Method and device for automatic driving safety decision-making by fusing human defensive driving experience in a perception-limited scenario

Abstract

The application relates to an automatic driving safety decision method and device fusing human defensive driving experience under a perception limited scene. The method comprises the following steps: acquiring state information of a target vehicle and perception information of a surrounding traffic environment; in the case that the target vehicle is determined to be in the perception limited scene according to the perception information, defensive driving experience judgment processing is performed on a current driving scene of the target vehicle based on the state information and a preset defensive driving experience knowledge base, and a processing result is obtained; in the case that the processing result represents that the target vehicle triggers the defensive driving experience, current scene parameters are input into a pre-trained phantom obstacle generation model according to the type characteristics of the perception limited scene, and a phantom obstacle is obtained; and a defensive safety trajectory is generated according to the phantom obstacle through a preset trajectory planning control algorithm, and the target vehicle is controlled to travel according to the defensive safety trajectory. The method can improve the accuracy of an automatic driving strategy.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to an autonomous driving safety decision-making method and device that integrates human defensive driving experience in perception-limited scenarios. Background Technology

[0002] As autonomous driving technology advances towards higher levels of intelligence, ensuring the safety and reliability of the system in complex traffic environments has become a core issue that urgently needs to be addressed. Perception-limited scenarios refer to situations where the vehicle cannot directly capture surrounding risk information through its perception system, leading to potential risks during driving.

[0003] Existing methods typically rely on real-time environmental information acquired by onboard sensors (such as cameras, LiDAR, and millimeter-wave radar) to identify and track the motion of detectable targets such as vehicles, pedestrians, and obstacles, and then build a risk model based on this information. The system predicts the future trajectories of these targets, calculates the collision risk between the vehicle and them, and plans a safe driving trajectory that avoids all identified risks based on optimization algorithms.

[0004] However, this "what you see is what you avoid" strategy suffers from a lack of accuracy. Summary of the Invention

[0005] Therefore, it is necessary to provide an autonomous driving safety decision-making method and device that integrates human defensive driving experience in perception-limited scenarios to address the above-mentioned technical problems and improve the accuracy of autonomous driving strategies.

[0006] Firstly, this application provides an autonomous driving safety decision-making method that integrates human defensive driving experience in perception-limited scenarios, including:

[0007] Acquire the target vehicle's status information and the surrounding traffic environment's perception information;

[0008] When the target vehicle is determined to be in a perception-limited scenario based on perception information, defensive driving experience judgment processing is performed on the target vehicle's current driving scenario based on state information and a pre-set defensive driving experience knowledge base to obtain the processing result.

[0009] Given that the processing results characterize the target vehicle's defensive driving experience, the current scene parameters are input into a pre-trained phantom obstacle generation model based on the type characteristics of the perception-limited scene to obtain phantom obstacles.

[0010] Based on the phantom obstacles, a pre-set trajectory planning and control algorithm generates a defensive safety trajectory and controls the target vehicle to travel along the defensive safety trajectory.

[0011] Secondly, this application also provides an autonomous driving safety decision-making device that integrates human defensive driving experience in perception-limited scenarios, comprising:

[0012] The acquisition module is used to acquire the status information of the target vehicle and the perception information of the surrounding traffic environment;

[0013] The processing module is used to perform defensive driving experience judgment processing on the current driving scenario of the target vehicle based on state information and a preset defensive driving experience knowledge base when it is determined that the target vehicle is in a perception-limited scenario based on perception information, and to obtain the processing result.

[0014] The input module is used to input the current scene parameters into the pre-trained phantom obstacle generation model based on the type characteristics of the perception-limited scene, when the processing result represents the target vehicle triggering defensive driving experience, in order to obtain phantom obstacles.

[0015] The generation module is used to generate a defensive safety trajectory based on phantom obstacles using a preset trajectory planning and control algorithm, and to control the target vehicle to drive along the defensive safety trajectory.

[0016] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0017] Acquire the target vehicle's status information and the surrounding traffic environment's perception information;

[0018] When the target vehicle is determined to be in a perception-limited scenario based on perception information, defensive driving experience judgment processing is performed on the target vehicle's current driving scenario based on state information and a pre-set defensive driving experience knowledge base to obtain the processing result.

[0019] Given that the processing results characterize the target vehicle's defensive driving experience, the current scene parameters are input into a pre-trained phantom obstacle generation model based on the type characteristics of the perception-limited scene to obtain phantom obstacles.

[0020] Based on the phantom obstacles, a pre-set trajectory planning and control algorithm generates a defensive safety trajectory and controls the target vehicle to travel along the defensive safety trajectory.

[0021] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0022] Acquire the target vehicle's status information and the surrounding traffic environment's perception information;

[0023] When the target vehicle is determined to be in a perception-limited scenario based on perception information, defensive driving experience judgment processing is performed on the target vehicle's current driving scenario based on state information and a pre-set defensive driving experience knowledge base to obtain the processing result.

[0024] Given that the processing results characterize the target vehicle's defensive driving experience, the current scene parameters are input into a pre-trained phantom obstacle generation model based on the type characteristics of the perception-limited scene to obtain phantom obstacles.

[0025] Based on the phantom obstacles, a pre-set trajectory planning and control algorithm generates a defensive safety trajectory and controls the target vehicle to travel along the defensive safety trajectory.

[0026] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0027] Acquire the target vehicle's status information and the surrounding traffic environment's perception information;

[0028] When the target vehicle is determined to be in a perception-limited scenario based on perception information, defensive driving experience judgment processing is performed on the target vehicle's current driving scenario based on state information and a pre-set defensive driving experience knowledge base to obtain the processing result.

[0029] Given that the processing results characterize the target vehicle's defensive driving experience, the current scene parameters are input into a pre-trained phantom obstacle generation model based on the type characteristics of the perception-limited scene to obtain phantom obstacles.

[0030] Based on the phantom obstacles, a pre-set trajectory planning and control algorithm generates a defensive safety trajectory and controls the target vehicle to travel along the defensive safety trajectory.

[0031] The aforementioned autonomous driving safety decision-making method and device, which integrates human defensive driving experience in perception-limited scenarios, formalizes and constructs this experience into a knowledge base. This enables the target vehicle to proactively identify potential risks in perception-limited scenarios, rather than passively responding to detected obstacles. Traditional methods suffer from missing decision-making information due to blind spots, resulting in aggressive strategies and a high risk of sudden accidents. This application, through defensive driving experience judgment processing, predicts hidden risks from the unexpected braking or steering behavior of vehicles in adjacent lanes, extending the decision-making basis from "what is seen is avoided" to "what is unseen can also be predicted," fundamentally solving the problem of missing information in perception-limited scenarios. Furthermore, through a phantom obstacle generation model, virtual obstacles simulating vulnerable road users are proactively generated within occluded areas, transforming unknown potential risks into knowable, quantifiable, and avoidable decision inputs. The trajectory planning algorithm then generates a defensive safety trajectory with sufficient safety margin, ensuring that the vehicle is already in a safe state of pre-deceleration and stable posture when facing suddenly appearing pedestrians or non-motorized vehicles, rather than a passive response of emergency braking. This technical solution improves the accuracy of autonomous driving decision-making from precise avoidance of known risks to proactive prediction of unseen risks, effectively making up for the inherent limitations of physical perception systems and significantly enhancing driving safety and system robustness in complex traffic environments. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is an internal structural diagram of a computer device in one embodiment;

[0034] Figure 2 This is a flowchart illustrating an autonomous driving safety decision-making method that integrates human defensive driving experience in a perception-limited scenario, as shown in one embodiment.

[0035] Figure 3 This is a flowchart illustrating an autonomous driving safety decision-making method that integrates human defensive driving experience in a perception-limited scenario, as described in another embodiment.

[0036] Figure 4 This is a flowchart illustrating an autonomous driving safety decision-making method that integrates human defensive driving experience in a perception-limited scenario, as described in another embodiment.

[0037] Figure 5This is a flowchart of an autonomous driving safety decision-making method that integrates human defensive driving experience in a perception-limited scenario, as shown in one embodiment.

[0038] Figure 6 This is a schematic diagram of an unexpected braking scenario in one embodiment;

[0039] Figure 7 This is a schematic diagram of an unexpected turning scenario in one embodiment;

[0040] Figure 8 Here is a flowchart of online defensive driving experience monitoring in one embodiment;

[0041] Figure 9 This is a schematic diagram of an occlusion area caused by limited perception in one embodiment;

[0042] Figure 10 Here is a flowchart of the process for generating a phantom obstacle model in one embodiment;

[0043] Figure 11 This is a structural block diagram of an autonomous driving safety decision-making device that integrates human defensive driving experience in a perception-limited scenario, as shown in one embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 1 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores relevant data during vehicle control. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an autonomous driving safety decision-making method that integrates human defensive driving experience in perception-limited scenarios.

[0046] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0047] In one exemplary embodiment, such as Figure 2 As shown, a method for autonomous driving safety decision-making that integrates human defensive driving experience in perception-limited scenarios is provided, and this method is applied to… Figure 1 Taking the server in the example, the explanation includes the following steps 201 to 204. Wherein:

[0048] Step 201: Obtain the status information of the target vehicle and the perception information of the surrounding traffic environment.

[0049] The state information refers to the set of data characterizing the current operating state of the target vehicle, including but not limited to dynamic parameters such as vehicle speed, acceleration, heading angle, position coordinates, and yaw rate. This information is collected in real time by onboard sensors and constitutes the basic representation of the vehicle's motion state.

[0050] The perceived information of the surrounding traffic environment refers to the surrounding traffic scene data obtained by the target vehicle through the vehicle-mounted perception system (including but not limited to cameras, lidar, millimeter-wave radar, ultrasonic sensors, etc.), including but not limited to: the position, speed, acceleration, heading angle, size, and turn signal status of other traffic participants (such as vehicles in adjacent lanes, vehicles in the same lane, and oncoming vehicles); static environmental information such as lane line positions, lane boundaries, and road speed limits; and identification information of vulnerable road users (pedestrians, non-motorized vehicles, etc.).

[0051] In this embodiment of the application, the server receives the status information uploaded by the target vehicle and the perception information of the surrounding traffic environment.

[0052] For example, the target vehicle is a passenger car equipped with a Level 4 autonomous driving system, traveling on a main urban road. The vehicle uses an IMU and wheel speed sensor to collect data showing its current speed as 45 km / h, acceleration as 0.2 m / s², and heading angle as 87.3°. Through a fusion perception system using a forward-facing camera and millimeter-wave radar, it identifies a large bus in the second lane from the left (the vehicle ahead in the adjacent lane), with coordinates (120, 3.5), speed of 38 km / h, acceleration of -1.8 m / s², heading angle of 86.5°, length of 12 meters, width of 2.5 meters, and turn signal not activated. Simultaneously, it detects the vehicle ahead in its lane with coordinates (115, 1.8) and speed of 42 km / h. The lane detection system identifies the current lane as a six-lane, two-way urban road with a lane width of 3.5 meters and a speed limit of 60 km / h. This data, after preprocessing on the vehicle, is uploaded to a cloud server via a 5G network at a frequency of 10 Hz.

[0053] In another embodiment, the target vehicle is an unmanned delivery vehicle operating within a closed park. Vehicle status information includes a current speed of 15 km / h and a heading angle of 270 degrees. Perception information includes a speed bump (static obstacle) 20 meters ahead, and an electric tricycle turning left in the adjacent lane on the right (vehicle ahead in the adjacent lane), with a lateral speed of 0.8 m / s and a lateral offset of 0.6 meters; its left turn signal is activated. The server receives and caches this information for subsequent scene recognition and decision-making.

[0054] Step 202: If it is determined from the perception information that the target vehicle is in a perception-limited scenario, defensive driving experience judgment processing is performed on the current driving scenario of the target vehicle based on the state information and the preset defensive driving experience knowledge base to obtain the processing result.

[0055] Among them, perception-limited scenarios refer to driving scenarios where, due to external obstructions, sensor performance limitations, or environmental interference, the target vehicle is unable to directly detect potential risk information in certain areas through its own perception system, thus creating blind spots. Typical scenarios include areas where the field of vision is obstructed by large vehicles, blind spots on curves, and areas where perception is degraded due to inclement weather. This application focuses on vision-limited scenarios caused by obstruction from vehicles in adjacent lanes.

[0056] A defensive driving experience knowledge base refers to a collection of knowledge that formally expresses and digitally stores the forward-looking and preventative driving experiences accumulated by human drivers over long-term driving practice, using metric temporal logic. This knowledge base includes the preconditions, triggering logic, judgment rules, and corresponding coping strategies for various risk scenarios, and possesses good interpretability and verifiability.

[0057] Defensive driving experience judgment processing refers to the process of making unexpected judgments on the driving behavior of the vehicle ahead in the adjacent lane based on the perceived and state information of the current scene and by invoking the metric timing logic rules in the defensive driving experience knowledge base. The processing results include unexpected braking and unexpected steering results, which are used to characterize whether the corresponding defensive driving experience has been triggered.

