conservatism level M and safety threshold, select the optimal strategy and execute it.

2. The hybrid decision-making arbitration method based on autonomous driving according to claim 1, characterized in that, Step 1 includes: Step 11: Obtain fused perception information and vehicle lateral and longitudinal constraint information. The fused perception information includes dynamic obstacle information, static obstacle information, map information, and vehicle status information. Step 12: The sensory information is fused through the encoder to obtain dynamic obstacle code, static obstacle code, map code, and vehicle status code; Step 13: Based on the vehicle's lateral and longitudinal constraint information, generate lateral and longitudinal queries and combine them into lateral and longitudinal query constraints; Step 14: Dynamic obstacle coding, static obstacle coding, map coding, vehicle status coding, and horizontal and vertical query constraints are respectively subjected to horizontal and vertical self-attention mechanisms, and then the scene interaction attention mechanism is applied. Step 15: Generate an end-to-end trajectory set using the encoder.

3. The hybrid decision-making arbitration method based on autonomous driving according to claim 1, characterized in that, Step 2 includes: Step 21: By fusing perception information, dynamic obstacle information, static obstacle information, map information, and vehicle information in the environment are obtained and uniformly mapped to a spatiotemporal semantic map for modeling. Step 22: Generate a behavior decision tree at the action semantic level by uniformly constructing spatiotemporal semantic information; Step 23: Construct multiple decision scenarios through a behavior decision tree, and obtain the vehicle trajectory and the surrounding vehicle trajectory by passing through the vehicle trajectory generator and the surrounding vehicle trajectory generator respectively. Then, conduct a decision scenario game between the vehicle trajectory and the surrounding vehicle trajectory to obtain a passable decision trajectory. Step 24: Generate a set of regular trajectories from the trajectories that can be accessed.

4. The hybrid decision-making arbitration method based on autonomous driving according to claim 1, characterized in that, Step 3 includes: Step 31: Construct an optional trajectory set from the end-to-end trajectory set and the rule-based trajectory set, obtain the optional trajectory set and refine the decision scenario to determine several strategies; Step 32, according to the scoring formula (score) all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each decision strategy, a weighted score is calculated to obtain the strategy's score, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety Indicates the score for the security dimension; Step 33: Match the strategy to the user's level of conservatism based on the rating.

5. The hybrid decision-making arbitration method based on autonomous driving according to claim 4, characterized in that, Step 4 includes: Step 41: Select the matching strategy X based on the user's level of conservatism; Step 42: Determine whether the score of strategy X meets the requirements. If it does, it becomes the optimal strategy and is executed. If it does not, proceed to step 43. Step 43: Downgrade the user's conservatism level, then proceed to step 41: Reselect the matching strategy X.

6. The hybrid decision-making arbitration method based on autonomous driving according to claim 5, characterized in that, Step 42 includes: Step 421: Determine whether the strategy score of strategy X is greater than the preset conservatism M. If yes, then strategy X is the optimal strategy and is executed. If not, proceed to step 412. Step 422: Determine whether the strategy score of strategy X is greater than the preset safety threshold. If yes, then strategy X is the optimal strategy and is executed. If not, proceed to step 423. Step 423: From all strategies with scores greater than the safety threshold, select the strategy with the highest score and proceed to step 43.

7. The hybrid decision-making arbitration method based on autonomous driving according to claim 6, characterized in that, Step 423 further includes: if the scores of all strategies are not greater than the safety threshold, then output a fault code.

8. A hybrid decision-making arbitration method based on autonomous driving according to claim 1, 5, or 6, characterized in that, The execution of the above measures specifically involves outputting the driving trajectory corresponding to the optimal strategy to the intelligent driving control module, so as to control the vehicle to drive according to the trajectory.

9. A system for implementing the hybrid decision-making arbitration method based on autonomous driving as described in any one of claims 1-8, characterized in that, The system includes: The trajectory set output module, installed on the vehicle, is used to output an end-to-end trajectory set through end-to-end data-driven decision-making, and output a regular trajectory set through rule-based spatiotemporal planning. The end-to-end trajectory set and the regular trajectory set are used to construct an optional trajectory set. The hybrid arbitration module, connected to the trajectory output module, is used to acquire a set of optional trajectories, refine decision-making scenarios, determine several strategies, and then apply the scoring formula. all (t)=(δ regulation *score regulation +δ efficiency *score efficiency +δ comfort *score comfort )*δ safety *score safety For each strategy, a weighted score is calculated to obtain the strategy's rating, where δ regulation The score represents the weight of the regulatory dimension. regulation The score representing the regulatory dimension, δ efficiency The score represents the weight of the efficiency dimension. efficiency δ represents the score for the efficiency dimension. comfort The score represents the weight of the comfort dimension. comfort The score represents the comfort dimension. safety Indicates the score for the security dimension; The strategy execution module, connected to the hybrid arbitration module, is used to select the optimal strategy and execute it based on the score, the user's conservatism level, and the preset conservatism level M and security threshold.

10. The hybrid decision-making arbitration system based on autonomous driving according to claim 9, characterized in that, The system also includes: The control module, connected to the strategy execution module, is used to obtain the driving trajectory of the optimal strategy determined by the strategy execution module and control the vehicle to drive according to the trajectory.

Images