Automobile automatic driving system and behavior safety method

Abstract

The invention discloses an automobile automatic driving system and a behavior safety method.The system comprises a main channel and a standby channel which are parallel and is composed of a first environment sensing module, a second environment sensing module, an environment fusion module, a track generation module, a standby track generation module, a safety verification module, a state switching module and a track execution module; the first environment sensing module and the second environment sensing module serve as environment sensing of the main channel, and the robustness of environment sensing is improved; the standby track generation module is combined with the second environment perception module to output a standby track, selection of the state switching module is facilitated, and the environment fusion module serves as environment perception fusion of the main channel to improve the environment perception result; the safety verification module verifies the main track and the standby track, output of the tracks does not need to be corrected, and the influence of safety verification and track correction on the availability of the system is reduced; and the state switching module realizes track output redundancy, so that the reliability of the system is improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent driving technology and relates to the safety of intelligent driving systems, and specifically to an automobile automatic driving system and a behavior safety method. Background Art

[0002] In the prior art, in order to achieve the safe behavior of autonomous driving vehicles, similar technical solutions are often used, such as the vehicle and its safe driving method and device disclosed in publication number CN111114540B. The specific principles are as follows: First, the environment perception module is responsible for collecting relevant information from the surrounding environment of the vehicle; secondly, the environment perception module outputs the collected relevant information to the trajectory generation module and the safety verification and trajectory correction module; the safety verification and trajectory correction module verifies and corrects the trajectory output by the trajectory generation module to ensure that a safe driving trajectory is generated, and outputs this safe trajectory to the trajectory execution module; finally, the trajectory execution module controls the vehicle to drive according to the safe trajectory output by the upstream module.

[0003] In the above technical solutions, the execution logic of the safety verification and trajectory correction module is usually constructed based on rules, that is, based on the positional relationship between the ego vehicle and surrounding objects, and the set acceleration and deceleration capabilities of the ego vehicle and surrounding objects, the safety constraints that the ego vehicle should have to ensure safety are inferred. This method has the following defects: 1. Impact of the environmental perception module: As an important component of the autonomous driving system, the performance of the environmental perception module directly affects the effectiveness of the trajectory generation, safety verification and trajectory correction modules.

[0004] If the environmental perception module has insufficient functions, such as a high rate of false detection and missed detection under certain lighting and weather conditions or for specific objects, then even after being processed by the safety verification and trajectory correction modules, it may not be possible to ensure that the output trajectory is safe.

[0005] This is because erroneous information from the environmental perception module will be passed on to downstream modules, causing the entire system to make decisions based on inaccurate data, thereby increasing safety risks.

[0006] 2. Severity of environmental perception module failure: If a serious fault occurs in the environment perception module, such as module crash, the downstream trajectory generation, safety verification, and trajectory correction modules will not work properly.

[0007] In this case, the system cannot output a safe vehicle trajectory, and cannot ensure the safe driving of the vehicle.

[0008] This shows that the safety of the entire autonomous driving system depends to a large extent on the stability and reliability of the environmental perception module.

[0009] 3. Mismatch between preset rules and actual scenarios: Safety verification and trajectory correction modules usually work based on preset rules, which are formulated based on general scenarios and assumptions.

[0010] However, in actual driving, you may encounter various complex and changeable scenarios that may not fully conform to the preset rules.

[0011] If the preset rules do not match the actual scenario, the safety verification and trajectory correction module may misjudge an unsafe scenario as safe, or misjudge a safe scenario as unsafe, thus affecting the overall usability and accuracy of the intelligent driving system.

[0012] In summary, although the safety verification and trajectory correction module is based on rule-based execution logic, this method has obvious shortcomings and needs to be continuously optimized and improved in practical applications. In order to improve the safety and reliability of the autonomous driving system, it is necessary to strengthen the research and development and testing of the environmental perception module to ensure that it can accurately and stably perceive the surrounding environment; at the same time, it is also necessary to continuously improve and optimize the rules and algorithms of the safety verification and trajectory correction module to adapt to more complex and changeable actual driving scenarios. Summary of the invention

[0013] In view of the above problems, the main purpose of the present invention is to design an automobile automatic driving system and a behavioral safety method, which constitute a parallel main channel and a backup channel through a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a backup trajectory generation module, a safety verification module, a state switching module, and a trajectory execution module, so as to solve the problem that the environment perception module has insufficient functions, resulting in incorrect trajectory output or inability to output the trajectory of the entire system, as well as technical problems such as poor system accuracy caused by misjudgment of safety verification and trajectory correction.