[0058] In this embodiment, the server identifies whether the target vehicle is currently in a perception-limited scenario based on the perception information obtained in step 201. If it is identified as a perception-limited scenario, the server performs experience-based judgment processing by combining state information and a defensive driving experience knowledge base.

[0059] For example, in step 201, the server receives information that a large bus (a vehicle in the adjacent lane) is in front of the target vehicle. The server analyzes the perception data and finds that the bus is large (12 meters long and 2.5 meters wide), and its position overlaps laterally with the target vehicle's view. Based on the geometric occlusion model, it is confirmed that a continuous blind spot is formed behind and to the right of the bus, and the target vehicle cannot directly perceive whether other vehicles or pedestrians are within this blind spot. Therefore, the server determines that the current scenario is one of limited perception, specifically a "visual restriction scenario caused by occlusion from a vehicle in the adjacent lane".

[0060] The server then invokes the timing logic rules for measuring "unintended braking" scenarios from the defensive driving experience knowledge base. These rules include: preconditions (the presence of a vehicle in the adjacent lane and the vehicle is braking), exclusion conditions (braking is not due to speeding or a collision risk with the vehicle in front), and state continuity conditions (effective braking has not ended). From the state information obtained in step 201, the server extracts the acceleration of the vehicle in the adjacent lane as -1.8 m / s² (for 3 sampling periods), confirming it is in an effective braking state; it extracts the vehicle speed as 38 km / h and the road speed limit as 60 km / h, indicating braking is not triggered by speeding; it extracts the relative motion information between the vehicle and the vehicle in front in the same lane, calculating the longitudinal collision time to be much greater than the threshold (TTCx > 5s), indicating braking is not triggered by a rear-end collision risk. Once all the above conditions are met, the server outputs the unexpected braking result as "triggered".

[0061] In another embodiment, the server detects that a vehicle (electric tricycle) in the adjacent lane is turning, with a lateral speed of 0.8 m / s and a lateral deviation of 0.6 meters. The server invokes the measurement timing logic rules for "unintended turning" scenarios from the defensive driving experience knowledge base. The rules include: preconditions (the vehicle in the adjacent lane is turning) and triggering conditions (turning without signal, or turning with signal but with a risk of collision with the vehicle in front). The server obtains from the perception information that the vehicle's left turn signal is active ("on"), and calculates the longitudinal collision time between the vehicle and the vehicle in front in the same lane as 2.3 seconds, which is lower than the preset risk threshold (3 seconds), thus determining that there is a collision risk. Therefore, this turning behavior belongs to the case of "turning with turn signal on and turning due to collision risk", and the server outputs the unexpected turning result as "triggered".

[0062] Step 203: Given that the processing result represents the target vehicle triggering defensive driving experience, the current scene parameters are input into the pre-trained phantom obstacle generation model according to the type characteristics of the perception-limited scene to obtain phantom obstacles.

[0063] The phantom obstacle generation model is a deep learning-based generative model trained on a large amount of historical driving scenario data. Based on input parameters of a perception-limited scene (including geometric features of the occluded area, the motion state of vehicles in adjacent lanes, lane structure information, etc.), this model outputs the position, size, and conservative motion parameters of the most likely virtual obstacle within the currently occluded area. The phantom obstacle is not a real physical entity, but rather a virtual risk representation generated to trigger defensive driving strategies.

[0064] Phantom obstacles refer to virtual obstacle objects generated within occluded areas of a perception-limited scenario, based on pre-defined rules or generated models derived from defensive driving experience. Their attributes include longitudinal position, lateral position, length, width, speed, and heading angle. They are used to replace potential risk targets not directly detected by the perception system and participate in subsequent trajectory planning and risk avoidance decisions.

[0065] The current scene parameters refer to the set of input data used to characterize the features of a scene with limited perception, including at least: geometric boundary parameters of the occluded area, size, position, heading angle, motion state parameters of the vehicle in front in the adjacent lane, lane boundary information, etc.

[0066] In this embodiment of the application, when the processing result output in step 202 indicates that the target vehicle has triggered defensive driving experience, the server inputs the current scene parameters into the pre-trained phantom obstacle generation model to generate the corresponding phantom obstacle.

[0067] For example, in step 202, the server determines the unexpected braking result as "triggered," confirming that the current scenario is of the "unexpected braking of the vehicle in front in the adjacent lane" type. The server first determines the occlusion area based on the state information: using the target vehicle's position as the viewpoint, and the right rear corner of the vehicle in front in the adjacent lane (a large passenger bus) as the starting boundary of the occlusion, combined with the right lane line, it calculates a polygonal occlusion area enclosed by five points: A, B, C, D, and E. The server extracts the geometric boundary coordinates of this occlusion area, the dimensions (12m × 2.5m), position (120, 3.5), heading angle (86.5°), speed (38km / h), acceleration (-1.8m / s²), and lane width (3.5m) of the vehicle in front in the adjacent lane, forming a scene parameter vector.

[0068] The server inputs the aforementioned scene parameters into a pre-trained phantom obstacle generation model. This model employs an encoder-decoder architecture, where the encoder maps the input parameters to latent space feature vectors, and the decoder outputs the attributes of the phantom obstacle based on these feature vectors. The phantom obstacle output by the model is located slightly to the right of the center of the occlusion region, with coordinates (125.3, 4.2), dimensions of 1.8m × 0.8m (simulating a pedestrian silhouette), and a lateral velocity of 1.2m / s (simulating a pedestrian crossing a road) and a longitudinal velocity of 0m / s. The server stores this phantom obstacle data and transmits it to the trajectory planning module.

[0069] In another embodiment, the server determines the unexpected steering result as "triggered" in step 202, confirming that the current scenario is of the "unexpected steering by the vehicle in front in the adjacent lane" type. The server extracts scenario parameters: the vehicle in front in the adjacent lane (electric tricycle) is 3.2m × 1.2m in size, located at (45, 3.2), with a heading angle of 272°, a lateral speed of 0.8m / s, and its left turn signal is on; the current steering state is a continuous leftward deviation. The server inputs the above parameters into the phantom obstacle generation model. Based on the implicit rule learned from the training data that "non-motorized vehicles easily emerge from the area obscured by the turning vehicle," the model outputs a phantom obstacle located at the front edge of the obscured area, with coordinates (46.5, 3.8), a size of 1.6m × 0.6m (simulating a bicycle), a longitudinal speed of 2.5m / s (simulating a rapid approach in the same direction), and a lateral speed of 0.3m / s. The server uses the generated phantom obstacle for subsequent trajectory planning.

[0070] Step 204: Based on the phantom obstacles, a defensive safety trajectory is generated using a preset trajectory planning and control algorithm, and the target vehicle is controlled to travel along the defensive safety trajectory.

[0071] The trajectory planning and control algorithm refers to the Artificial Potential Field-Model Predictive Control (APF-MPC) hybrid framework that integrates defensive driving experience. This framework first constructs an artificial potential field model that includes both phantom obstacles and actually perceived obstacles, quantifying collision risk through a rejection potential field. Then, it constructs a model predictive control problem based on the vehicle dynamics model, integrating the artificial potential field value as a penalty term into the cost function. Finally, it generates the optimal trajectory that satisfies dynamic constraints, safety constraints, and comfort constraints through rolling optimization.

[0072] Defensive safety trajectories refer to preventative driving trajectories generated by trajectory planning algorithms in scenarios with limited perception and triggered defensive driving experiences. These trajectories aim to avoid both phantom and actual obstacles, involving early deceleration, moderate steering, and sufficient safety margins. This trajectory is more conservative than conventional trajectories, designed to allow for response time and operational space for potential risks within obstructed areas.

[0073] In this embodiment, the server generates a defensive safety trajectory based on the phantom obstacle generated in step 203 using a preset trajectory planning and control algorithm, and sends the trajectory command to the target vehicle for execution.

[0074] For example, after receiving a phantom obstacle (coordinates (125.3, 4.2), size 1.8m × 0.8m, lateral speed 1.2m / s), the server invokes the artificial potential field-model prediction control framework. In the artificial potential field construction phase, the server calculates defensive driving potential field values ​​for both the phantom obstacle and the actually perceived vehicle in the adjacent lane. For the phantom obstacle, a high intensity coefficient K is set (simulating high alertness towards vulnerable road users), and the potential field distribution at the current moment is calculated based on parameters such as relative distance dx = 15.3m, dy = 2.4m, relative speed difference, and relative heading angle. For the vehicle in the adjacent lane, a normal intensity coefficient is set. The server then superimposes the potential fields of each obstacle to form the total artificial potential field.

[0075] In the model predictive control phase, the server constructs a predictive model based on the three-degree-of-freedom dynamics model of the target vehicle, with the prediction time domain set to 3 seconds and the control time domain set to 1.5 seconds. The cost function includes three terms: trajectory tracking term (to keep the vehicle as close to the centerline of the original lane as possible), control smoothing term (to limit the rate of change of front wheel steering angle and acceleration / deceleration), and potential field penalty term (to ensure the vehicle trajectory avoids high potential field regions). Using the current vehicle state as initial conditions, the server solves a quadratic programming problem to obtain the optimal control sequence (front wheel steering angle, longitudinal force) for the next 3 seconds. The first control variable of the sequence is applied to the vehicle to generate the defensive safety trajectory for this period: the vehicle smoothly decelerates from 45 km / h to 32 km / h within 2 seconds, while simultaneously generating a small left turn of 0.3 to avoid obstacles, ensuring that the minimum distance between the vehicle and the phantom obstacle is maintained at more than 3.5 meters when passing through the obstructed area.

[0076] The server sends the trajectory command to the target vehicle via the 5G network. The vehicle's chassis control system parses the command and executes the corresponding steering and braking operations to achieve defensive and safe driving.

[0077] In another embodiment, the server generates a standard safety trajectory in normal scenarios or in perception-limited scenarios where defensive driving experience is not triggered. The standard safety trajectory does not include conservative avoidance maneuvers for phantom obstacles; it only avoids actually perceived obstacles. For example, on the same road segment where the vehicle in the adjacent lane does not trigger unexpected braking, the server only performs follow-the-car control for the vehicle in front, generating a constant speed or slightly adjusted trajectory to maintain a safe distance, without generating additional deceleration or avoidance actions. The vehicle travels normally according to this standard trajectory.

[0078] In the aforementioned autonomous driving safety decision-making method that integrates human defensive driving experience in perception-limited scenarios, by formalizing human defensive driving experience and constructing it into a knowledge base, the target vehicle can proactively identify potential risks in perception-limited scenarios, rather than passively responding only to detected obstacles. Traditional methods suffer from a lack of decision-making information due to perception blind spots, resulting in aggressive strategies and a high risk of sudden accidents. This application, through defensive driving experience judgment processing, predicts hidden risks from the unexpected braking or steering behavior of vehicles in adjacent lanes, extending the decision-making basis from "what is seen is avoided" to "what is unseen can also be predicted," fundamentally solving the problem of information loss in perception-limited scenarios. Furthermore, through a phantom obstacle generation model, virtual obstacles simulating vulnerable road users are proactively generated within the occluded area, transforming unknown potential risks into knowable, quantifiable, and avoidable decision inputs. The trajectory planning algorithm then generates a defensive safety trajectory with sufficient safety margin, ensuring that the vehicle is already in a safe state of pre-deceleration and stable posture when facing suddenly appearing pedestrians or non-motorized vehicles, rather than a passive response of emergency braking. This technical solution improves the accuracy of autonomous driving decision-making from precise avoidance of known risks to proactive prediction of unseen risks, effectively making up for the inherent limitations of physical perception systems and significantly enhancing driving safety and system robustness in complex traffic environments.

[0079] In an exemplary embodiment, the aforementioned perception-restricted scenarios include a first type of restricted scenario formed by an unexpected braking by a vehicle in the adjacent lane, and a second type of restricted scenario formed by an unexpected steering by a vehicle in the adjacent lane; the processing results include unexpected braking results and unexpected steering results. Based on this, the aforementioned "performing defensive driving experience judgment processing on the current driving scenario of the target vehicle based on state information and a preset defensive driving experience knowledge base to obtain processing results" includes:

[0080] Scenario 1: In the case of a perception-restricted scenario of type I, the braking behavior of the vehicle in front in the adjacent lane is judged and processed unexpectedly based on state information and defensive driving experience knowledge base, resulting in an unexpected braking result.

[0081] The first type of restricted scenario refers to a perception-restricted scenario caused by an unexpected braking action by a vehicle in an adjacent lane. In this scenario, the vehicle in the adjacent lane decelerates or brakes without any obvious external cause, creating a continuous obstruction of the target vehicle's view, and its braking action itself constitutes a potential source of traffic conflict risk.

[0082] Braking behavior refers to the driving operation of actively applying braking force to reduce vehicle speed or maintain a standstill. In the defensive driving experience knowledge base, braking behavior is quantitatively characterized by parameters such as longitudinal acceleration, brake pedal status, and rate of change of vehicle speed. Effective braking state refers to the state interval in which the vehicle's longitudinal acceleration is continuously lower than a preset braking threshold for a duration exceeding the judgment window.

