Automatic driving lane changing method and system based on safety verification

Abstract

The invention discloses an automatic driving lane changing method and system based on safety verification, and the method comprises the steps: S1, generating a lane changing scheme of a vehicle under the current driving working condition based on the obtained real-time information of the vehicle driving working condition and a trained lane changing planning model; s2, making a safe lane changing rule based on a rule engine algorithm, and performing safety verification on the lane changing scheme under the current driving working condition; s3, judging whether the lane changing scheme under the current driving working condition meets a safe lane changing rule or not, if not, entering the step S1, and if yes, entering the step S4; and S4, controlling the vehicle to change the lane according to the lane changing scheme passing the verification. The safe lane changing rule is formulated based on the rule engine algorithm, the lane changing scheme generated by the lane changing planning model is subjected to safety verification by using the safe lane changing rule, the performability and reliability of the automatic driving lane changing scheme are improved, and thus the safety of automatic driving lane changing and the driving experience of a user can be enhanced.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and particularly to an autonomous driving lane-changing method and system based on safety verification. Background Art

[0002] With the development of autonomous driving technology, lane changing, as a key decision-making behavior, directly affects the driving efficiency and safety performance of vehicles. Traditional autonomous driving lane-changing systems are mostly based on rule-driven or end-to-end deep learning models, which have problems such as poor flexibility, weak adaptability to complex traffic environments, and poor interpretability. Especially in the face of dynamic traffic scenarios, emergencies, or high-density traffic flows, existing systems are prone to misjudgment or delay in judging lane-changing opportunities, selecting target lanes, and planning trajectories, affecting the driving experience and even causing potential risks.

[0003] In recent years, large language models (LLMs) have shown strong generalization and reasoning abilities in the fields of natural language understanding, multimodal reasoning, etc., and have gradually been explored for use in advanced decision-making tasks in autonomous driving scenarios. However, directly using the lane-changing plans output by large models often fails to meet the strict requirements for real-time performance and reliability in the actual road traffic environment, making the lane-changing plans generated by some models theoretically feasible but having insufficient reliability problems in the actual execution process. Summary of the Invention

[0004] This application provides an autonomous driving lane-changing method and system based on safety verification, which can solve the technical problem of poor reliability in current autonomous driving lane-changing technologies.

[0005] To achieve the above object, in a first aspect, this application provides an autonomous driving lane-changing method based on safety verification, and the method includes: S1. Generate a lane-changing plan for the vehicle under the current driving condition based on the real-time information of the vehicle driving condition obtained and the trained lane-changing planning model.

[0006] S2. Formulate safety lane-changing rules based on the rule engine algorithm and perform a safety verification on the lane-changing plan under the current driving condition.

[0007] S3. Determine whether the lane-changing plan under the current driving condition meets the safety lane-changing rules. If not, go to step S1; if so, go to step S4.

[0008] S4. Control the vehicle to change lanes according to the lane-changing plan that has passed the verification.

[0009] Further, in an embodiment, in the step S1, the training method of the lane-changing planning model includes: Obtain the training sample information of the vehicle driving condition and preprocess the training sample information.

[0010] Use the preprocessed training sample information to pre-train the large language model.

[0011] Based on the supervised fine-tuning algorithm and the annotation information of the vehicle driving condition for specific lane-changing tasks, fine-tune the pre-trained large language model.

[0012] Based on the reinforcement learning algorithm with human feedback, update and iterate the parameters of the fine-tuned large language model to obtain a trained lane-changing planning model.

[0013] Further, in one embodiment, in the step S1, the vehicle driving condition includes the destination, vehicle speed, vehicle position, traffic density, road type, and driving behavior.

[0014] Further, in one embodiment, the preprocessing of the training sample information includes: Perform standardization processing on the training sample information.

[0015] Extract key feature variables from the standardized training sample information.

[0016] Combine the context information of the key feature variables and encode the key feature variables with time series or spatial features.

[0017] Use the feature fusion technology to fuse the encoded key features into a high-dimensional semantic vector, perform dimension conversion, and finally generate training sample information with a unified dimension.