[0014] In order to achieve the above purpose, the present invention adopts the following technical solutions: An automobile automatic driving system, the system comprising a parallel main channel and a backup channel; wherein the main channel and the backup channel are composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a backup trajectory generation module, a safety verification module, a state switching module, and a trajectory execution module; The main channel is composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a safety verification module, and a state switching module; The first environment perception module and the second environment perception module receive sensor data respectively, and transmit the perception results to the environment fusion module; the trajectory generation module receives the output of the environment fusion module, generates a main trajectory, and outputs the main trajectory to the security verification module and the state switching module; The backup channel is composed of a second environment perception module, a backup trajectory generation module, a safety verification module, and a state switching module; The second environment perception module receives sensor data and transmits the perception result to the backup trajectory generation module, the backup trajectory generation module generates a backup trajectory according to the received perception result, and outputs the backup trajectory to the security verification module and the state switching module; The state switching module decides to adopt the main trajectory or the backup trajectory based on the verification result output by the safety verification module, and outputs the adopted trajectory; The trajectory execution module receives and executes the trajectory output by the state switching module.

[0015] As a further description of the present invention, the input data of the first environment perception module and the second environment perception module are raw sensor data, including camera data, lidar data, and millimeter wave radar data, and the output data are perception results, including the position, speed, category, and road information of objects in the environment; The first environment perception module processes the sensor data based on the deep learning model to obtain a perception result; The second environmental perception module processes sensor data based on traditional algorithms and lightweight deep learning models to obtain perception results.

[0016] As a further description of the present invention, the environment fusion module takes the perception results of the first environment perception module and the second environment perception module as input to fuse, obtains a fused perception result, and takes the fused perception result as output; The fusion mode of the environment fusion module includes one of a target level fusion algorithm, a probability fusion method, and a weighted voting mechanism.

[0017] As a further description of the present invention, the trajectory generation module takes the fused perception result output by the environment fusion module and the vehicle navigation target as input, outputs the planned driving path of the vehicle, and defines the path as the main trajectory; The trajectory generation module uses a hierarchical planning architecture, a path planning algorithm, and a speed planning algorithm to generate the main trajectory.

[0018] As a further description of the present invention, the backup trajectory generation module takes the perception result of the second environment perception module as input, outputs the planned driving path of the vehicle, and defines the path as the backup trajectory; The backup trajectory generation module uses a pure tracking algorithm, a safety distance setting, and a preset emergency operation to generate a backup trajectory.

[0019] As a further description of the present invention, the safety verification module takes the main trajectory generated by the trajectory generation module, the backup trajectory generated by the backup trajectory generation module, and the perception result as input, performs safety verification, and outputs the verification results of the main trajectory and the backup trajectory; The safety verification method of the safety verification module is one or more of rule-based collision detection, spatiotemporal safety corridor verification, and probabilistic risk assessment.

[0020] As a further description of the present invention, the state switching module selects the final execution trajectory according to the verification result of the safety verification module; the input of the state switching module includes the main trajectory, the backup trajectory, and the safety verification result, and outputs the final execution trajectory; The selection process of the state switching module includes normal working mode, conditions for triggering switching, switching logic, and special case processing.

[0021] As a further description of the present invention, the trajectory execution module realizes the movement of the vehicle according to the trajectory through control instructions based on the output trajectory of the state switching module; The trajectory execution module is implemented by using a lateral controller, a longitudinal controller, and an execution monitoring mechanism.

[0022] A method for the safety of vehicle automatic driving behavior, the method is based on the above system, and includes a main trajectory generation process, a backup trajectory generation process, and a main trajectory and backup trajectory control process; The main trajectory generation process is based on the main channel and includes the following steps: S11: The first environment perception module and the second environment perception module receive sensor data respectively and obtain perception results; S12: The first environment perception module and the second environment perception module output the perception results to the environment fusion module, and the perception result of the first environment perception module is fused with the perception result of the second environment perception module to obtain a fused perception result; S13: The environment fusion module outputs the fused perception results to the trajectory generation module to generate the main trajectory; S14: The trajectory generation module outputs the main trajectory to the safety verification module and the state switching module to verify the safety of the main trajectory; The backup trajectory generation process is based on the backup channel and includes the following steps: S21: The second environment perception module receives sensor data and obtains a perception result; S22: The second environment perception module outputs the perception result to the backup trajectory generation module to generate the backup trajectory; S23: The trajectory generation module outputs the backup trajectory to the safety verification module and the state switching module to verify the safety of the backup trajectory; The main trajectory and backup trajectory control process includes the following steps: S31: The safety verification module outputs the verification results of the main trajectory and the backup trajectory, the main trajectory, and the backup trajectory to the state switching module; S32: The state switching module selects to output the main trajectory or the backup trajectory to the trajectory execution module according to the verification result; S33: The trajectory execution module receives the main trajectory or the backup trajectory and controls the vehicle to execute it.