[0083] Unexpected behavior determination processing refers to the process of analyzing the causal relationship and inferring the intent of the braking behavior of the vehicle in the adjacent lane based on the metric time sequence logic rules in the defensive driving experience knowledge base. This processing verifies various possible triggers for braking behavior one by one, including speeding, collision risk with the vehicle in front, and sudden changes in road conditions. After eliminating all reasonable explanations, the braking behavior is determined to be "unexpected," and the corresponding trigger determination result is output.

[0084] Unexpected braking result refers to the binary judgment conclusion output after unexpected judgment processing, with a value of "triggered" or "not triggered". When the judgment result is "triggered", it indicates that the current braking behavior of the vehicle in front in the adjacent lane conforms to the "unexpected braking" characteristic pattern defined in the defensive driving experience knowledge base, and the system needs to activate the corresponding risk prediction and avoidance strategy.

[0085] A defensive driving experience knowledge base refers to a collection of knowledge that formally expresses and digitally stores the forward-looking and preventative driving experiences accumulated by human drivers over long-term driving practice, using metric temporal logic. For unexpected braking scenarios, the knowledge base stores a complete chain of reasoning rules, including preconditions, exclusion conditions, and state persistence determinations.

[0086] In this embodiment of the application, when the server identifies that the target vehicle is in a first type of restricted scenario, it performs unexpected judgment processing on the braking behavior of the vehicle in front in the adjacent lane based on state information and a defensive driving experience knowledge base, and obtains unexpected braking results.

[0087] For example, the target vehicle is a Level 4 autonomous passenger car traveling at 65 km / h in the middle lane of an urban expressway. The server, through perception information uploaded from the vehicle, identifies: a heavy-duty van is in the adjacent lane to the left, with a current speed of 58 km / h and a longitudinal acceleration of -2.1 m / s², which has persisted for 0.6 seconds; approximately 45 meters ahead of the van is a sedan in the same lane, with low relative speeds and a stable longitudinal distance; the current speed limit is 80 km / h, and the van's speed is not exceeded. Simultaneously, based on a geometric occlusion model, the van's outline and the vehicle's perspective form a continuous occlusion area, preventing the target vehicle from perceiving traffic conditions to the right and behind the van.

[0088] The server calls upon the rules for determining unexpected braking from the defensive driving experience knowledge base to process each case. First, the server confirms that there is a vehicle in the adjacent lane ahead and that the vehicle is effectively braking, thus fulfilling the precondition. Second, the server extracts the road speed limit information and the truck's current speed to confirm that the truck is not speeding; the server calculates the longitudinal collision time between the truck and the vehicle in front in the same lane, which is much greater than the preset risk threshold, confirming that the truck is not braking due to a rear-end collision risk. Both exclusion conditions are met. Third, the server continuously monitors the truck's acceleration information and has not detected any indication of the effective braking ending, confirming that the vehicle is still in a state of effective braking not having ended.

[0089] Based on the above item-by-item judgment, the server confirms that the vehicle in the adjacent lane continues to brake without speeding or posing a risk of collision, and that this braking behavior cannot be reasonably interpreted as a necessary response to an explicit risk. Therefore, the server determines that this braking behavior conforms to the "unintended braking" characteristic pattern defined in the defensive driving experience knowledge base, and outputs the unexpected braking result as "triggered".

[0090] In another embodiment, the target vehicle is traveling on a main urban road, and the vehicle in front in the adjacent lane is a city bus. The server detects that the bus's longitudinal acceleration is -2.5 m / s², and it is continuously braking. The server first determines that the preconditions are met; then it verifies the exclusion conditions: the current speed limit for this road segment is 50 km / h, the bus's current speed is 32 km / h, which is not exceeding the speed limit; the longitudinal distance between the bus and the vehicle in front in the same lane is approximately 30 meters, the relative speed is stable, and there is no risk of a rear-end collision. The server further checks whether there are other reasonable explanations, such as the bus stopping at a bus stop—but the perception information shows that there is no bus stop in the current road segment, and the bus has not activated its right turn signal or stop indicator. After eliminating all conventional braking causes, the server determines that the braking behavior is unexpected braking and outputs the unexpected braking result as "triggered".

[0091] In another embodiment, the target vehicle is traveling on a highway, and the vehicle in the adjacent lane ahead is a small passenger car. The server monitors that the vehicle's acceleration is -1.8 m / s², and it is continuously braking. The preconditions are met; the exclusion condition verification shows that the vehicle's current speed is 110 km / h, the speed limit for this section is 120 km / h, and it is not exceeding the speed limit; there are no other vehicles within 200 meters ahead in the same lane, and there is no risk of rear-end collision. However, the server further detects that there is a construction area ahead of the vehicle, with cones and warning signs clearly visible. The server determines that this braking behavior is a reasonable response to the construction area ahead and does not constitute unexpected braking, therefore the unexpected braking result is output as "not triggered".

[0092] Scenario 2: In the case of a perception-restricted scenario of type II, an unexpected judgment is made on the steering behavior of the vehicle in front in the adjacent lane based on state information and a defensive driving experience knowledge base, resulting in an unexpected steering outcome.

[0093] The second type of restricted scenario refers to a perception-restricted scenario caused by an unexpected turning action by a vehicle in an adjacent lane. In this scenario, the vehicle in the adjacent lane makes a lateral movement into the target lane or a neighboring lane, which not only dynamically obstructs the target vehicle's view, but its turning action itself also constitutes a direct or potential risk of entry.

[0094] Steering behavior refers to the driving action of a vehicle that causes a lateral deviation in its direction of travel by changing the steering angle of the front wheels. In the defensive driving experience knowledge base, steering behavior is quantitatively characterized by parameters such as lateral velocity, lateral displacement, rate of change of heading angle, and turn signal status. The initiation of a steering action is identified when the lateral velocity exceeds a preset threshold or the lateral position deviation exceeds a preset threshold.

[0095] Unexpected turning judgment processing refers to the process of determining the compliance and rationality of the turning behavior of a vehicle in an adjacent lane based on the metric-sequential logic rules in a defensive driving experience knowledge base. This processing determines whether the turning behavior is "unexpected" by comprehensively evaluating the matching degree between the turn signal usage status, the turning intention, and the surrounding traffic environment. Unexpected turning includes two typical scenarios: one is initiating a turning action without activating the turn signal; the other is that although the turn signal is activated, the turning behavior is a forced avoidance maneuver due to a collision risk with the vehicle in front in the same lane.

[0096] Unexpected steering result refers to the binary judgment conclusion output after unexpected judgment processing, with a value of "triggered" or "not triggered". When the judgment result is "triggered", it indicates that the current steering behavior of the vehicle in front in the adjacent lane conforms to the "unexpected steering" characteristic pattern defined in the defensive driving experience knowledge base, and the system needs to activate the corresponding risk prediction and avoidance strategy.

[0097] The defensive driving experience knowledge base stores a complete inference rule chain for unexpected turning scenarios, including preconditions, triggering conditions, and state persistence determination. Triggering conditions cover two types of situations: turning without signaling and turning with signaling due to risk association.

[0098] In this embodiment of the application, when the server identifies that the target vehicle is in a second type of restricted scenario, it performs unexpected judgment processing on the turning behavior of the vehicle in front in the adjacent lane based on state information and defensive driving experience knowledge base, and obtains unexpected turning results.

[0099] For example, a target vehicle is traveling at 40 km / h on a secondary urban road. An electric tricycle is in the adjacent lane on its right. The server identifies through sensor information that the tricycle's lateral speed is 0.8 m / s, its lateral position deviation is 0.6 meters, and it is continuously deviating to the left from the lane centerline; its left turn signal is active; simultaneously, approximately 8 meters ahead of the tricycle is a slowly moving sanitation vehicle. The relative speed between the tricycle and the vehicle in front of it in the same lane is significant, and the longitudinal distance is rapidly decreasing.

[0100] The server calls upon the rules for determining unintended turns in the defensive driving experience knowledge base to process each rule. First, the server confirms that there is a vehicle in the adjacent lane ahead and that vehicle is initiating a turning action, thus fulfilling the precondition. Second, the server obtains the turn signal status information and confirms that the vehicle has activated its left turn signal. The server further calculates the longitudinal collision time between the vehicle and the vehicle in front in the same lane and finds that this value is below a preset risk threshold, determining that there is a risk of rear-end collision. According to the knowledge base rules, when a vehicle activates its turn signal and turns due to a collision risk with the vehicle in front in the same lane, this turning behavior is classified as a risk-related unintended turn. The server outputs the unintended turn result as "triggered".

[0101] In another embodiment, the target vehicle is traveling on a two-lane rural road. A large SUV is in the opposite lane (in an oncoming scenario, the adjacent lane refers to the lane facing the opposite direction). The server detects that the SUV has a lateral speed of 1.2 m / s and its heading angle is continuously shifting to the left, clearly encroaching on the target vehicle's lane. The server obtains turn signal status information, showing that the vehicle has not activated any turn signals. The server determines that this behavior meets the unexpected steering characteristic of "initiating a steering action without activating the turn signal," and outputs the unexpected steering result as "triggered."

[0102] In another embodiment, the target vehicle is traveling on a highway, with a sedan in the adjacent lane to its left. The server detects that the car's lateral speed is 0.3 m / s, its lateral deviation is 0.2 meters, and its left turn signal is activated. The server calculates that the longitudinal distance between the car and the car in front in the same lane is sufficient, with no risk of collision; simultaneously, the perception information indicates that there are no obstacles, accident vehicles, or lane reduction abnormalities in front of the car. The server determines that this turning behavior is a normal lane change operation and not an unexpected turn, therefore outputting the unexpected turn result as "not triggered". After the car completes the lane change, the server continues to monitor its subsequent behavior.

[0103] In another embodiment, the target vehicle is waiting to turn left at an intersection, with a van in the adjacent lane on its right. The server detects that the van has a lateral speed of 0.5 m / s, is veering to the right, and its right turn signal is activated. The server's perception information indicates that there are no vehicles or obstacles in front of the van, posing no risk of collision; at the same time, the van's right turn direction is towards a side road entrance. The server determines that this turning behavior is a normal right turn operation and not an unexpected turn, outputting the result "Not triggered".

[0104] In one exemplary embodiment, the method further includes:

[0105] If an unexpected braking or steering result indicates the triggering of defensive driving experience, and the target vehicle meets the preset risk conditions, then the defensive driving experience of the target vehicle is determined to have been triggered.

[0106] Among them, the unexpected braking result refers to the judgment conclusion output by the server after unexpectedly judging the braking behavior of the vehicle in front in the adjacent lane, with the value being "triggered" or "not triggered". When the result is "triggered", it indicates that the current braking behavior of the vehicle in front in the adjacent lane conforms to the "unexpected braking" characteristic pattern defined in the defensive driving experience knowledge base, that is, the braking is continuously implemented without reasonable explanations such as speeding or risk of collision with the vehicle in front.

[0107] Unexpected steering result refers to the judgment conclusion output by the server after processing the unexpected steering behavior of the vehicle in front in the adjacent lane. The value is "triggered" or "not triggered". When the result is "triggered", it means that the current steering behavior of the vehicle in front in the adjacent lane conforms to the "unexpected steering" characteristic pattern defined in the defensive driving experience knowledge base. This includes two situations: initiating a steering action without turning on the turn signal, or being forced to turn due to the risk of collision with the vehicle in front in the same lane even though the turn signal is on.

[0108] Triggering defensive driving experience is the final decision signal output by the server after comprehensive judgment. It indicates that the current perception-limited scenario has met all the activation conditions of the defensive driving experience knowledge base, and the system must immediately activate risk prediction and avoidance strategies. This signal is a prerequisite for calling the phantom obstacle generation model.

[0109] Preset risk conditions refer to the set of judgment conditions used to assess whether there is a substantial collision threat between the target vehicle and potential risk sources in the current perception-limited scenario. This set of conditions includes two types of scenarios: first, there is a direct collision risk between the target vehicle and the vehicle in front in the adjacent lane; second, there is a collision risk between the target vehicle and a vulnerable road user who may exist within the area obscured by the vehicle in front in the adjacent lane. Only when at least one type of risk condition is met does the triggering of defensive driving experience have practical significance for risk avoidance.

[0110] Collision risk assessment is based on quantitative calculations using the target vehicle's state information and surrounding environment perception information. The collision risk between the target vehicle and the vehicle ahead in the adjacent lane is determined by calculating the distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration and comparing this distance with the current longitudinal distance between the two vehicles. The collision risk between the target vehicle and vulnerable road users within an obstructed area is determined by assuming the existence of a vulnerable road user moving at any position and speed within the obstructed area and assessing whether a collision is likely to occur with the target vehicle's current trajectory.

[0111] In this embodiment of the application, after obtaining unexpected braking or unexpected steering results, the server further verifies whether the target vehicle meets the preset risk conditions, and if both conditions are met, it finally determines that the target vehicle has triggered defensive driving experience.

[0112] For example, in Case 1, the server has already completed the unexpected judgment and processing of the braking behavior of the vehicle (heavy van) in the adjacent lane, and outputs the unexpected braking result as "triggered". The server then initiates the verification process of the preset risk conditions.