[0018] Further, in one embodiment, the standardization processing includes format normalization, outlier removal, and missing value filling.

[0019] Further, in one embodiment, in the step S1, the lane-changing plan includes the target lane-changing lane and the lane-changing trajectory. When checking the safety of the lane-changing plan, when both the target lane-changing lane and the lane-changing trajectory meet the safe lane-changing rules, the lane-changing plan meets the safe lane-changing rules and the check passes.

[0020] Further, in one embodiment, the judgment criterion for the target lane-changing lane to meet the safe lane-changing rules is: On the premise that the user has set waypoints, the target lane-changing lane is the same as the lane where the waypoints are located.

[0021] On the premise that the user has not set waypoints and the difference between the traffic efficiency of the current driving lane and that of the adjacent lane is greater than or equal to the preset threshold, if the traffic efficiency of the current driving lane is lower than that of the adjacent lane, the target lane-changing lane is the adjacent lane with the highest traffic efficiency.

[0022] On the premise that the user has not set waypoints and the difference between the traffic efficiency of the current driving lane and that of the adjacent lane is less than a preset threshold, the target lane for lane change is the middle lane.

[0023] Further, in one embodiment, the judgment criteria for the lane change trajectory to meet the safe lane change rule are as follows: The distance from the starting point of the lane change trajectory to the preset necessary waypoint is greater than or equal to the distance from the preset latest lane change point to the preset necessary waypoint, and the curvature of the lane change trajectory is less than or equal to the preset curvature threshold.

[0024] Further, in one embodiment, the preset curvature threshold is set based on the riding comfort of the user when the vehicle turns.

[0025] In a second aspect, based on the above-mentioned safe-check-based autonomous driving lane change method, the present application provides an autonomous driving lane change system for the safe-check-based autonomous driving lane change method. The system includes: A scenario generation module, which is used to generate a lane change scenario for the vehicle under the current driving scenario based on the real-time information of the vehicle driving scenario obtained and the trained lane change planning model.

[0026] A verification module, which is used to formulate safe lane change rules based on the rule engine algorithm and perform safety verification on the lane change scenario. If the lane change scenario does not meet the safe lane change rules, a new lane change scenario will be generated until the lane change scenario meets the safe lane change rules.

[0027] A lane change module, which is used to control the vehicle to change lanes according to the verified lane change scenario.

[0028] The beneficial effects brought by the technical solutions provided in the embodiments of the present application include: Based on the real-time information of the vehicle driving scenario obtained and the trained lane change planning model, the present application generates a lane change scenario for the vehicle under the current driving scenario; formulates safe lane change rules based on the rule engine algorithm and performs safety verification on the lane change scenario. If the lane change scenario does not meet the safe lane change rules, a new lane change scenario will be generated until the lane change scenario meets the safe lane change rules; controls the vehicle to change lanes according to the verified lane change scenario. By formulating safe lane change rules based on the rule engine algorithm and using the safe lane change rules to perform safety verification on the lane change scenario generated by the lane change planning model, the executability and reliability of the autonomous driving lane change scenario are improved, thereby enhancing the safety of autonomous driving lane change and the driving experience of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flowchart of the safe-check-based autonomous driving lane change method according to the embodiment of the present application.

[0030] Figure 2 It is a flowchart of the training method of the lane-changing planning model in step S1 of the embodiment of the present application.

[0031] Figure 3 It is a flowchart of preprocessing the training sample information in step A11 of the embodiment of the present application.

[0032] Figure 4 It is a block diagram of an autonomous driving lane-changing system based on safety verification according to an embodiment of the present application.

[0033] Figure 5 It is a block diagram of an autonomous driving lane-changing device based on safety verification according to an embodiment of the present application. Detailed implementation manners

[0034] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0035] First, some technical terms in the present application are explained to facilitate the understanding of the present application by those skilled in the art.

[0036] Instruction fine-tuning: It is a technique for fine-tuning a large model on a labeled dataset with instruction prompts and corresponding outputs.