[0023] Compared with the prior art, the technical effects of the present invention are: The present invention provides an automobile automatic driving system and a behavior safety method, the system comprises a parallel main channel and a backup channel, and the main channel and the backup channel are composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a backup trajectory generation module, a safety verification module, a state switching module, and a trajectory execution module; the present invention improves the robustness of the entire system to the failure and insufficient function of the first environment perception module by using the first environment perception module and the second environment perception module as the environment perception of the main channel; the backup trajectory is output by combining the backup trajectory generation module with the second environment perception module, which is convenient for the selection of the state switching module and improves the reliability of the system; the environment fusion module is used as the output of the first environment perception module and the second environment perception module in the main channel, and is output to the trajectory generation module of the main channel, so as to improve the environmental perception result; the main trajectory and the backup trajectory are verified by the safety verification module, and there is no need to correct the output of the trajectory, so as to reduce the possibility that the safety verification and trajectory correction affect the overall availability of the system; the state switching module is used to switch the trajectory output to the downstream, so as to achieve trajectory output redundancy and improve the reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the system structure of the present invention; Figure 2 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0025] The present invention is described in detail below in conjunction with the accompanying drawings: In one embodiment of the present invention, a vehicle automatic driving system is disclosed, referring to Figure 1As shown, the system includes a parallel main channel and a backup channel; wherein the main channel and the backup channel are composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a backup trajectory generation module, a safety verification module, a state switching module, and a trajectory execution module.

[0026] Specifically, in this embodiment, the main channel is composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a security verification module, and a state switching module; the backup channel is composed of a second environment perception module, a backup trajectory generation module, a security verification module, and a state switching module.

[0027] In the main channel, the first environment perception module and the second environment perception module receive sensor data respectively, and transmit the perception results to the environment fusion module. The trajectory generation module receives the output of the environment fusion module, generates a main trajectory, and outputs the main trajectory to the security verification module and the state switching module. In the backup channel, the second environment perception module receives sensor data and transmits the perception result to the backup trajectory generation module, the backup trajectory generation module generates a backup trajectory according to the received perception result, and outputs the backup trajectory to the security verification module and the state switching module; The state switching module decides to adopt the main trajectory or the backup trajectory based on the verification result output by the safety verification module, and outputs the adopted trajectory; the trajectory execution module receives and executes the trajectory output by the state switching module.

[0028] A detailed analysis of each module included in the above main channel and backup channel is as follows: 1. Environmental perception (first environmental perception module, second environmental perception module) The first environment perception module and the second environment perception module perceive and identify objects, road information, etc. in the vehicle's surrounding environment through raw sensor data.

[0029] Input: raw sensor data, including camera data, lidar data, millimeter wave radar data (images, point clouds, radar data, etc.); Output: Perception results (object detection results), including the location, speed, category, and road information of objects in the environment.

[0030] Implementation of the first environment perception module and the second environment perception module: The first environment perception module processes the sensor data based on the deep learning model to obtain a perception result; The second environmental perception module processes sensor data based on traditional algorithms and lightweight deep learning models to obtain perception results.

[0031] 2. Environmental fusion module The environment fusion module fuses the perception results of the first environment perception module and the second environment perception module to provide more reliable environment perception results.

[0032] Input: perception results of the first environment perception module and the second environment perception module; Output: fused perception result (fused scene understanding result).

[0033] The fusion mode of the environment fusion module includes one of a target level fusion algorithm, a probability fusion method, and a weighted voting mechanism.

[0034] The environment fusion module is used to fuse the perception results of the first environment perception module and the second environment perception module before the trajectory generation module, thereby reducing the possibility of mistakenly entering a safe driving behavior state due to the safety verification module misdetecting that the first environment perception module is insufficiently functional.