[0113] First, the server verifies whether there is a collision risk between the target vehicle and the vehicle in front in the adjacent lane. The server extracts the current speed (65 km / h, approximately 18.1 m / s) from the target vehicle's status information, reads the preset comfortable deceleration value of 2.5 m / s², and the preset driver reaction time of 1.2 seconds. The server calculates the first distance required for the target vehicle to decelerate to a stop at the comfortable deceleration: the sum of the reaction distance (18.1 m / s × 1.2 s = 21.7 meters) and the braking distance (18.1² / (2 × 2.5) = 65.5 meters), totaling approximately 87.2 meters. The server obtains the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane from the perception information, which is 95 meters. Comparing these, the first distance (87.2 meters) is less than the current longitudinal distance (95 meters), and the server determines that there is no direct collision risk between the target vehicle and the vehicle in front in the adjacent lane.

[0114] Subsequently, the server verifies whether there is a collision risk between the target vehicle and potentially vulnerable road users within the obscured area. Based on the outline dimensions (12m × 2.5m), position coordinates, and heading angle of the adjacent vehicle (truck) in the adjacent lane, as well as the target vehicle's position and lane boundary information, the server determines a continuous obscured area enclosed by five points: A, B, C, D, and E. Within this obscured area, the server, according to preset rules in a defensive driving experience knowledge base, assumes the existence of vulnerable road users (such as pedestrians and non-motorized vehicles) moving at arbitrary positions and speeds (0-5m / s laterally / longitudinally). Using the target vehicle's current motion state (speed 65km / h, heading angle 87.3°) as a benchmark, the server simulates and predicts the vehicle's trajectory over the next 3 seconds and verifies whether this trajectory might intersect with any assumed vulnerable road user within the obscured area in time and space. Simulation results show that near the right rear boundary of the obstructed area, there exists a hypothetical pedestrian with positional parameters (x=128.7m, y=4.3m) and velocity parameters (lateral 1.8m / s). The pedestrian's trajectory spatially overlaps with the predicted trajectory of the target vehicle after 2.3 seconds, with the minimum distance being less than the safety threshold (1.5 meters). The server determines that there is a collision risk between the target vehicle and a vulnerable road user within the obstructed area.

[0115] At this point, the server confirms that the unexpected braking result is "triggered," and the target vehicle meets the second type of pre-defined risk conditions (a collision risk with a vulnerable road user within the obstructed area). Therefore, the server ultimately determines that the target vehicle has triggered defensive driving experience and transmits this decision signal to the next stage to initiate phantom obstacle generation and defensive trajectory planning.

[0116] In one exemplary embodiment, the aforementioned preset risk conditions include any of the following conditions:

[0117] There is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane; there is a potential risk of collision between the target vehicle and a vulnerable road user within the area obscured by the vehicle in front in the adjacent lane.

[0118] The preset risk conditions refer to the set of judgment conditions used to assess whether there is a substantial collision threat between the target vehicle and potential risk sources in the current perception-limited scenario. This set of conditions is a necessary component of the determination of defensive driving experience triggering. Only when the unexpected braking result or unexpected steering result is "triggered" and at least one preset risk condition is met can the server finally determine that the target vehicle has triggered defensive driving experience.

[0119] The preset risk conditions include any of the following:

[0120] Condition 1: There is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

[0121] The risk of collision between the target vehicle and the vehicle in the adjacent lane refers to the possibility, based on the current motion of both vehicles, that the target vehicle will rear-end or collide with the vehicle in front at some point in the future if it does not take effective evasive action. This risk is determined by calculating the first distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration and comparing it to the current longitudinal distance between the two vehicles. If the first distance is greater than or equal to the current longitudinal distance, a collision risk is considered to exist; otherwise, no collision risk is considered to exist. This method fully embodies the conservative principle of defensive driving, assuming that the vehicle will actively decelerate and avoid a collision, and assesses the safety margin under the most unfavorable conditions.

[0122] Condition 2: There is a potential collision risk between the target vehicle and a vulnerable road user within the area obscured by the vehicle in front in the adjacent lane.

[0123] The occlusion area refers to the blind spot where the target vehicle's perception system cannot directly detect it due to the physical outline of the vehicle in front in the adjacent lane. This area is defined by the outline of the vehicle in front in the adjacent lane, the position of the target vehicle, and the lane boundary. Its geometric boundary can be accurately calculated through viewpoint projection and spatial geometric model.

[0124] Vulnerable road users refer to road users who are in a relatively vulnerable position in the road traffic safety system, including but not limited to pedestrians, non-motorized vehicles (bicycles, electric bicycles, tricycles), and wheelchair users. Because these road users lack physical protection, they are more susceptible to serious injury in collisions; therefore, defensive driving experience is given a higher risk weighting for them.

[0125] Potential collision risk refers to a situation within an obstructed area where, although the actual presence of a vulnerable road user cannot be directly detected by sensors, a conservative assumption based on defensive driving experience presupposes the existence of such a user at any position and speed, and that a collision is "possible" if the target vehicle continues in its current state of motion. Determining this risk does not require proving the actual existence of a vulnerable road user; instead, it employs a defensive approach of "assuming existence and proactive avoidance." This involves a comprehensive test across all position and speed spaces within the obstructed area to determine if any set of position and speed parameters would allow a collision to occur. If such a set exists, a potential collision risk is identified.

[0126] In this embodiment of the application, after the server obtains an unexpected braking result or an unexpected steering result as "triggered", it verifies whether the above two preset risk conditions are met one by one, and finally determines whether the target vehicle has triggered defensive driving experience based on the verification results.

[0127] For condition one, for example, the target vehicle is a Level 4 autonomous passenger vehicle traveling at 65 km / h on an urban expressway. In condition one, the server has already determined that the unexpected braking result of the vehicle (a heavy-duty van) in the adjacent lane was "triggered". Now, we verify the preset risk condition one.

[0128] The server extracts the current speed from the target vehicle's status information as 65 km / h (approximately 18.1 m / s). The system's preset comfortable deceleration value is 2.5 m / s², a parameter calibrated based on typical deceleration behavior of human drivers in normal traffic flow, balancing safety and ride comfort. The system's preset driver reaction time is 1.2 seconds, a parameter encompassing the entire link delay from risk recognition to braking execution. The server calculates the first distance required for the target vehicle to decelerate to a stop at a comfortable deceleration: the sum of the reaction distance (18.1 m / s × 1.2 s = 21.7 meters) and the braking distance (18.1² / (2 × 2.5) = 65.5 meters), totaling 87.2 meters.

[0129] The server extracts the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane from the perceived information, which is 95 meters. Upon comparison, the first distance (87.2 meters) is less than the current longitudinal distance (95 meters). The server determines that even if the target vehicle immediately decelerates at a comfortable speed, it can safely stop before reaching the truck's current position, and there is no risk of a rear-end collision. Therefore, the first preset risk condition is not met.

[0130] In another embodiment, the target vehicle is traveling at 40 km / h on a main urban road. An unexpected turn by a vehicle (an electric tricycle) in the adjacent lane is considered a "trigger." The server calculates the target vehicle's first distance: speed 40 km / h (approximately 11.1 m / s), comfortable deceleration 2.0 m / s², reaction time 1.0 s, the first distance is (11.1 × 1.0) + (11.1² / (2 × 2.0)) = 11.1 + 30.8 = 41.9 meters. The current longitudinal distance between the two vehicles is 18 meters, and the first distance (41.9 meters) is greater than the current longitudinal distance (18 meters). The server determines that if the tricycle suddenly decelerates or cuts in, the target vehicle cannot avoid it within a safe distance, posing a direct collision risk. The preset risk condition is met.

[0131] For condition two, for example, the target vehicle is traveling at 65 km / h, and the vehicle in front in the adjacent lane is a heavy truck that braked unexpectedly. The server has already determined that there is no direct collision risk in the verification of condition one, and now it will verify the preset risk condition two.

[0132] The server first determines the occlusion area. Based on the outline dimensions (12 meters long, 2.5 meters wide), position coordinates (120 meters longitudinally, 3.5 meters laterally), and heading angle of the vehicle (truck) in the adjacent lane, as well as the target vehicle's position (0 meters longitudinally, 1.8 meters laterally) and lane boundary (4.2 meters laterally from the right lane line), the server calculates a continuous polygonal occlusion area enclosed by five points A, B, C, D, and E through viewpoint projection and a spatial geometric model. This area covers a certain range to the right and rear of the truck, constituting a complete blind spot for the target vehicle's perception system.

[0133] The server then assesses potential collision risks. Following the pre-defined rules of the defensive driving experience knowledge base, the server traverses all possible location coordinates within the obstructed area in a grid-like manner, assigning each location any possible speed within the range of 0-5 m / s (including both lateral and longitudinal dimensions). Using the target vehicle's current motion state (speed 65 km / h, heading angle 87.3°) as a baseline, the server employs a kinematic model to predict the vehicle's trajectory over the next 5 seconds, and iterates through each set of hypothetical vulnerable road users (position + speed) to determine whether their trajectory will spatially overlap with the target vehicle's trajectory at some future moment.

[0134] After comprehensive calculations, the server detected a pedestrian moving at a lateral speed of 1.8 m / s (simulating a pedestrian crossing the road) near the right rear boundary of the obscured area (coordinates x=128.3m, y=4.1m). The relative motion between this pedestrian and the target vehicle would result in a spatial intersection after 2.1 seconds, with the minimum distance less than the system's preset safety threshold (1.5 meters). The server determined that although no pedestrian was actually detected within the obscured area, based on the conservative principle of defensive driving, this combination of location and speed parameters is physically possible and would lead to a collision. Therefore, there is a potential collision risk between the target vehicle and a vulnerable road user within the obscured area, and the second preset risk condition is met.

[0135] In one exemplary embodiment, such as Figure 3 As shown, the process for determining whether there is a collision risk between the target vehicle and the vehicle in front in the adjacent lane includes:

[0136] Step 301: Based on the state information of the target vehicle, calculate the first distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration.

[0137] The preset comfort deceleration refers to the maximum braking deceleration threshold set to ensure ride comfort. This threshold is lower than the vehicle's physical deceleration limit and is typically between 2.0 m / s² and 3.0 m / s². The specific value can be dynamically adjusted or pre-calibrated based on factors such as vehicle type, load condition, and road surface adhesion coefficient. Using comfort deceleration rather than the limit deceleration for calculation reflects the conservative principle of "leaving a safety margin and avoiding emergency operations" in defensive driving experience.

[0138] The first distance refers to the total distance the target vehicle travels from the current moment, after the driver's reaction time and the braking system delay time, as it continuously decelerates at a preset comfort deceleration until its speed reaches zero. This distance consists of two parts: the reaction distance, which is the distance the target vehicle travels at a constant speed within the reaction time; and the braking distance, which is the distance the target vehicle travels from the start of deceleration to a complete stop.

[0139] In this embodiment of the application, the server calculates the first distance required for the target vehicle to decelerate to a stop at a preset comfortable deceleration based on the target vehicle's state information.

[0140] For example, the target vehicle is a Level 4 autonomous passenger car traveling at 72 km / h on an urban expressway. The server extracts the current speed as 72 km / h (20 m / s) from the status information uploaded by the target vehicle. The system's preset comfort deceleration value is 2.5 m / s², a parameter derived from extensive passenger car comfort calibration experiments, ensuring braking safety while preventing excessive vehicle pitch that could cause occupant discomfort. The system's preset reaction time is 1.2 seconds, a parameter that covers the entire chain latency from risk identification and decision-making to actuator response.

[0141] The server calculates the reaction distance as: 20 m / s × 1.2 s = 24 meters. The server calculates the braking distance as: 20² / (2 × 2.5) = 400 / 5 = 80 meters. The server adds the reaction distance and the braking distance, obtaining the first distance as 24 meters + 80 meters = 104 meters.

[0142] In another embodiment, the target vehicle is an unmanned delivery vehicle traveling at 18 km / h within a closed park. The server extracts the current speed as 18 km / h (5 m / s). For low-speed, lightly loaded vehicles, the system's preset comfortable deceleration value is 1.5 m / s², and the reaction time is 0.8 seconds. The server calculates the first distance: reaction distance 5 m / s × 0.8 s = 4 meters, braking distance 5² / (2 × 1.5) = 25 / 3 ≈ 8.33 meters, for a total first distance of approximately 12.33 meters.

[0143] In another embodiment, the target vehicle is a fully loaded heavy truck traveling at 80 km / h on a highway. The server extracts the current speed as 80 km / h (22.2 m / s). Considering the vehicle's load and braking performance, the system dynamically adjusts the comfort deceleration value to 1.8 m / s², and the reaction time is preset to 1.5 seconds (considering the response delay of the heavy vehicle's braking system). The server calculates the first distance: reaction distance 22.2 m / s × 1.5 s = 33.3 meters, braking distance 22.2² / (2 × 1.8) = 492.84 / 3.6 ≈ 136.9 meters, and the total first distance is approximately 170.2 meters.

[0144] Step 302: Based on the first distance and the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane, determine whether there is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

[0145] The current longitudinal distance refers to the projected distance between the target vehicle and the vehicle in front in the adjacent lane along the road direction in the same lane coordinate system. This distance is obtained in real time through onboard perception systems (cameras, lidar, millimeter-wave radar) and is a core parameter for assessing the relative positional relationship between the two vehicles.