[0037] Parameter fine-tuning: It refers to the adjustment of a pre-trained model in machine learning to adapt to a specific task or dataset, including methods such as LoRA (Low-Rank Adaptation of Large Language Models) and Adapter (Adapter-based Fine-tuning).

[0038] Transformer encoder-decoder architecture: It is a deep learning model architecture used for natural language processing and sequence-to-sequence tasks.

[0039] To make the purpose, technical solutions and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail in conjunction with the accompanying drawings.

[0040] In a first aspect, an embodiment of the present application provides an autonomous driving lane-changing method based on safety verification.

[0041] In one embodiment, referring to Figure 1 as shown, the above-mentioned autonomous driving lane-changing method includes: S1. Generate a lane-changing plan for the vehicle under the current driving condition based on the real-time information of the vehicle driving condition obtained and the trained lane-changing planning model.

[0042] S2. Formulate safety lane-changing rules based on the rule engine algorithm and perform safety verification on the lane-changing plan under the current driving condition.

[0043] S3. Determine whether the lane-changing plan under the current driving condition meets the safety lane-changing rules. If not, go to step S1; if so, go to step S4.

[0044] S4. Control the vehicle to change lanes according to the verified lane-changing plan.

[0045] In this embodiment, the real-time information of the above vehicle driving condition is the information collected in real time during the actual driving process of the vehicle, which may include user information, vehicle information, and road information. Specifically: User information may include destination, driving preference, driving style, tendency towards overtaking or conservative driving, driving behavior, and subjective lane-changing road selection priority, etc.; vehicle information may include vehicle speed, vehicle position, vehicle acceleration, vehicle size, wheelbase, steering angle, power system response ability, braking performance, current lane keeping state, and relative distance between the vehicle and surrounding vehicles, etc.; road information may include traffic density, road type, lane structure (such as lane width, type, whether lane-changing is allowed, etc.), adjacent lane state (such as whether it is idle, adjacent vehicle type, speed, etc.), forward and backward traffic flow speeds, speed limit information, intersection or ramp information, lane line type (solid line or dashed line), influence of weather and road conditions on road feasibility, etc. This application does not limit the real-time information of the vehicle driving condition.

[0046] This application first generates a lane-changing plan for the vehicle under the current driving condition based on the real-time information of the vehicle driving condition obtained and the trained lane-changing planning model. Then, use the safety lane-changing rules formulated based on the rule engine algorithm to perform safety verification on the generated lane-changing plan. If the verification fails, that is, the lane-changing plan does not meet the safety lane-changing rules, then regenerate the lane-changing plan and enter the safety verification process again. If the verification passes, control the vehicle to perform a lane-changing operation according to this lane-changing plan. By combining the reasoning ability of the large language model and the safety verification mechanism driven by the rule engine, the method of this application not only has the intelligent automatic driving lane-changing decision-making ability, but also can systematically ensure the reliability and safety of the execution of the lane-changing plan, realizing the efficient generation, reasonable evaluation of the lane-changing plan, and the safety control of automatic driving lane-changing.

[0047] Further, in one embodiment, as shown in Figure 2 The specific steps of the training method of the lane-changing planning model in the above step S1 are as follows: A11. Obtain the training sample information of the vehicle driving conditions and preprocess the training sample information.

[0048] A12. Use the preprocessed training sample information to pre-train the large language model.

[0049] A13. Fine-tune the pre-trained large language model based on the supervised fine-tuning algorithm and the annotation information of the vehicle driving conditions for specific lane-changing tasks.

[0050] In this embodiment, the fine-tuning method adopts a combination of instruction fine-tuning and parameter fine-tuning.

[0051] A14. Based on the reinforcement learning algorithm with human feedback, update and iterate the parameters of the fine-tuned large language model to obtain a trained lane-changing planning model.

[0052] The training sample information of the vehicle driving conditions in this embodiment is derived from a large-scale collected multi-modal data set, which may include the above user information, vehicle information, and road information, but is not necessarily real-time data and can cover a variety of driving scenarios. The above training sample information can be collected in real-time synchronization through sensors, map systems, and vehicle-mounted controllers, and preliminary screening and timestamp alignment are performed.