[0035] 3. Trajectory generation module The trajectory generation module plans the vehicle's driving path based on the fused perception results and defines the path as the main trajectory.

[0036] Input: fused perception results and vehicle navigation targets; Output: The planned driving path of the vehicle (including position, speed, acceleration, etc.).

[0037] The trajectory generation module uses a hierarchical planning architecture, a path planning algorithm, and a speed planning algorithm to generate the main trajectory.

[0038] 4. Backup trajectory generation module The backup trajectory generation module plans a vehicle driving path based on the perception result of the second environment perception module and defines the path as a backup trajectory.

[0039] Input: the perception result of the second environment perception module; Output: The planned driving path of the vehicle (including position, speed, acceleration, etc.).

[0040] The backup trajectory generation module uses a pure tracking algorithm, a safety distance setting, and a preset emergency operation to generate a backup trajectory.

[0041] 5. Security Verification Module The safety verification module is used to verify the safety of the main trajectory and the backup trajectory. This module is different from the "safety verification and trajectory correction module" in the traditional solution. In this embodiment, the safety verification module is not responsible for trajectory correction, thereby reducing the complexity of the module, improving the reliability of verification, and reducing false triggering caused by inappropriate preset rules. By verifying the main trajectory and the backup trajectory at the same time, a reliable decision-making basis is provided for the state switching module.

[0042] Input: the main trajectory generated by the trajectory generation module, the backup trajectory generated by the backup trajectory generation module, and the perception results of the first environment perception module and the second environment perception module; Output: Safety verification results of the main trajectory and backup trajectory.

[0043] The safety verification method of the safety verification module is one or more of rule-based collision detection, spatiotemporal safety corridor verification, and probabilistic risk assessment.

[0044] It should be noted that the safety verification module simultaneously receives the main trajectory generated by the main channel and the backup trajectory generated by the backup channel, reducing false triggering caused by inappropriate preset rules. When the safety verification module verifies that the main channel fails, the vehicle automatically switches to the backup channel by default, thereby ensuring safe driving of the vehicle.

[0045] 6. State switching module The state switching module selects the final execution trajectory based on the security verification results.

[0046] Input: main trajectory, backup trajectory, safety verification results; Output: Final execution trace.

[0047] The selection process of the state switching module includes normal working mode, conditions for triggering switching, switching logic, and special case processing.

[0048] 7. Trajectory execution module The trajectory execution module executes the trajectory selected by the state switching module and realizes the movement of the vehicle along the trajectory through control instructions.

[0049] Input: Finalized execution trace; Output: control instructions (steering, acceleration, braking, etc.).

[0050] The trajectory execution module is implemented by using a lateral controller, a longitudinal controller, and an execution monitoring mechanism.

[0051] Through the above content, the automatic driving system of the present invention is disclosed, which has the following advantages over the prior art: 1. The present invention uses the first environment perception module and the second environment perception module as the main channel environment perception, thereby improving the robustness of the entire system against the failure and insufficient function of the first environment perception module; 2. The present invention combines the backup trajectory generation module with the second environment perception module to output the backup trajectory, which facilitates the selection of the state switching module and improves the reliability of the system; 3. The present invention adopts an environment fusion module as the output of the first environment perception module and the second environment perception module in the main channel, and outputs it to the trajectory generation module of the main channel to improve the environment perception result; 4. The present invention verifies the main trajectory and the backup trajectory through the safety verification module, without the need to correct the output of the trajectory, thereby reducing the possibility that safety verification and trajectory correction affect the overall availability of the system; 5. The present invention adopts a state switching module to switch the trajectory output to the downstream, realize trajectory output redundancy, and improve the reliability of the system.

[0052] In another embodiment of the present invention, a method for ensuring the safety of an automatic driving behavior of an automobile is disclosed. Figure 2 As shown, the method is based on the above-mentioned automatic driving system, including a main trajectory generation process, a backup trajectory generation process, and a main trajectory and backup trajectory control process.

[0053] The main trajectory generation process is based on the main channel and includes the following steps: S11: The first environment perception module and the second environment perception module receive sensor data respectively and obtain perception results; S12: The first environment perception module and the second environment perception module output the perception results to the environment fusion module, and the perception result of the first environment perception module is fused with the perception result of the second environment perception module to obtain a fused perception result; S13: The environment fusion module outputs the fused perception results to the trajectory generation module to generate the main trajectory; S14: The trajectory generation module outputs the main trajectory to the safety verification module and the state switching module to verify the safety of the main trajectory.