[0146] Collision risk refers to the possibility, based on the current motion of the two vehicles, that the target vehicle will rear-end or side-impact the vehicle in the adjacent lane at some point in the future if it does not take effective evasive action. This step uses "active deceleration and avoidance capability" as the judgment criterion: if the first distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration rate is less than the current longitudinal distance, it is determined that the target vehicle has sufficient space to stop safely before reaching the position of the vehicle in front, and there is no collision risk; if the first distance is greater than or equal to the current longitudinal distance, it is determined that the target vehicle cannot complete deceleration and avoidance within a safe distance, and there is a collision risk.

[0147] This judgment method embodies the core idea of ​​defensive driving experience: not only assessing whether a collision will occur at the current moment, but also assessing whether the vehicle has sufficient safety margin to deal with unexpected situations.

[0148] In this embodiment of the application, the server determines whether there is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane based on the first distance calculated in step 301 and the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane.

[0149] For example, continuing from the embodiment of the passenger car traveling on the expressway in step 301. The server has calculated the first distance to be 104 meters. The server extracts the current longitudinal distance between the target vehicle and the vehicle in front (heavy van) in the adjacent lane from the perception information, which is 110 meters. The server compares the first distance with the current longitudinal distance: 104 meters < 110 meters. The server determines that if the target vehicle immediately decelerates at a comfortable speed, it can come to a complete stop before reaching the van's current position, and there is no risk of a rear-end collision. Therefore, the server determines that there is no risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

[0150] In another embodiment, the target vehicle is traveling at 40 km / h on a main urban road, and the vehicle in front in the adjacent lane is an electric tricycle. The server calculates the first distance as 41.9 meters (reaction distance 11.1 meters + braking distance 30.8 meters). The server extracts the current longitudinal distance from the perception information as 18 meters. The server compares: 41.9 meters > 18 meters. The server determines that even if the target vehicle immediately decelerates at a comfortable speed, it cannot come to a complete stop before reaching the tricycle's position, posing a risk of rear-end collision or cutting-in collision. Therefore, the server determines that there is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

[0151] In another embodiment, the target vehicle is traveling at 100 km / h (27.8 m / s) on a highway, and the vehicle in front in the adjacent lane is a small passenger car. The server calculates the first distance: comfortable deceleration 2.5 m / s², reaction time 1.2 s, reaction distance 27.8 × 1.2 = 33.36 meters, braking distance 27.8² / (2 × 2.5) = 772.84 / 5 = 154.57 meters, totaling 187.93 meters. The current longitudinal distance is 200 meters. The server compares: 187.93 meters < 200 meters, therefore determining there is no risk of collision.

[0152] In another embodiment, the target vehicle is traveling at 60 km / h (16.7 m / s). The vehicle in the adjacent lane slows down due to congestion ahead, and the current longitudinal distance is only 35 meters. The server calculates the first distance: reaction distance 16.7 × 1.2 = 20.04 meters, braking distance 16.7² / (2 × 2.5) = 278.89 / 5 = 55.78 meters, for a total first distance of 75.82 meters. The server compares this distance: 75.82 meters > 35 meters, therefore determining a collision risk.

[0153] In another embodiment, the target vehicle is an autonomous sanitation vehicle operating at 25 km / h (6.94 m / s) in the non-motorized vehicle lane, with an illegally parked vehicle in the adjacent lane ahead. The server uses more conservative parameters for this low-speed operation scenario: a comfortable deceleration of 1.2 m / s² and a reaction time of 1.0 s. The first distance calculation is as follows: reaction distance 6.94 × 1.0 = 6.94 meters, braking distance 6.94² / (2 × 1.2) = 48.16 / 2.4 = 20.07 meters, for a total first distance of 27.01 meters. The current longitudinal distance is 22 meters. The server compares the distance: 27.01 meters > 22 meters, therefore determining a collision risk.

[0154] In another embodiment, after the server completes the collision risk assessment, it uses the assessment result as the verification conclusion of preset risk condition one and inputs it into the comprehensive decision-making process triggered by defensive driving experience. If a collision risk is determined to exist, preset risk condition one is valid; if no collision risk is determined to exist, preset risk condition one is invalid, and the server needs to continue to verify preset risk condition two (whether there is a potential collision risk between the target vehicle and vulnerable road users in the obstructed area).

[0155] In one exemplary embodiment, such as Figure 4 As shown, the process for determining whether there is a potential collision risk between the target vehicle and vulnerable road users within the obstructed area includes:

[0156] Step 401: Determine the occlusion area based on the state information, which is jointly defined by the outline of the vehicle in front in the adjacent lane, the position of the target vehicle, and the lane boundary.

[0157] The occlusion area refers to the blind spot where the target vehicle's perception system cannot directly detect it due to the physical outline of the vehicle ahead in the adjacent lane. This area is the core characteristic of perception-limited scenarios and also the breeding ground for potential risks.

[0158] The outline of the vehicle in front in the adjacent lane refers to the three-dimensional geometric shape of the vehicle in front in the adjacent lane. In this method, parameters such as its length, width, and heading angle are mainly used. These parameters are obtained in real time through vehicle-mounted perception systems (LiDAR, cameras, millimeter-wave radar) by detection and fusion recognition, and are used to accurately describe the spatial occupancy of obstructions.

[0159] The target vehicle position refers to the spatial coordinates of the target vehicle in the current lane coordinate system, including its longitudinal and lateral positions. This position information is obtained by matching the Global Positioning System (GPS) with a high-precision map and recursively applying it to the vehicle's kinematics model. It serves as the benchmark for calculating the viewpoint position and line of sight direction.

[0160] Lane boundaries refer to the physical boundaries of the current lane and adjacent lanes, including lane lines, curbs, guardrails, and other traffic markings and facilities. Lane boundary information is obtained through a combination of high-precision map preloading and real-time visual perception detection, and is used to constrain the lateral extension range of occluded areas.

[0161] The geometric definition of the occlusion area adopts a viewpoint projection method: taking the typical sensor position of the target vehicle's perception system as the viewpoint, and the far edge of the outline of the vehicle in front in the adjacent lane as the starting boundary of the occlusion, the projection is made along the line of sight to the lane boundary, forming a continuous polygonal region enclosed by several feature points. The spatial position within this region cannot be directly observed by the target vehicle through its onboard sensors.

[0162] In this embodiment, the server determines the occlusion area defined by the outline of the vehicle in front in the adjacent lane, the position of the target vehicle, and the lane boundary based on the state information of the target vehicle.

[0163] For example, the target vehicle is a Level 4 autonomous passenger car traveling at 65 km / h in the middle lane of an urban expressway. The server identifies, through perception information, a heavy-duty van in the adjacent lane to the left, measuring 12 meters in length, 2.5 meters in width, with a heading angle of 86.5 degrees. Its current longitudinal position is 120 meters and its lateral position is 3.5 meters. The target vehicle itself has a longitudinal position of 0 meters and a lateral position of 1.8 meters. The current lane is a six-lane, two-way urban expressway, with each lane being 3.5 meters wide, and the right lane line has a lateral position of 4.2 meters.

[0164] The server uses the installation location of the target vehicle's forward-looking perception sensor as the viewpoint, and the right rear corner of the truck as the starting boundary of the occlusion. It projects along the line of sight to the right lane line and calculates the coordinates of five key feature points in the occluded area:

[0165] Point A is the rear right corner of the truck, with coordinates (120, 3.5); Point B is the front right corner of the truck, with coordinates (132, 3.5); Point C is the intersection of the projection of the right lane line and the front right corner of the truck along the line of sight, with coordinates (132, 4.2); Point D is the intersection of the projection of the right lane line and the rear right corner of the truck along the line of sight, with coordinates (120, 4.2); Point E is the intersection of the line connecting the rear right corner of the truck and the target vehicle's viewpoint, extended to the right lane line, with coordinates calculated geometrically as (118.5, 4.2).

[0166] The server connects the five points mentioned above in sequence to form a closed polygonal region, which is the occlusion area in the current perception-limited scene. This region is located on the right side and behind the truck, covering the continuous space behind and to the right of the adjacent lane's blind spot.

[0167] In another embodiment, the target vehicle is traveling at 40 km / h on an urban main road. An electric tricycle is positioned in the adjacent lane on the right as the vehicle ahead in that lane. The tricycle is 3.2 meters long and 1.2 meters wide, with a heading angle of 272 degrees. Its current longitudinal position is 45 meters and its lateral position is 3.2 meters. The target vehicle is at 0 meters longitudinally and 1.8 meters laterally. The lane width is 3.5 meters, the left lane line is at 0 meters laterally, and the right lane line is at 3.5 meters laterally.

[0168] The server uses the target vehicle's viewpoint as a reference and the left rear corner of the tricycle as the initial boundary for occlusion, projecting it onto the left lane line to calculate the coordinates of feature points in the occluded area. Since the tricycle is making a left turn, its heading angle has changed significantly, so the server needs to consider the vehicle's attitude on the occlusion boundary when calculating the contour projection. The final determined occlusion area is an irregular quadrilateral, covering the blind spot area to the left and rear of the tricycle.

[0169] In another embodiment, the target vehicle is traveling on a two-lane rural road, with a large SUV in the opposite lane as the vehicle ahead. Because it is an oncoming traffic scenario, the definition of the occlusion area is adjusted: the server projects the occlusion area onto the opposite lane boundary and shoulder, using the target vehicle's viewpoint as a reference and the near-end outline edge of the SUV as the starting boundary of the occlusion, forming the occlusion area located in the opposite lane. This area also constitutes a blind spot for the target vehicle.

[0170] Step 402: Within the obscured area, determine whether there is a potential collision on the current trajectory of the target vehicle based on the preset speed range, location distribution, and status information of vulnerable road users.

[0171] Among them, the pre-defined vulnerable road users refer to various vulnerable road users who are predefined in the defensive driving experience knowledge base and may appear in the obstructed area, including but not limited to pedestrians, cyclists, electric bicycle riders, tricycle riders, and wheelchair users. The common characteristics of these participants are lack of physical protection, flexible and changeable movement trajectories, and high perception difficulty.

[0172] The speed range refers to the predefined range of possible movement speeds for vulnerable road users. The defensive driving experience knowledge base sets typical speed ranges for different user types: pedestrians typically experience 0-2 m / s, bicycles 2-5 m / s, e-bikes 3-8 m / s, wheelchairs 0-1.5 m / s, etc. In potential collision detection, the server employs a strategy that covers all possible speed values ​​within this range to ensure the completeness of the judgment.

[0173] Location distribution refers to the set of spatial locations where vulnerable traffic participants may appear within an obstructed area. Defensive driving experience follows the "worst-case scenario principle," making no optimistic assumptions about the possible locations of participants, but rather assuming that participants may exist at any location within the obstructed area. In its implementation, the server discretizes the obstructed area into a uniform grid, with each grid point representing a potential participant's possible location.

[0174] The target vehicle's current trajectory refers to the spatiotemporal path predicted over a future period of time based on the target vehicle's current motion state (speed, acceleration, heading angle, yaw rate, etc.) and preset driving intentions (lane keeping, following, lane changing, etc.) through kinematic or dynamic models. The trajectory prediction time domain is typically set to 3-5 seconds, covering the time window required for typical avoidance maneuvers.

[0175] A potential collision refers to a situation in the prediction time domain where there exists at least one combination of position and velocity parameters of a vulnerable road user, such that the predicted trajectories of these users and the target vehicle spatially overlap at a certain moment, and the relative distance at the point of overlap is less than a preset safety threshold. Determining a potential collision does not require proving the actual existence of the users; instead, it employs a defensive approach of "assuming existence and proactively avoiding," aiming to identify any potentially hazardous scenarios.

[0176] In this embodiment of the application, within the occlusion area determined in step 401, the server determines whether there is a potential collision on the current driving trajectory of the target vehicle based on the preset speed range, location distribution of vulnerable traffic participants and the target vehicle status information.

[0177] For example, consider the embodiment of the truck obstruction scenario in step 401. The server has determined the obstruction area formed by the heavy-duty van, which is a polygon enclosed by five points A, B, C, D, and E, covering a continuous space of approximately 8 square meters on the right side and rear of the van.

[0178] The server first loads the preset speed ranges of vulnerable road users from the defensive driving experience knowledge base. For urban expressway scenarios, the knowledge base presets typical users as pedestrians (0~2m / s) and electric bicycles (3~6m / s), with the speed coverage range taken as the union of 0~6m / s, including both lateral and longitudinal motion dimensions.

[0179] The server then performs a grid-based discretization process on the obstructed area. With a resolution of 0.5 meters, the polygonal obstructed area is divided into approximately 32 grid points, and the coordinates of each grid point represent the possible location of a vulnerable road user.

[0180] Based on the target vehicle's current state information (speed 65km / h, heading angle 87.3, acceleration 0.2m / s²) and driving intention (keeping in lane), the server uses a uniform kinematics model to predict the vehicle's trajectory within the next 4 seconds. The sampling interval is 0.1 seconds, generating 40 discrete trajectory points. Each trajectory point contains a timestamp, longitudinal coordinate, and lateral coordinate.