[0053] In this embodiment, first, pre-train the large language model with a large amount of training sample information of vehicle driving conditions, so that the large language model has the basic ability of traffic language understanding and multi-source information fusion. Secondly, based on the vehicle driving condition information and the supervised fine-tuning algorithm under specific lane-changing tasks with annotations, adopt a combination of instruction fine-tuning and parameter fine-tuning to fine-tune the pre-trained large language model, so that the model can perform conditional understanding and reasoning on the fused input features and output a structured lane-changing plan, making the model have good robustness and generalization ability. Finally, apply the RLHF (Reinforcement Learning from Human Feedback) algorithm to update and iterate the fine-tuned large language model, so that the model behavior is consistent with human preferences.

[0054] Further, in one embodiment, as shown in Figure 3 In step A11 above, the steps of preprocessing the training sample information are as follows: A111. Standardize the training sample information, including format normalization, outlier removal, and missing value filling.

[0055] A112. Extract key feature variables from the standardized training sample information. The key feature variables may include driving style, vehicle dynamics parameters, and traffic flow information.

[0056] A113. Encode the key feature variables with time series or spatial features in combination with the context information of the key feature variables. Among them, the context information of the key feature variables may include timestamps, vehicle position coordinates, and the vehicle's historical driving trajectory.

[0057] A114. Use feature fusion technology to fuse the encoded key features into a high-dimensional semantic vector, and perform dimensionality conversion to finally generate training sample information with a unified dimension.

[0058] Among them, the above feature fusion technology may be feature splicing, attention mechanism, or fusion network. The dimensionality conversion of the high-dimensional semantic vector can be performed through an embedding layer to map the multi-dimensional semantic vector to a vector with a unified dimension.

[0059] In this embodiment, by performing normalization processing on the training sample information, the data quality in the training sample information can be ensured to be consistent. By performing unified dimensionality conversion on the high-dimensional semantic vector, the requirements for the input format and semantic expression in the inference stage of the large language model to be consistent can be met.

[0060] Further, in one embodiment, in the above step A12, the large language model is pre-trained using the preprocessed training sample information, which can be implemented through a Transformer encoder-decoder architecture, and this architecture includes an encoder and a decoder.

[0061] First, model the input and output of the Transformer encoder-decoder architecture. The input sequence is represented by the following expression (1): (1), Among them, X represents the input sequence, represents t the user information at time represents t the vehicle information at time represents t the road information at time.

[0062] The encoder is used to extract high-dimensional features from the input training sample information sequence. It consists of multiple layers of self-attention and fully connected neural networks, with a total of 16 layers. Its expression (2) is: (2), Among them, H represents the output of the encoder, E(X) represents the encoder function.

[0063] The decoder is used to generate a lane-changing plan. It also uses multiple layers of self-attention and fully connected neural networks, with a total of 16 layers. Its expression (3) is: (3), Among them, represents the output of the decoder at t+1 moment, represents the decoder function, which receives the output of the encoder H and additional specified input information O .

[0064] The goal of the Transformer encoder-decoder architecture is to learn the internal logical relationship from X to .

[0065] Secondly, in order to make the generated lane-changing plan closer to the real lane change, a loss function within a certain time period is defined, and its expression (4) is: (4), Among them, L represents the output of the loss function, T represents the total time within the time period, represents the predicted control command at t moment, represents the real measured control command at t moment, λ and μ respectively represent the weight coefficients of different losses, CL represents the collision loss, LKL represents the lane-keeping loss.

[0066] CL The expression (5) of (5), Among them, N represents the number of surrounding vehicles, represents the position of the vehicle itself at t moment, represents the i th surrounding vehicle at t moment, d safe represents the pre-given safety distance.

[0067] LKL The expression (6) of (6), Among them, y ego (t) represents the lateral position of the autonomous vehicle at t moment, y lane(t) Represents t The lateral position of the lane centerline at a moment.