[0054] The backup trajectory generation process is based on the backup channel and includes the following steps: S21: The second environment perception module receives sensor data and obtains a perception result; S22: The second environment perception module outputs the perception result to the backup trajectory generation module to generate the backup trajectory; S23: The trajectory generation module outputs the backup trajectory to the safety verification module and the state switching module to verify the safety of the backup trajectory; The main trajectory and backup trajectory control process includes the following steps: S31: The safety verification module outputs the verification results of the main trajectory and the backup trajectory, the main trajectory, and the backup trajectory to the state switching module; S32: The state switching module selects to output the main trajectory or the backup trajectory to the trajectory execution module according to the verification result; S33: The trajectory execution module receives the main trajectory or the backup trajectory and controls the vehicle to execute it.

[0055] Specifically, in this embodiment, the steps of cooperating among the modules in the above-mentioned autonomous driving system to implement the behavior safety method are described in detail as follows: In this embodiment, in the above step S11 and step S21, the first environment perception module and the second environment perception module perceive the input raw sensor data and output the perception results.

[0056] The first environment perception module uses a deep learning model for high-precision perception. The perception process mainly includes data preprocessing, feature extraction, target detection and recognition, and result output, as follows: Data preprocessing: Preprocess the input raw sensor data (such as images, point clouds, radar data, etc.), including cleaning, denoising, calibration, etc., to ensure the accuracy and consistency of the data; Feature extraction: The preprocessed data is input into the deep learning model, which automatically extracts features from the data, including edges, textures, shapes, depth, etc., through structures such as convolutional layers and pooling layers. Object Detection and Recognition: Based on the extracted features, the deep learning model is used to detect objects and determine the location, speed, and category information of the objects. Result output: The deep learning model outputs the detection results in the form of object detection results, including information such as the position, speed (if measurable), and category of each detected object.

[0057] The second environment perception module uses traditional algorithms and lightweight deep learning models for high-precision perception. The perception process mainly includes data preprocessing, feature extraction, target detection and preliminary recognition, deep learning and further recognition, result output and post-processing, as follows: Data preprocessing: Preprocess the input raw sensor data (such as images, point clouds, radar data, etc.), including cleaning, denoising, calibration, etc., to ensure the accuracy and consistency of the data; Feature extraction: The preprocessed data uses traditional algorithms (such as edge detection and filtering in image processing) and simplified deep learning models (such as lightweight CNN networks, such as MobileNets) to extract key features in the data, including edges, textures, shapes, depth, etc.; Target detection and preliminary recognition: Use traditional algorithms to determine the location of the target (such as a bounding box) and preliminarily determine the target category (such as pedestrians, vehicles, obstacles, etc.); Deep learning and further recognition: For targets that require higher precision recognition, the first environment perception module inputs the above recognition results into a more complex deep learning model for further recognition. The deep learning model includes a deeper neural network, a more complex feature extraction layer, etc.; Result output and post-processing: The detection results of the second environment perception are output in the form of object detection results, including information such as the position, speed (if measurable), and category of each detected object.

[0058] In this embodiment, in the above step S12, the environment fusion module fuses the perception results of the first environment perception module and the second environment perception module, and outputs the fused perception results.

[0059] In this embodiment, the fusion method of the environment fusion module includes but is not limited to a target level fusion algorithm, a probabilistic fusion method, and a weighted voting mechanism.

[0060] The fusion process of the target level fusion algorithm is as follows: the environment fusion module receives the perception results from the first environment perception module and the second environment perception module; the target tracking algorithm is used to perform data association on the perception results of the first environment perception module and the second environment perception module; a fusion decision is made based on the association results, and for objects that are successfully associated, the reliability and accuracy of their position, speed and other information are weightedly fused; the fused perception results are output, including the position, speed and category of the object.

[0061] The fusion process of the probabilistic fusion method (Bayesian fusion) is as follows: the environmental fusion module receives the perception results from the first environmental perception module and the second environmental perception module; the probability of the object appearing is calculated for the perception results of the first environmental perception module and the second environmental perception module; the probability results of the first environmental perception module and the second environmental perception module are fused using the Bayesian theorem or related probabilistic fusion methods to obtain the posterior probability of the object's existence; based on the fused probability results, determine whether to include the object in the fused perception results (usually an object whose probability exceeds a set threshold), determine its position, speed, and category information, and output the fused perception results.