[0181] The server initiates a dual traversal judgment: the outer layer traverses all grid positions within the occluded area, and the inner layer traverses all possible combinations of lateral and longitudinal velocities within the range of 0~6m / s (velocity step size 0.2m / s). For each set of parameters (position, lateral velocity, longitudinal velocity), the server simulates the movement trajectory of vulnerable traffic participants and checks at each time step whether their Euclidean distance from the target vehicle's trajectory point is less than the safety threshold (1.5 meters).

[0182] After traversing the calculations, the server found that at grid point (128.3, 4.1), if there is a vulnerable road user (simulating a pedestrian crossing the road) moving with a lateral velocity of 1.8 m / s and a longitudinal velocity of 0.2 m / s, the trajectory of this user and the predicted trajectory of the target vehicle will spatially intersect at coordinates (129.1, 3.9) after 2.3 seconds, with a minimum distance of 1.2 meters, which is less than the safety threshold of 1.5 meters. The server determines that this combination of parameters is physically possible and would constitute a collision risk if it occurs. Therefore, there is a potential collision between the target vehicle's current trajectory and the vulnerable road user within the occluded area.

[0183] In another embodiment, the target vehicle is traveling at 40 km / h, and the vehicle in front in the adjacent lane is an electric tricycle making an unexpected turn. After determining the occlusion area, the server loads the preset speed ranges of vulnerable road users. For the urban main road scenario, the knowledge base presets participants including pedestrians (0~2 m / s), bicycles (2~5 m / s), and electric bicycles (3~6 m / s), with the speed range taken as the union of 0~6 m / s.

[0184] The server divides the obstructed area into a grid, generating approximately 18 grid points. The server predicts the target vehicle's trajectory for the next 3 seconds. After double-passing and evaluation, for all combinations of position and speed, the minimum distance between the vulnerable road user and the target vehicle's trajectory is greater than the safety threshold of 2.0 meters (a more lenient safety threshold is used in urban scenarios). The server determines that there is no potential collision.

[0185] In another embodiment, the target vehicle is traveling at 30 km / h on a residential road, and the vehicle in front in the adjacent lane is an illegally parked van. After determining the obstruction area, the server loads a preset speed range for vulnerable road users. For the residential scenario, the knowledge base focuses on children and pedestrians, whose speed range is extended to 0~3 m / s, and whose movement direction is highly random.

[0186] The server divides the obstructed area into a high-density grid (0.3 meters resolution) and traverses it using a finer velocity step (0.1 m / s). It is determined that at the grid point (22.5, 2.8) near the front of the van, if a child suddenly darts out with a lateral velocity of 1.2 m / s and a longitudinal velocity of 0.5 m / s, the minimum distance between the child and the target vehicle's trajectory after 1.1 seconds is only 0.8 meters, less than the safety threshold of 1.2 meters. The server determines that a potential collision exists.

[0187] In another embodiment, the target vehicle is traveling at 100 km / h on a highway, and the vehicle in front in the adjacent lane is a container truck traveling normally. After determining the occlusion area, the server loads a preset speed range for vulnerable road users. In the highway scenario, the knowledge base assumes that pedestrians or non-motorized vehicles should not appear within the occlusion area, therefore the preset speed range is an empty set. The server directly determines that there is no potential collision.

[0188] In another embodiment, after the server completes the potential collision determination, it uses the determination result as the verification conclusion of preset risk condition two. If a potential collision is determined to exist, preset risk condition two is valid; if no potential collision is determined to exist, preset risk condition two is invalid. This conclusion is input into the comprehensive decision-making process for triggering defensive driving experience, and together with unexpected braking / steering results, it determines whether defensive driving experience is ultimately triggered.

[0189] In one exemplary embodiment, the method further includes:

[0190] The system obtains acceleration information of the vehicle in the adjacent lane from the status information and uses this information to determine whether the braking behavior is in an effective braking state. It also obtains vehicle speed information and road speed limit information from the status information and uses this information to determine whether the braking behavior is triggered by speeding. Furthermore, it obtains relative motion information between the vehicle in the adjacent lane and the vehicle in the same lane and uses this relative motion information to determine whether the braking behavior is triggered by a collision risk with the vehicle in the same lane. If the vehicle in the adjacent lane is in an effective braking state, and the braking behavior is neither triggered by speeding nor by a collision risk with the vehicle in the same lane, the unexpected braking result is determined to trigger defensive driving experience.

[0191] In this context, the vehicle in the adjacent lane refers to a traffic participant that is in the adjacent lane to the target vehicle and is located within a certain distance in front of the target vehicle. This vehicle, due to its physical outline, obstructs the target vehicle's view and is a core component of perception-limited scenarios.

[0192] Acceleration information refers to the rate of change of velocity of the vehicle ahead in the adjacent lane in the longitudinal direction, measured in meters per second², and is obtained through Doppler effect measurement by onboard millimeter-wave radar or inter-frame difference calculation by visual tracking algorithms. Acceleration information is a key indicator for identifying vehicle braking behavior.

[0193] The "effective braking not yet terminated" state refers to the interval in which the vehicle in the adjacent lane continues to apply braking force, the longitudinal acceleration remains below the preset braking threshold, and has not yet returned to the normal driving acceleration range. This state must simultaneously meet the amplitude condition (acceleration below -1.5 m / s²) and the duration condition (lasting for more than 3 sampling periods, i.e., 0.3 seconds). The "effective braking not yet terminated" state is a prerequisite for determining "unintended braking."

[0194] Vehicle speed information refers to the instantaneous speed of the vehicle ahead in the adjacent lane at the current moment, in kilometers per hour or meters per second, obtained through radar speed measurement or visual tracking. Road speed limit information refers to the maximum permitted speed for the current road segment, obtained in real time through high-precision map preloading or visual recognition of traffic signs. Overspeed-triggered braking refers to compliant driving behavior where a vehicle actively decelerates due to exceeding the road speed limit. This type of braking is reasonable and predictable and should not be considered unintended braking.

[0195] Relative motion information refers to the motion relationship data between a vehicle in an adjacent lane and a vehicle in the same lane, including parameters such as longitudinal distance, relative speed, and collision time. A vehicle in the same lane refers to a vehicle in the same lane as the vehicle in the adjacent lane and located in front of it. Collision risk refers to the probability, based on the current motion of the two vehicles, that a rear-end collision will occur between the vehicle in the adjacent lane and the vehicle in the same lane within a certain timeframe if the vehicle in the adjacent lane does not brake. The quantifiable indicator of this risk is the longitudinal collision time; when the collision time is less than a preset threshold (usually 3 seconds), a collision risk is considered to exist. Braking due to a collision risk is a reasonable avoidance maneuver and should not be considered unintended braking.

[0196] Unexpected braking result refers to the binary conclusion output after the above series of judgments, with a value of "triggered defensive driving experience" or "not triggered defensive driving experience". When the vehicle in the adjacent lane is in an effective braking state that has not ended, and its braking behavior is neither triggered by speeding nor by the risk of collision with the vehicle in front, the braking behavior cannot be reasonably interpreted as a necessary response to explicit risk, which meets the "unexpected braking" characteristic pattern defined in the defensive driving experience knowledge base. Therefore, the judgment result is that defensive driving experience has been triggered.

[0197] In this embodiment of the application, the server sequentially obtains the acceleration information, speed information and road speed limit information of the vehicle in front in the adjacent lane and the relative motion information with the vehicle in front in the same lane from the status information, and performs the determination and processing of unexpected braking behavior based on the above information.

[0198] For example, the target vehicle is a Level 4 autonomous passenger car traveling at 65 km / h in the middle lane of an urban expressway. The server identifies a heavy-duty van in the adjacent lane to the left based on perception information uploaded by the vehicle.

[0199] The server first extracts the truck's acceleration information from the status information. The longitudinal acceleration values ​​for three consecutive sampling periods (0.3 seconds) are -2.1 m / s², -2.0 m / s², and -2.1 m / s², respectively, all lower than the system's preset effective braking threshold of -1.5 m / s². The server determines that the truck is in a state where effective braking has not ended.

[0200] The server then extracted the truck's speed information from the status information, finding it to be 58 km / h. Simultaneously, the server retrieved the speed limit information for the current road segment from the high-precision map, indicating it was an urban expressway with a speed limit of 80 km / h. The truck's current speed of 58 km / h was far below the speed limit, and there was no record of speeding. The server determined that the braking action was not triggered by speeding.

[0201] The server then extracts the relative motion information between the truck and the vehicle in front in the same lane from the status information. The perception data shows that a small car is located approximately 45 meters ahead of the truck in the same lane, with a relative speed of approximately 3 km / h. The calculated longitudinal collision time is 45 meters / (3 km / h) ≈ 54 seconds, far exceeding the system's preset risk threshold of 3 seconds. The server determines that the braking action was not triggered by a collision risk with the vehicle in front in the same lane.

[0202] At this point, the server confirms that the vehicle in the adjacent lane is still under effective braking, and that the braking action was not triggered by speeding or by a risk of collision with the vehicle in front in the same lane. With all three conditions met, the server determines the unexpected braking result as "triggered defensive driving experience."

[0203] In another embodiment, the target vehicle is traveling on a main urban road, and the vehicle in the adjacent lane ahead is a city bus. The server detects that the bus's longitudinal acceleration is -2.5 m / s², lasting for 0.5 seconds, and determines that it is in a state of effective braking that has not ended. The server extracts the bus's current speed as 32 km / h, the speed limit for the section is 50 km / h, and the braking is not triggered by speeding. The server further extracts the relative motion information between the bus and the vehicle in front in the same lane: there is a taxi 30 meters ahead, the relative speed between the two vehicles is about 2 km / h, the collision time is 30 meters / (2 km / h) = 54 seconds, and there is no risk of rear-end collision. All three conditions are met, and the server determines that the unexpected braking result is "triggered defensive driving experience".

[0204] In another embodiment, the target vehicle is traveling on a highway, and the vehicle in the adjacent lane ahead is a small passenger car. The server detects that the vehicle's acceleration is -1.8 m / s², lasting for 0.4 seconds, and determines that it is in a state of effective braking that has not ended. The server extracts the vehicle's current speed as 110 km / h, and the speed limit for this section of the road is 120 km / h, indicating that the braking was not triggered by speeding. However, the server further detects that there are no other vehicles within 200 meters ahead of the vehicle, indicating that the braking was clearly not triggered by collision risk. All three conditions are met, and the server initially determines that the unexpected braking result is "triggered defensive driving experience".

[0205] However, the server continues to execute the verification process based on preset risk conditions. Calculations show that the longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane is sufficient, with no direct collision risk; the occulted area also shows no potential collision risk after thorough testing. Therefore, although the unexpected braking result is "triggered defensive driving experience," the target vehicle does not meet the preset risk conditions, and ultimately, the triggering of defensive driving experience is uncertain. This embodiment illustrates that the triggering of an unexpected braking result is a necessary condition, but not a sufficient condition, for triggering defensive driving experience.

[0206] In another embodiment, the target vehicle is traveling on a mountain road, with a medium-sized truck in the adjacent lane ahead. The server detects that the vehicle's acceleration is -2.2 m / s², determining that it is in a state of effective braking that has not yet ended. The server extracts the vehicle's current speed as 45 km / h, but visual recognition reveals a temporary speed limit sign 200 meters ahead, with a speed limit of 40 km / h. The server determines that the braking behavior is triggered by speeding (45 km / h > 40 km / h), which is a reasonable and compliant deceleration. Therefore, although the effective braking state is established, the exclusion condition is not met, and the server determines the unexpected braking result as "defensive driving experience not triggered".

[0207] In another embodiment, the target vehicle is traveling near an intersection, with a van in the adjacent lane ahead. The server detects that the van's acceleration is -2.3 m / s², determining that it is in a state of effective braking that has not yet ended. The server extracts the vehicle's current speed as 35 km / h, the speed limit for the section is 60 km / h, and the braking is not triggered by speeding. However, the server calculates that the longitudinal collision time between the vehicle and the van in the same lane is only 2.1 seconds, which is lower than the risk threshold of 3 seconds. The server determines that the braking behavior is an avoidance maneuver taken due to the risk of collision with the vehicle in front, and is reasonable and predictable. Therefore, although the effective braking state is established, the exclusion condition is not met, and the server determines the unexpected braking result as "defensive driving experience not triggered".

[0208] In another embodiment, after the server determines the unexpected braking result, it outputs the result to the comprehensive decision-making module that triggers defensive driving experience. If the result is "defensive driving experience triggered", the server will continue to verify the preset risk conditions; if the result is "defensive driving experience not triggered", the current defensive driving experience judgment process will be terminated directly, and the target vehicle will maintain its normal driving strategy.