[0068] In this embodiment, by CL Punishing behaviors that may lead to collisions, and by LKL Encouraging vehicles to stay within the lane and maintain smooth driving behaviors can improve the prediction accuracy and generalization ability of the model.

[0069] Furthermore, in one embodiment, the supervised fine-tuning (SFT) in the above step A13 is achieved by minimizing the supervised loss function, and the expression (7) of the supervised loss function is: (7), Wherein, L SFT Represents the output of the supervised fine-tuning loss function, E Represents the expected value operator, (M,Y) Represents the input data and the corresponding labels, D fine-tune Represents the fine-tuning dataset, M Represents the input features, Y <t Represents t The label sequence before the moment, θ SFT Represents the model parameters after fine-tuning, P(Y t |X,Y <t ;θ SFT ) Represents given X And Y <t , under the model parameters θ SFT After fine-tuning, predicting t The label at the moment Y t Probability.

[0070] Furthermore, in one embodiment, the expression (8) of RLHF in the above step A14 is: (8), Wherein, L RLHF Represents the output of RLHF, W Represents the input features, D RL Represents the RLHF dataset, V Represents the lane-changing trajectory, aIndicates a lane change decision, Indicates the model parameters updated by RLHF θ RLHF Parameterized policy, Indicates the optimized reward function.

[0071] Furthermore, in one embodiment, in the above step S2, based on the rule engine algorithm, a safe lane change rule is formulated to perform a safety check on the lane change plan under the current driving condition. Among them, the lane change plan includes the target lane to change and the lane change trajectory. When performing the safety check on the lane change plan, when both the target lane to change and the lane change trajectory meet the safe lane change rule, the lane change plan meets the safe lane change rule and the check passes.

[0072] Furthermore, in one embodiment, the judgment criterion for the target lane to change to meet the safe lane change rule is: On the premise that the user has set waypoints, the target lane to change is the same as the lane where the waypoints are located.

[0073] In this embodiment, the waypoints set by the user can be represented by the following expression (9), and the waypoint lane change decision can be represented by the following expression (10): (9), Wherein, Represents the waypoint sequence.

[0074] (10), Wherein, Represents the waypoint lane change decision function, which is used to judge whether a lane change is required at the waypoint. 1 means a lane change is required, and 0 means a lane change is not required.

[0075] On the premise that the user has not set waypoints and the difference between the traffic efficiency of the current driving lane and that of the adjacent lane is greater than or equal to the preset threshold, if the traffic efficiency of the current driving lane is lower than that of the adjacent lane, the target lane to change is the adjacent lane with the highest traffic efficiency.

[0076] In this embodiment, the traffic efficiency is calculated by formula (11), and the traffic efficiency lane change decision is represented by the following expression (12): (11), Wherein, Represents the lane j At the t Moment traffic efficiency, Represents t Moment lane j The average speed of the vehicles on, Represents t Moment lane j The vehicle density on.

[0077] (12), wherein, represents the traffic efficiency lane change decision, 1 means it is necessary to change to the adjacent lane with the highest traffic efficiency, and 0 means it is not necessary to change to the adjacent lane with the highest traffic efficiency.

[0078] On the premise that the user does not set waypoints and the difference between the traffic efficiency of the current driving lane and that of the adjacent lane is less than the preset threshold, the target lane for lane change is the middle lane.

[0079] In this embodiment, the middle lane change decision is represented by the following expression (13): (13), wherein, represents the middle lane change decision, 1 means it is necessary to change to the middle lane, and 0 means it is not necessary to change to the middle lane.

[0080] Furthermore, in one embodiment, the judgment criterion for the above lane change trajectory to satisfy the safe lane change rule is: The distance from the starting point of the lane change trajectory to the preset necessary waypoint is greater than or equal to the distance from the preset latest lane change point to the preset necessary waypoint, and the curvature of the lane change trajectory is less than or equal to the preset curvature threshold.

[0081] In this embodiment, the latest lane change point is defined as the last position on the current lane where a safe lane change can be made. The calculation formula (14) for the distance from the latest lane change point to the next nearest waypoint is: (14) wherein, represents the distance from the latest lane change point to the next nearest waypoint, represents the distance from the actual driving lane change point of the vehicle to the next waypoint, represents the vehicle speed, represents the time required for the vehicle to complete a safe lane change during actual driving.