[0062] The fusion process of the weighted voting mechanism is as follows: the environment fusion module receives the perception results from the first environment perception module and the second environment perception module; based on historical performance evaluation, real-time sensor status or other relevant factors, weights are assigned to the perception results of the first environment perception module and the second environment perception module; weighted voting is performed on the perception results of each object, and whether to include the object in the fused perception results is determined based on the support of the weighted vote; based on the results of the weighted vote, the fused perception results are determined and output.

[0063] In this embodiment, in the above step S13, the trajectory generation module plans the driving trajectory based on the fused perception result, including the following steps: The trajectory generation module receives the fused perception results and navigation targets, and preprocesses the input data, including data cleaning, verification, formatting, coordinate conversion, time synchronization, etc. Use the path planning algorithm to search the optimal or feasible path from the current position to the navigation target in the environment map; evaluate the length, safety, feasibility, etc. of the multiple paths searched, and select the optimal path as the vehicle's driving path; When selecting a path, a speed planning algorithm (such as dynamic programming, quadratic programming, etc.) is used to generate a smooth speed curve based on factors such as road conditions, traffic rules, and obstacle distribution; Based on the speed curve, calculate the acceleration and deceleration requirements of the vehicle at each position point to ensure that the vehicle can smoothly travel according to the planned speed curve; The driving path obtained above is combined with the speed curve to generate a detailed motion trajectory, and the generated trajectory is optimized, and the optimized trajectory is output as the main trajectory.

[0064] In this embodiment, in the above step S22, the backup trajectory generation module plans the driving trajectory based on the perception result of the second environment perception module, including the following steps: The backup trajectory generation module receives the perception results of the second environment perception module and uses a simplified planning algorithm, such as the Pure Pursuit algorithm, to follow the path. Based on the current environment perception results and the dynamic model of the vehicle, it directly calculates the path that the vehicle should follow to ensure fast response and calculation stability. Perform speed planning and set a safety distance. The expression is: d_safe = v * t_response + v² / (2*a_max) + margin; Where v is the current speed, t_response is the system response time, a_max is the maximum deceleration, and margin is the additional safety margin (larger than the main channel); According to the possible emergency situations of the vehicle, a series of emergency operation plans are preset, including but not limited to emergency braking plan, emergency avoidance plan, and safe parking plan; emergency braking plan: when the distance to the obstacle ahead is detected to be less than the safe distance, the vehicle will brake at the maximum allowable deceleration; emergency avoidance plan: when there is available avoidance space, the vehicle will execute the predefined avoidance trajectory; safe parking plan: when the system cannot ensure the safety of continued driving, the vehicle will execute the safe parking procedure on the shoulder or in the safe passage; Based on the results of the above steps, a backup safety trajectory is generated and output as a backup trajectory.

[0065] In this embodiment, in the above step S14 and step S23, the safety verification module performs safety verification of the main trajectory and the backup trajectory based on the input main trajectory, backup trajectory, and environmental perception results, and outputs the safety verification result. Specifically, the process of the safety verification module includes the following steps: The safety verification module receives the main trajectory, the backup trajectory, and the environmental perception result; Based on the received information, collision detection is performed based on predefined rules, including but not limited to the minimum safe distance between the vehicle and the obstacle, speed limit, traffic rules, etc.; by comparing the vehicle position on the main trajectory and the backup trajectory with the obstacle position in the environment, it is determined whether there is a potential collision risk; For the trajectory after collision detection, the time-space safety corridor verification is carried out to check whether the main trajectory and the backup trajectory always remain in the time-space safety corridor to ensure the safety of the vehicle during driving; For the trajectory verified in the spatiotemporal safety corridor, a probabilistic risk assessment is conducted to evaluate the safety of the main trajectory and backup trajectory based on the uncertainty factors in the environment (including weather conditions, the behavior of traffic participants, etc.); more comprehensive safety assessment results are obtained by calculating indicators such as collision probability and accident severity; Based on the above verification and evaluation, the safety verification module outputs the safety verification results of the main trajectory and the backup trajectory, including the safety level of the trajectory, potential risk points, recommended risk avoidance measures, etc.