[0209] In one exemplary embodiment, the method further includes:

[0210] The system obtains lateral movement information of the vehicle in front in the adjacent lane from the status information and uses this information to determine whether to initiate a steering action. It also obtains the turn signal status information of the vehicle in front in the adjacent lane from the status information. Furthermore, it obtains the relative movement information between the vehicle in front in the adjacent lane and the vehicle in front in the same lane from the status information and uses this relative movement information to determine whether there is a risk of collision. If the vehicle in front in the adjacent lane initiates a steering action without activating its turn signal, or if the vehicle in front in the adjacent lane initiates a steering action with its turn signal activated and turns due to a risk of collision with the vehicle in front in the same lane, the system determines that the unexpected steering result triggers defensive driving experience.

[0211] In this context, the vehicle in the adjacent lane refers to a traffic participant that is in the adjacent lane to the target vehicle and is located within a certain distance in front of the target vehicle. This vehicle obstructs the target vehicle's view due to its physical outline, and the obstruction area changes dynamically during turning, making it a core component of perception-limited scenarios.

[0212] Lateral motion information refers to the motion data of the vehicle in the adjacent lane in the direction perpendicular to the lane, including parameters such as lateral velocity, lateral displacement, and rate of change of heading angle. Lateral velocity is obtained by calculating the lateral coordinate change of the vehicle's center point through inter-frame difference calculation using a visual tracking algorithm, and is measured in meters per second. Lateral displacement refers to the lateral offset of the vehicle relative to the lane centerline, measured in meters. The rate of change of heading angle reflects the degree of vehicle steering. This information is a key indicator for identifying vehicle steering actions.

[0213] Steering is a driving maneuver that causes a significant lateral deviation in the vehicle's direction of travel by changing the steering angle of the front wheels. In the defensive driving experience knowledge base, a steering maneuver is initiated based on any of the following conditions: lateral velocity exceeds a preset threshold (typically 0.3 m / s), lateral displacement exceeds a preset threshold (typically 0.5 m), or the rate of change of heading angle exceeds a preset threshold (typically 3 / s). The recognition of steering maneuvers is independent of turn signal status and relies solely on vehicle kinematic parameters.

[0214] Turn signal status information refers to the activation status of the turn indicator lights in the external lighting signal system of the vehicle ahead in the adjacent lane, including the left turn signal, right turn signal, and off status. This information is obtained by identifying the optical characteristics of the vehicle's front and rear lights through a visual perception system (camera). Turn signals are legally mandated signal devices for drivers to convey their turning intentions to other road users. Failure to use turn signals as required constitutes a traffic violation and increases the difficulty for surrounding vehicles to predict the driver's intentions.

[0215] Relative motion information refers to the motion relationship data between the vehicle in the adjacent lane and the vehicle in the same lane, including parameters such as the longitudinal distance between the two vehicles, relative speed, and collision time. The vehicle in the same lane refers to the vehicle in the same lane as the vehicle in the adjacent lane and located in front of it. The collision risk assessment method is the same as that for braking scenarios: when the longitudinal collision time is less than a preset threshold (usually 3 seconds), a collision risk is determined to exist.

[0216] Unexpected steering result refers to the binary conclusion output after the above series of judgments, with a value of "defensive driving experience triggered" or "defensive driving experience not triggered". Unexpected steering includes two typical scenarios:

[0217] Scenario 1: The vehicle in front in the adjacent lane initiates a turning motion but does not activate its turn signal. In this situation, the vehicle's intention is not effectively communicated to surrounding road users, making it unpredictable and constituting an unintended turn.

[0218] Scenario 2: The vehicle in front in the adjacent lane initiates a turning maneuver, even with its turn signal on. However, this turning action is a forced evasive maneuver due to the risk of a collision with the vehicle in front in the same lane. In this scenario, the turning action is essentially a passive response to the risk of a rear-end collision, rather than a normal expression of lane changing or turning intentions, and thus constitutes an unintended turn.

[0219] If any of the above conditions are met, the server will determine that the unexpected steering result is "triggered defensive driving experience".

[0220] In this embodiment of the application, the server sequentially obtains the lateral movement information of the vehicle in front in the adjacent lane, the turn signal status information, and the relative movement information with the vehicle in front in the same lane from the status information, and performs the determination and processing of unexpected turning behavior based on the above information.

[0221] For example, the target vehicle is traveling at 40 km / h on a secondary urban road, and there is an electric tricycle in the adjacent lane on the right as the vehicle in front of it.

[0222] The server first extracts the tricycle's lateral movement information from the status information. The lateral velocity values ​​for three consecutive sampling periods (0.3 seconds) are 0.6 m / s, 0.8 m / s, and 0.9 m / s, respectively, all exceeding the system's preset steering action threshold of 0.3 m / s; the cumulative lateral displacement has reached 0.7 meters, exceeding the 0.5-meter threshold. The server determines that the tricycle has initiated a steering action, heading left (encroaching on the target vehicle's lane).

[0223] The server then extracts the turn signal status information of the tricycle from the status information. The perception system detects the front and rear lights of the vehicle and identifies that the left turn signal and the right turn signal are not activated. The server determines that the vehicle did not activate any turn signals when initiating the turning action.

[0224] The server confirms that the current situation meets the unexpected steering judgment rule of "initiating a steering action without activating the turn signal", and no further verification of collision risk is required. The server directly determines the unexpected steering result as "triggering defensive driving experience".

[0225] In another embodiment, the target vehicle is traveling on an urban expressway, with an SUV in the adjacent lane to its left. The server detects that the SUV has a lateral speed of 0.5 m / s and a lateral displacement of 0.4 meters (continuously increasing), and determines that it has initiated a left turn. The server extracts the turn signal status information and identifies that the left turn signal is active (blinking periodically). The server further extracts the relative motion information between the vehicle and the vehicle in front in the same lane: there is a slow-moving vehicle 20 meters ahead, the relative speed between the two vehicles is about 8 km / h, and the longitudinal collision time is calculated to be 20 m / (8 km / h) = 20 / 2.22 ≈ 9 seconds. This value is greater than the risk threshold of 3 seconds, and it is determined that there is no collision risk.

[0226] The server confirms the current situation as "initiating a steering action and activating the turn signal, but with no risk of collision." This situation constitutes a normal lane change and does not meet any of the criteria for unexpected steering. The server determines the unexpected steering result as "defensive driving experience not triggered."

[0227] In another embodiment, the target vehicle is traveling on a highway, and the vehicle in front of it in the adjacent lane on the left is a container truck. The server detects that the vehicle's lateral speed is 0.4 m / s and its lateral displacement is 0.3 meters, and determines that it has initiated a left turn. The server extracts the turn signal status information and recognizes that the left turn signal is on. The server further extracts the relative motion information between the vehicle and the vehicle in front in the same lane: there is a heavy truck 15 meters ahead, the relative speed between the two vehicles is about 5 km / h, and the longitudinal collision time is calculated to be 15 m / (5 km / h) = 15 / 1.39 ≈ 10.8 seconds, which is greater than the risk threshold of 3 seconds, and it is determined that there is no collision risk.

[0228] The server determines this situation as a normal lane change, and the unexpected steering result is "defensive driving experience not triggered".

[0229] In another embodiment, the target vehicle is traveling on a main urban road, with a van in the adjacent lane on its right. The server detects that the van has a lateral speed of 0.7 m / s and a lateral displacement of 0.6 meters, and determines that it has initiated a right turn. The server extracts the turn signal status information and recognizes that the right turn signal is on. The server further extracts the relative motion information between the vehicle and the vehicle in front in the same lane: there is a sanitation vehicle 8 meters ahead, the relative speed between the two vehicles is about 6 km / h, and the longitudinal collision time is calculated to be 8 m / (6 km / h) = 8 / 1.67 ≈ 4.8 seconds, which is still greater than the risk threshold of 3 seconds, so it is determined that there is no risk of collision.

[0230] The server determined this to be a normal lane change, and the unexpected steering result was "defensive driving experience not triggered." The target vehicle maintained normal following, and the van entered the right lane after completing the lane change.

[0231] In another embodiment, the target vehicle is traveling near an intersection, with a sedan in the adjacent lane on its left. The server detects that the vehicle's lateral speed is 0.6 m / s and its lateral displacement is 0.5 meters, determining that it has initiated a left turn. The server extracts the turn signal status information and identifies that the left turn signal is on. The server further extracts the relative motion information between the vehicle and the vehicle in front in the same lane: there is a vehicle slowing down due to a red light 5 meters ahead, the relative speed between the two vehicles is approximately 10 km / h, and the longitudinal collision time is calculated to be 5 meters / (10 km / h) = 5 / 2.78 ≈ 1.8 seconds, which is lower than the risk threshold of 3 seconds.

[0232] The server determined that although the vehicle had its turn signal on, its turning action was a hazard avoidance maneuver forced by the risk of collision with the vehicle in front in the same lane—a rear-end collision would occur if the vehicle did not change lanes immediately. This situation meets the second criterion for determining unintended steering. The server determined the unintended steering result to be "triggered defensive driving experience".

[0233] In another embodiment, the target vehicle is driving in a rainy night environment, where perception conditions are limited. The vehicle in front in the adjacent lane on the left is a pickup truck. The server detects that the vehicle's lateral speed is 0.5 m / s and its lateral displacement is 0.4 meters, and determines that it has initiated a turning action. However, due to the low visibility and glare interference from the lights at night, the server cannot reliably identify the turn signal status information, and the perception system outputs "status unknown".

[0234] Based on the conservative principle of defensive driving experience, when the turn signal status cannot be confirmed, the server defaults to the worst-case assumption—that is, "turn signal not activated." If the server determines that the current situation meets the judgment rule of "initiating a turn action and not activating the turn signal" (including the default assumption), the server determines that the unexpected turn result is "triggering defensive driving experience."

[0235] In another embodiment, after the server determines the unexpected steering result, it outputs the result to the comprehensive decision-making module that triggers defensive driving experience. If the result is "defensive driving experience triggered", the server will continue to verify the preset risk conditions (whether there is a collision risk between the target vehicle and the vehicle in front in the adjacent lane, or whether there is a potential collision risk with vulnerable road users in the obstructed area); if the result is "defensive driving experience not triggered", the current defensive driving experience judgment process will be terminated directly, and the target vehicle will maintain the normal driving strategy.

[0236] In one exemplary embodiment, the method further includes:

[0237] If, based on perception information, it is determined that the target vehicle is not in a perception-limited scenario, or is in a perception-limited scenario but has not triggered defensive driving experience, a normal safe trajectory is generated, and the target vehicle is controlled to drive along the normal safe trajectory.

[0238] The standard safe trajectory refers to a normal driving trajectory generated by a trajectory planning algorithm in typical scenarios without defensive driving experience. This trajectory is based on actually perceived obstacle information (excluding phantom obstacles) and aims to maintain lane position, follow the vehicle's path, and avoid real obstacles, satisfying dynamic constraints, traffic rule constraints, and ride comfort constraints. This trajectory does not include defensive maneuvers such as additional deceleration or steering avoidance caused by phantom obstacles.

[0239] In this embodiment of the application, when the server confirms that the target vehicle is not in a perception-limited scenario, or is in a perception-limited scenario but has not triggered defensive driving experience, it generates a normal safety trajectory and controls the target vehicle to drive according to the normal safety trajectory.

[0240] For example, the target vehicle is a Level 4 autonomous passenger vehicle traveling at 65 km / h on an urban expressway. The server continuously monitors the surrounding traffic environment. Based on geometric occlusion model calculations, there are no large vehicles obstructing the view in adjacent lanes, and the forward field of vision is unobstructed. The perception system can completely detect all traffic participants and static obstacles within a 200-meter radius ahead. The server determines that the target vehicle is not in a perception-limited scenario.

[0241] The server directly calls the conventional trajectory planning module. This module uses the target vehicle's current state (speed 65 km / h, heading angle 87.3°) as initial conditions, the lane centerline as a reference trajectory, and the perceived preceding vehicle (95 meters away, speed 62 km / h) as the car-following target. It generates a smooth car-following trajectory for the next 3 seconds using a model predictive control algorithm. The trajectory parameters are: maintaining the current speed, maintaining a safe distance of approximately 3.2 seconds from the preceding vehicle, no steering maneuvers, and no additional deceleration. The server sends the trajectory instructions to the target vehicle, which executes the system's parsed instructions and maintains normal driving.

[0242] In another embodiment, the target vehicle is traveling at 40 km / h on a main urban road, with a heavy van blocking the view in the adjacent lane on its left. The server, after processing the defensive driving experience assessment, outputs an unexpected braking result as "triggered." However, the server further verifies the preset risk conditions: calculations show that the first distance required for the target vehicle to decelerate to a stop at a comfortable speed is 41.9 meters, and the current longitudinal distance to the van is 60 meters, indicating no direct collision risk; a comprehensive test of the blocked area shows no potential collisions for any combination of positions and speeds. The server determines that although the target vehicle is in a perception-limited scenario, it does not meet the preset risk conditions, and ultimately concludes that defensive driving experience was not triggered.

[0243] The server enters the normal trajectory planning process. Since the obstructed area exists but is deemed to pose no potential risk, the server does not generate phantom obstacles and only plans the trajectory based on the actually perceived truck (as an obstacle in the adjacent lane). The truck is currently in the left-hand adjacent lane, has not encroached on this lane, and its relative speed is stable. The server generates a normal, safe trajectory that maintains the lane and travels at a constant speed. The trajectory does not include any deceleration or avoidance maneuvers for phantom obstacles. The vehicle travels normally through the obstructed area according to this trajectory.