[0082] In this embodiment, the calculation formula (15) for the curvature is: (15), wherein, represents the curvature at the s point on the lane change trajectory, represents the first derivative of the position vector along the lane change trajectory, [[ID=�6]] represents the second derivative of the position vector along the lane change trajectory.

[0083] The lane change trajectory needs to satisfy the following formula (16): , (16), wherein, represents a preset curvature threshold, which is set based on the riding comfort of the user when the vehicle turns, represents the starting point on the lane change trajectory, represents the ending point on the lane change trajectory.

[0084] In a second aspect, based on the embodiments of the above-mentioned safety-check-based automatic driving lane change method, embodiments of a safety-check-based automatic driving lane change system are provided. Refer to Figure 4 As shown, the above system includes a plan generation module, a verification module, and a lane change module. Specifically: The plan generation module is used to generate a lane change plan for the vehicle under the current driving condition based on the real-time information of the vehicle driving condition obtained and the trained lane change planning model.

[0085] The verification module is used to formulate safety lane change rules based on the rule engine algorithm, perform safety verification on the lane change plan. If the lane change plan does not meet the safety lane change rules, a new lane change plan is regenerated until the lane change plan meets the safety lane change rules.

[0086] The lane change module is used to control the vehicle to change lanes according to the verified lane change plan.

[0087] Furthermore, in one embodiment, refer to Figure 5 As shown, the present application also provides a safety-check-based automatic driving lane change device. The structure of the device includes: a perception and acquisition module 101, a data processing and feature fusion module 102, an AI large model training and inference module 103, a lane selection verification module 104, a lane change trajectory verification module 105, and a control and execution module 106, which are used to implement the aforementioned safety-check-based automatic driving lane change method. The functions of each module are as follows: The perception and acquisition module 101 is used to obtain user information, vehicle information, and environmental information (such as vehicle speed, steering angle, road structure, traffic density, etc.) in real time as the original input for subsequent processing.

[0088] The data processing and feature fusion module 102 is used to standardize the format of the information collected by the perception and acquisition module 101, fill in missing values, eliminate anomalies, extract multi-dimensional features, and transform them into a unified semantic vector through joint embedding to construct a structured semantic vector.

[0089] The AI large model training and inference module 103 is used to perform inference based on the semantic vector of the data processing and feature fusion module 102, use a pre-trained and fine-tuned large language model to generate a preliminary target lane change lane and lane change trajectory, and output the target lane change lane and the sequence of lane change path points.

[0090] The lane selection verification module 104 is used to perform rule verification on the target lane change lane generated by the AI large model training and inference module 103, including path legality, traffic efficiency, and priority judgment, to ensure that the selection of the target lane change lane is reasonable and safe.

[0091] The lane change trajectory verification module 105 is used to conduct a dynamic executability review of the lane change trajectory generated by the AI large model training and inference module 103, including the judgment of the latest lane change point, the verification of curvature continuity, and the matching verification with vehicle dynamics, to ensure that the trajectory is controllable and stable.

[0092] The control execution module 106 is used to, after all verifications of the lane selection verification module 104 and the lane change trajectory verification module 105 are passed, receive the final lane change instruction and transmit it to the vehicle control system to drive the execution unit to complete the lane change operation, forming a closed-loop control.

[0093] In this embodiment, the above device can be deployed at the vehicle end or the edge computing platform, with high-efficiency decision-making capabilities and a rigorous safety mechanism, and is applicable to the autonomous driving system in a complex traffic environment.