[0066] In this embodiment, in the above step S32, the state switching module outputs the final execution trajectory based on the verification results of the main trajectory and the backup trajectory. Specifically, the switching logic of the state switching module includes the following: In normal working mode, the main trace output by the main channel is used by default to continuously monitor the verification results of the safety verification module; The conditions for triggering the switch include: when the safety verification module determines that there is a safety risk in the main trajectory, when the verification result of the safety verification module cannot be obtained, when it is detected that the first environment perception module of the main channel has insufficient function or failure; The switching logic includes immediate switching, smooth switching, and conservative strategy. In the case of immediate switching, the system immediately switches to the backup track when a safety risk is detected. In the case of smooth switching, the system uses a smooth transition of the track in non-emergency situations. In the case of conservative strategy, the system switches to the backup track by default when the status is uncertain. Special situation handling includes: when the first environmental perception module of the main channel is insufficient in function, switching to the backup channel to ensure safe driving behavior and monitoring the output quality of the environmental fusion module; when the first environmental perception module of the main channel fails, the output of the trajectory generation module is detected to be abnormal, and automatic switching to the backup trajectory is performed; when the safety verification module has no response, it switches to the backup trajectory by default and starts the fault diagnosis program.

[0067] In this embodiment, in the above step S33, the trajectory execution module controls the vehicle based on the execution trajectory finally determined by the state switching module. Specifically, a specific control instruction is generated according to the trajectory to control the vehicle to travel along the predetermined trajectory, including the following steps: The trajectory execution module receives the execution trajectory finally determined by the state switching module, including the main trajectory or the backup trajectory; Analyze the received trajectory and extract key information, such as target position, target speed, target acceleration, etc., as well as the change patterns of the above parameters over time; Generate a steering control command using a lateral controller (such as a PID controller, an MPC controller, etc.) according to lateral information of the trajectory, wherein the lateral information includes but is not limited to the curvature of the path and the deviation of the current position of the vehicle from the path; generating an acceleration or braking control command using a longitudinal controller according to longitudinal information of the trajectory, the longitudinal information including but not limited to a difference between a target speed and a current speed; Longitudinal controllers may involve strategies such as speed tracking and distance keeping to ensure that the vehicle can smoothly follow the predetermined speed changes; The control instructions generated by the lateral controller and the longitudinal controller are integrated to form a final control command, which is output to the vehicle's actuators, such as the steering system, throttle and brake system, to achieve real-time control of the vehicle.

[0068] At the same time, a monitoring mechanism is implemented for the vehicle. If large deviations or abnormal situations are found, timely adjustments will be made or the emergency response mechanism will be triggered. The status information during the execution process will be fed back to the upper-level control system for further optimization and decision-making, and the difference between the vehicle's actual response and the control instructions will be detected in real time.

[0069] Through the above content, the behavioral safety method of the present invention is disclosed. The method constructs two environment perception algorithm models. The model of the first environment perception module of the main channel is the main perception algorithm model, which has complete functions and balances the false detection rate and the missed detection rate; the model of the second environment perception module of the backup channel is the secondary perception algorithm model, which may not have complete functions, but should ensure the detection rate of safety risk objects; the construction and training of the second environment perception module ensures that it has a more robustness than the first environment perception module model, therefore, the models of the first environment perception module and the second environment perception module have different construction and training methods. It should be noted that in this embodiment, the construction and training methods of the first environment perception module and the second environment perception module are any method in the prior art and do not belong to the scope of protection of this embodiment.

[0070] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. An automatic driving system for a vehicle, characterized in that: The system includes a parallel main channel and a backup channel; wherein the main channel and the backup channel are composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a backup trajectory generation module, a safety verification module, a state switching module, and a trajectory execution module; The main channel is composed of a first environment perception module, a second environment perception module, an environment fusion module, a trajectory generation module, a safety verification module, and a state switching module; The first environment perception module and the second environment perception module receive sensor data respectively, and transmit the perception results to the environment fusion module; the trajectory generation module receives the output of the environment fusion module, generates a main trajectory, and outputs the main trajectory to the security verification module and the state switching module; The backup channel is composed of a second environment perception module, a backup trajectory generation module, a safety verification module, and a state switching module; The second environment perception module receives sensor data and transmits the perception result to the backup trajectory generation module, the backup trajectory generation module generates a backup trajectory according to the received perception result, and outputs the backup trajectory to the security verification module and the state switching module; The state switching module decides to adopt the main trajectory or the backup trajectory based on the verification result output by the safety verification module, and outputs the adopted trajectory; The trajectory execution module receives and executes the trajectory output by the state switching module.