[0094] Further, in one embodiment, the timing relationship between the modules of the above autonomous driving lane change device based on safety verification is as follows: The perception and acquisition module 101 acquires multi-source information and transmits it to the data processing and feature fusion module 102; the feature fusion module 102 preprocesses the information acquired by the perception and acquisition module 101 and transmits it to the AI large model training and inference module 103; the AI large model training and inference module 103 completes the training of the lane change planning model, and uses the trained lane change planning model and the real-time information of the vehicle driving conditions obtained to generate a lane change plan, and transmits the lane change plan to the lane selection verification module 104 and the lane change trajectory verification module 105; the lane selection verification module 104 completes the safety inspection of the target lane change lane; the lane change trajectory verification module 105 completes the safety inspection of the lane change trajectory; the target plan passed through the verification of the selection verification module 104 and the lane change trajectory verification module 105 is transmitted to the control execution module 106, and the control execution module 106 drives the execution unit to complete the lane change operation.

[0095] This application collects multi-source data such as user information, vehicle information, and road information to construct a multi-modal feature input, and uses joint embedding technology to generate a unified semantic vector, which is input into a trained lane-changing planning model. Based on the real-time information of the vehicle driving conditions obtained, a preliminary lane-changing intention and trajectory plan are generated to achieve intelligent reasoning for target lane selection and lane-changing trajectory. On this basis, a dual safety verification mechanism is introduced to perform safety verification on the target lane selection and lane-changing trajectory output by the model. By combining the rule engine with physical constraints, it avoids the situation in the current technology that when focusing on geometric reachability in lane-changing trajectories, there is a lack of detailed verification of vehicle dynamics constraints (such as steering angle limits, latest lane-changing points, etc.) and trajectory curvature smoothness. As a result, although some trajectories can be executed, they are insufficient in terms of driving comfort, controllability, and continuity, ensuring that lane-changing behavior is not only reasonable in strategy but also controllable and stable at the execution level.

[0096] Compared with the current autonomous driving lane-changing technology, the advantages of this application are as follows: Strong intelligent decision-making ability: This application introduces a large language model as the core reasoning engine, which can integrate multi-source information such as users, vehicles, and roads, understand complex driving contexts, and generate a more global and adaptive lane-changing plan, significantly improving the model's intelligent decision-making ability in diverse traffic environments.

[0097] High safety: This application formulates safety verification rules based on the rule engine to verify the target lane-changing lane and lane-changing trajectory generated by the model respectively, realizing a full-process safety closed-loop control from decision-making to execution, and solving the problem of poor executability in the existing technology where the model is reasonable but not practical.

[0098] High execution reliability: By introducing the latest lane-changing point control, curvature smoothness constraint, and vehicle dynamic boundary matching mechanism, it effectively prevents sudden lane-changing, continuous lane-changing, or over-limit operations, improves the smoothness of trajectory planning and the controllability of execution, and enhances the stability and comfort of actual driving.

[0099] The autonomous driving lane-changing device based on safety verification proposed in this application further enhances the deployment flexibility of the system. Each module collaborates to achieve a complete functional chain of perception processing, model reasoning, safety review, and control execution. The generated lane-changing instruction can directly drive the vehicle to perform actual operations, support autonomous driving applications in complex traffic scenarios, effectively improve the efficiency of autonomous driving and road safety, and provide technical support for intelligent travel.

[0100] It should be noted that the serial numbers of the embodiments of this application above are only for description and do not represent the superiority or inferiority of the embodiments.

[0101] In the description of the specification, claims and the above-mentioned drawings of this application, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices. Descriptions such as "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit that "first", "second" and "third" are different types.

[0102] In the description of the embodiments of this application, words such as "exemplary", "for example" or "for instance" are used to indicate examples, illustrations or explanations. Any embodiment or design solution described as "exemplary", "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary", "for example" or "for instance" is intended to present relevant concepts in a specific manner.

[0103] In the description of the embodiments of this application, unless otherwise specified, " / " means "or". For example, A / B may mean A or B; "and / or" in the text is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "a plurality of" means two or more than two.

[0104] In some processes described in the embodiments of this application, a plurality of operations or steps appear in a specific order. However, it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of this application or may be executed in parallel. The serial numbers of the operations are only used to distinguish different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0105] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium as described above (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions to enable a terminal device to execute the methods described in various embodiments of this application.

[0106] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present application.

Claims

1. An automatic driving lane-changing method based on security verification, characterized in that, The method includes: S1. Generate a lane change plan for the vehicle under the current driving condition based on the real-time information of the vehicle driving condition obtained and the trained lane change planning model; S2. Formulate safety lane change rules based on the rule engine algorithm and perform safety verification on the lane change plan under the current driving condition; S3. Determine whether the lane change plan under the current driving condition meets the safety lane change rules. If not, enter step S1; if so, enter step S4; S4. Control the vehicle to change lanes according to the lane change plan that passes the verification.

2. The method for automatic driving lane change based on security verification according to claim 1, characterized in that, In the above step S1, the training method of the lane change planning model includes: Obtain the training sample information of the vehicle driving condition and preprocess the training sample information; Use the preprocessed training sample information to pre-train the large language model; Based on the supervised fine-tuning algorithm and the annotation information of the vehicle driving condition for specific lane change tasks, fine-tune the pre-trained large language model; Based on the reinforcement learning algorithm with human feedback, update and iterate the parameters of the fine-tuned large language model to obtain the trained lane change planning model.

3. The method for automatic driving lane change based on security verification according to claim 1, wherein In the above step S1, the vehicle driving condition includes the destination, vehicle speed, vehicle position, traffic density, road type, and driving behavior.

4. The method for an autonomous vehicle to change lanes based on security verification according to claim 2, wherein The preprocessing of the training sample information includes: Perform standardization processing on the training sample information; Extract key feature variables from the standardized training sample information; Combine the context information of the key feature variables and encode the key feature variables with time series or spatial features; Use the feature fusion technology to fuse the encoded key features into a high-dimensional semantic vector, perform dimension conversion, and finally generate the training sample information with a unified dimension.

5. The method for automatic driving lane change based on security verification according to claim 4, characterized in that The standardization processing includes format normalization, outlier removal, and missing value filling.

6. The method for automatic driving lane change based on security verification according to claim 1, wherein In the above step S1, the lane change plan includes the target lane change lane and the lane change trajectory. When performing safety verification on the lane change plan, when both the target lane change lane and the lane change trajectory meet the safety lane change rules, the lane change plan meets the safety lane change rules and the verification passes.

7. The method for automatic driving lane change based on security verification according to claim 6, wherein The judgment criteria for the target lane change lane to meet the safety lane change rules are: On the premise that the user has set a waypoint, the target lane change lane is the same as the lane where the waypoint is located; On the premise that the user has not set a waypoint and the difference between the traffic efficiency of the current driving lane and that of the adjacent lane is greater than or equal to the preset threshold, if the traffic efficiency of the current driving lane is lower than that of the adjacent lane, the target lane change lane is the adjacent lane with the highest traffic efficiency; On the premise that the user has not set a waypoint and the difference between the traffic efficiency of the current driving lane and that of the adjacent lane is less than the preset threshold, the target lane change lane is the middle lane.

8. The method for automatic driving lane change based on security verification according to claim 6, wherein The judgment criteria for the lane change trajectory to meet the safety lane change rules are: The distance from the starting point of the lane change trajectory to the preset mandatory waypoint is greater than or equal to the distance from the preset latest lane change point to the preset mandatory waypoint, and the curvature of the lane change trajectory is less than or equal to the preset curvature threshold.

9. The method for automatic driving lane change based on security verification according to claim 8, wherein The preset curvature threshold is set based on the riding comfort of the user when the vehicle turns.

10. An automatic driving lane-changing system based on the automatic driving lane-changing method based on security verification according to any one of claims 1-9, characterized in that, The system includes: A scenario generation module, which is used to generate a lane change scenario for the vehicle under the current driving scenario based on the real-time information of the vehicle driving scenario obtained and the trained lane change planning model; A verification module, which is used to formulate safe lane change rules based on the rule engine algorithm, perform safety verification on the lane change scenario, and if the lane change scenario does not meet the safe lane change rules, regenerate the lane change scenario until the lane change scenario meets the safe lane change rules; A lane change module, which is used to control the vehicle to change lanes according to the verified lane change scenario.

Images