2. The automatic driving system for a vehicle according to claim 1, characterized in that: The input data of the first environment perception module and the second environment perception module are raw sensor data, including camera data, lidar data, and millimeter wave radar data, and the output data are perception results, including the position, speed, category, and road information of objects in the environment; The first environment perception module processes the sensor data based on the deep learning model to obtain a perception result; The second environmental perception module processes sensor data based on traditional algorithms and lightweight deep learning models to obtain perception results.

3. The automatic driving system for a vehicle according to claim 1, characterized in that: The environment fusion module takes the perception results of the first environment perception module and the second environment perception module as input to fuse, obtains a fused perception result, and takes the fused perception result as output; The fusion mode of the environment fusion module includes one of a target level fusion algorithm, a probability fusion method, and a weighted voting mechanism.

4. The automatic driving system for a vehicle according to claim 1, characterized in that: The trajectory generation module takes the fused perception result output by the environment fusion module and the vehicle navigation target as input, outputs the planned driving path of the vehicle, and defines the path as the main trajectory; The trajectory generation module uses a hierarchical planning architecture, a path planning algorithm, and a speed planning algorithm to generate the main trajectory.

5. The automatic driving system for a vehicle according to claim 1, characterized in that: The backup trajectory generation module takes the perception result of the second environment perception module as input, outputs the planned driving path of the vehicle, and defines the path as the backup trajectory; The backup trajectory generation module uses a pure tracking algorithm, a safety distance setting, and a preset emergency operation to generate a backup trajectory.

6. The automatic driving system for a vehicle according to claim 1, characterized in that: The safety verification module takes the main trajectory generated by the trajectory generation module, the backup trajectory generated by the backup trajectory generation module, and the perception result as input, performs safety verification, and outputs the verification results of the main trajectory and the backup trajectory; The safety verification method of the safety verification module is one or more of rule-based collision detection, spatiotemporal safety corridor verification, and probabilistic risk assessment.

7. The automatic driving system for a vehicle according to claim 1, characterized in that: The state switching module selects the final execution trajectory according to the verification result of the safety verification module; the input of the state switching module includes the main trajectory, the backup trajectory, and the safety verification result, and outputs the final execution trajectory; The selection process of the state switching module includes normal working mode, conditions for triggering switching, switching logic, and special case processing.

8. The automatic driving system for a vehicle according to claim 1, characterized in that: The trajectory execution module realizes the movement of the vehicle according to the trajectory through control instructions based on the output trajectory of the state switching module; The trajectory execution module is implemented by using a lateral controller, a longitudinal controller, and an execution monitoring mechanism.

9. A method for ensuring the safety of vehicle automatic driving behavior according to the system of any one of claims 1 to 8, characterized in that: The method includes a main trajectory generation process, a backup trajectory generation process, and a main trajectory and backup trajectory control process; The main trajectory generation process is based on the main channel and includes the following steps: S11: The first environment perception module and the second environment perception module receive sensor data respectively and obtain perception results; S12: The first environment perception module and the second environment perception module output the perception results to the environment fusion module, and the perception result of the first environment perception module is fused with the perception result of the second environment perception module to obtain a fused perception result; S13: The environment fusion module outputs the fused perception results to the trajectory generation module to generate the main trajectory; S14: The trajectory generation module outputs the main trajectory to the safety verification module and the state switching module to verify the safety of the main trajectory; The backup trajectory generation process is based on the backup channel and includes the following steps: S21: The second environment perception module receives sensor data and obtains a perception result; S22: The second environment perception module outputs the perception result to the backup trajectory generation module to generate the backup trajectory; S23: The trajectory generation module outputs the backup trajectory to the safety verification module and the state switching module to verify the safety of the backup trajectory; The main trajectory and backup trajectory control process includes the following steps: S31: The safety verification module outputs the verification results of the main trajectory and the backup trajectory, the main trajectory, and the backup trajectory to the state switching module; S32: The state switching module selects to output the main trajectory or the backup trajectory to the trajectory execution module according to the verification result; S33: The trajectory execution module receives the main trajectory or the backup trajectory and controls the vehicle to execute it.

Images