[0244] In another embodiment, the target vehicle is traveling at 100 km / h on a highway, and there is a container truck in the adjacent lane to its left. The server, after processing the defensive driving experience assessment, outputs an unexpected braking result of "not triggered"—although the truck is braking, it is determined that the braking is due to the speed limit at the construction zone ahead, which is considered reasonable braking. The server directly determines that the defensive driving experience has not been triggered.

[0245] The server generates a standard, safe trajectory. Since the truck is in an adjacent lane and has not initiated any cutting-in intent, the server treats it as a normally traveling vehicle in the adjacent lane, applying only standard lateral safety distance constraints without additional deceleration or avoidance. The trajectory planning output maintains the current lane and travels at a constant speed, following a standard trajectory. The vehicle proceeds normally.

[0246] In another embodiment, the target vehicle is driving on a suburban road at night. Due to insufficient light, the confidence level of the perception system is reduced. However, through sensor fusion and confidence assessment, the server can still confirm that there are no obstacles or vehicles obstructing the view within a 200-meter radius ahead. The server determines that the target vehicle is not in a perception-limited scenario (although the environmental conditions are harsh, no actual blind spots are formed).

[0247] The server generates a standard, safe trajectory. Since there are no target vehicles ahead, the trajectory planning module uses the road centerline as a reference, sets the cruising speed to 60 km / h, and generates a uniform, straight-line driving trajectory. The vehicle drives stably according to the trajectory instructions.

[0248] In another embodiment, the target vehicle is queuing slowly at an intersection at a speed of 15 km / h. A illegally parked van in the adjacent lane on the right is obstructing the view. The server, after processing the defensive driving experience assessment, outputs an unexpected steering result of "not triggered"—the van is stationary and has not initiated any steering action. The server determines that the target vehicle has not triggered defensive driving experience.

[0249] The server generates a standard, safe trajectory. Since the van is a static obstacle located in an adjacent lane and does not encroach on the current lane, the server only applies static obstacle boundary constraints at its location. The trajectory planning output is a car-following trajectory that moves slowly along the current lane, without generating any additional defensive maneuvers. The vehicle passes through the intersection normally.

[0250] In another embodiment, after the server completes the generation and distribution of the regular safety trajectory, it continuously monitors the status information of the target vehicle and the perceived information of the surrounding traffic environment, and then returns to step 201 to enter the next round of scene recognition and decision-making loop. The generation and execution of the regular safety trajectory is the default working mode of the system, covering the vast majority of normal driving scenarios; the defensive safety trajectory is only activated in special risk scenarios where perception is limited and defensive driving experience is triggered.

[0251] In one exemplary embodiment, the method further includes:

[0252] Figure 5 Here is a flowchart of an autonomous driving safety decision-making method that integrates human defensive driving experience in a perception-limited scenario, as shown in one embodiment. Figure 5As shown, the specific features include: 1. Obtaining information on the motion status of the vehicle and surrounding vehicles through onboard sensors and identifying scenarios with limited perception; 2. Monitoring triggering results online based on human defensive driving experience; 3. Designing a phantom obstacle model; 4. Generating a defensive safety trajectory or a conventional safety trajectory.

[0253] Step 1: Obtain motion status information of the vehicle and surrounding vehicles through onboard sensors and identify scenarios with limited perception.

[0254] Considering the varying degrees of potential risks in perception-limited scenarios, they can be further divided into two typical scenarios: unexpected braking and unexpected braking steering by the DPV of the vehicle ahead in the adjacent lane. For example... Figure 6 As shown, in unintended turning scenarios, the vehicles involved include the EV (electric vehicle), the interfering vehicle (DV) in the adjacent lane, the preceding vehicle (PV) in the same lane as the EV, the preceding vehicle (DPV) in the same lane as the DV, and other vehicles (OV) in the other lane. In addition, vulnerable road users (VRUs), such as pedestrians and non-motorized vehicles, are also involved. To ensure the EV's normal operation, onboard sensors are needed to obtain information such as the position, speed, acceleration, and heading angle of other road users within the observable range. Figure 7 As shown, in more urgent driving situations, the DV makes an unexpected turn, which also obstructs the driver's view, creating a scenario with limited perception.

[0255] Step 2: Online monitoring of trigger results based on human defensive driving experience

[0256] For the monitoring system proposed in this method, the preconditions for defensive driving experiences should first be clearly defined. Monitoring of a defensive driving experience should only begin after the corresponding preconditions are met, which helps improve monitoring accuracy and reduce computational costs. Once the preconditions for a defensive driving experience are met, the relevant logical definitions and the temporal logical relationships between different definitions need to be determined. The temporal states include real-time states and continuous states. Real-time states indicate whether the vehicle's current state triggers a defensive driving experience, such as whether the turn signal is currently activated. Continuous states indicate whether the vehicle's movement over a period of time triggers a defensive driving experience, such as whether the vehicle is within an effective braking or steering timeframe. The final result is as follows: Figure 8 The flowchart shown.

[0257] Unexpected braking is predicated on the presence of a vehicle in the adjacent lane that is braking.

[0258]

[0259]

[0260] In the formula, For the presence of interfering vehicle DV, , Vehicles in the right front area and left front area respectively, due to and The difference in location does not fundamentally change the vehicle's motion state, therefore, the following... For example, we will conduct research. This indicates that the DV has started braking. for longitudinal acceleration, To effectively brake and decelerate the vehicle.

[0261] Once the preconditions are met, the triggering logic needs further verification. Since there are numerous traffic participants in the driving scenario, the driving status of vehicles in adjacent lanes may be affected by surrounding vehicles and the driving environment. Only after excluding other factors that cause the expected braking can it be determined that it is unexpected braking. The specific Metric Temporal Logic (MTL) expression is as follows:

[0262]

[0263]

[0264] In the formula, express The logic for triggering speed limits express Located in a speed-limited area, for longitudinal velocity, , These are the upper and lower limits of the speed limit sign. Indicates that It exists, but it does not exist. The logic that triggers the conflict The vehicle in front of the vehicles in the right front area. The time of longitudinal collision between the two vehicles. It is the threshold for collision time.

[0265] The effective braking phase is considered complete when the vehicle accelerates again after effective braking.

[0266]

[0267] In the formula, express The trigger logic for the end of effective braking. for The start time of effective braking time. It refers to the current moment.

[0268] Taking into account the above propositions, the triggering logic for unexpected braking can be obtained as follows:

[0269]

[0270] In the formula, for The triggering logic for unexpected braking, namely Driving normally within the speed limit area, without braking due to the speed limit, and... There is no risk of collision, nor is there any braking, but its effective braking has not yet ended.

[0271] Although as long as it exists This can create obstructed areas, potentially leading to danger. However, the vehicle only cares about the area where it can interact with other vehicles sufficiently. If the vehicle will not interact with other vehicles... A potential collision occurs, but it is not triggered. At the same time, the state of the Vulnerable Rides (VRUs) in the occluded area also needs to be considered. If the vehicle does not collide with any phantom obstacle, the currently occluded area does not constitute a valid area of ​​concern.

[0272] When the vehicle comfortably decelerates to a stop, it still does not exceed The front of the vehicle will not collide with obstacles in the obstructed area, so the area of ​​interest can be considered invalid.

[0273]

[0274] In the formula, yes and Triggering logic that will not result in a collision. yes longitudinal velocity, For reaction time, For maximum acceleration, To reduce speed for vehicle comfort, In the current state and The longitudinal distance between them.

[0275] like Figure 9 As shown, the obstructed area is determined by the vehicle's viewpoint and The outline, along with the lane boundary, together form the occluded area boundary, which is composed of... , , , , It is formed by connecting points.

[0276]

[0277]

[0278]

[0279]

[0280]

[0281] In the formula, , , , , They are respectively , , , , The vertical position of the point , , , , They are respectively , , , , The horizontal position of the point , , , and They are Its longitudinal position, lateral position, length, width, and heading angle. for Right lane line, , , and They are Its longitudinal position, lateral position, length, and heading angle.

[0282] When a vehicle is comfortably decelerating, it needs to consider VRUs at any position and speed. If it does not collide with any VRU in the obstructed area, the vehicle is considered to be driving safely.

[0283]

[0284] The constraint in the above equation is described in detail as follows:

[0285]

[0286] In the formula, yes Triggering logic that will not collide with VRU. , , These are the VRU's longitudinal position, lateral position, and lateral velocity, respectively. To cover the area, This represents the minimum VRU speed. This represents the maximum speed of the VRU.

[0287] Taking into account the above propositions, the triggering logic for the effective region of interest can be obtained as follows:

[0288]

[0289] In the formula, yes The triggering logic of the effective attention area, i.e. and There is a risk of collision, or although and There is no risk of collision between them, but there is a risk of collision with the VRU.

[0290] Preconditions for an unexpected turn And the vehicle in front in the adjacent lane is turning, that is:

[0291]

[0292] In the formula, This indicates that the DV has started to turn. The threshold for the vehicle's lateral speed at the start of a turn. yes Lateral drift of the vehicle, The threshold value for the lateral position of the vehicle at the start of steering.

[0293] Once the prerequisite conditions are met, it is necessary to comprehensively consider the situation of unexpected turning. The trigger logic for turn signals and movement status is as follows:

[0294]

[0295] In the formula, yes The trigger logic for turn signal activation for The status of the turn signals, To turn on the left turn signal.

[0296] Referring to unexpected braking, when The steering is considered complete when the lateral speed returns to zero again after the steering has begun.

[0297]

[0298] In the formula, express The logic for triggering the end of the turnaround.

[0299] Since the effective attention area triggering logic remains unchanged, the triggering logic expression for unexpected redirection is obtained by synthesis.

[0300]

[0301] In the formula, for The triggering logic for unexpected turns, i.e. Initiating a turn signal without activating the turn signal, or with... When there is a risk of collision Turn on the turn signal and perform the turning action.

[0302] Step 3: Design the phantom obstacle model

[0303] The decision-makers consider the surrounding vehicles in the driving environment and quantify the collision risks between their own vehicle and other vehicles. Among these, they need to pay special attention to the state of the phantom obstacle after the online defensive driving experience monitoring system is triggered. This part can optimize the driving state of the own vehicle in advance and improve the driving safety of the own vehicle.

[0304] like Figure 10 As shown, when the online monitoring system is triggered, phantom obstacles are generated. When the monitoring system is not triggered, if the vehicle observes an obstacle, the actual obstacle is input; otherwise, the obstacle is not considered in the potential field.

[0305]

[0306] In the formula, , , , , These represent the longitudinal position, lateral position, speed, length, and width of the phantom obstacle, respectively. The length of the obstacle. The width of the obstacle.

[0307] Step 4: Generate a defensive security trajectory or a regular security trajectory.

[0308] In constructing the defensive driving potential field, this application's embodiments consider not only the basic positional relationship between the vehicle and other vehicles, but also different vehicle types and different driving behaviors, such as changes in heading angle during lane changes. Based on different types of obstacles, different types of trajectories are generated, including defensive safety trajectories and regular safety trajectories.

[0309]

[0310] In the formula, For the first Defensive driving potential field of individual traffic participants and It is the first The strength and shape coefficients of the defensive driving potential field for each traffic participant. and It is a car and the first The relative lateral and longitudinal distances of each traffic participant and It is the first Longitudinal and lateral safe distances for each traffic participant , , , , , , The first The lateral position, longitudinal position, speed, lateral speed, longitudinal speed, length, and width of each traffic participant. It is a regulatory factor. For the vehicle's heading angle, It is a safe time interval.

[0311] Based on the rolling optimization characteristics and multi-constraint collaborative processing capabilities of model predictive control, defensive or conventional safety trajectories are generated. A three-degree-of-freedom dynamic model is used, and the state-space expression of the vehicle system is:

[0312]

[0313] in:

[0314]

[0315] In the formula, and These are the vertical and horizontal positions in global coordinates. It is the longitudinal force of the tire. It's the front wheel steering angle.

[0316] When designing the cost function, it is necessary to drive along the path with the least penalty in the defensive driving potential field as much as possible. In addition, the vehicle should follow its desired reference trajectory as closely as possible. To improve vehicle comfort and avoid frequent acceleration, deceleration, or steering, the vehicle's control input also needs to be optimized. The cost function of model predictive control is as follows:

[0317]

[0318] In the formula, , These are the forecast range and the control range, respectively. Indicates time Before The predicted value of the step, For the expected value, and These are the state weight matrix and the control weight matrix, respectively. and These are the lower and upper limits that control the input, respectively. and These are the maximum values ​​of the steering angle and the driving torque, respectively. That is the wheel radius.

[0319] In one exemplary embodiment, the method further includes:

[0320] Step 1: Obtain the status information of the target vehicle and the perception information of the surrounding traffic environment.

[0321] Step 2: In the case of a perception-restricted scenario of type I, the braking behavior of the vehicle in front in the adjacent lane is judged and processed unexpectedly based on state information and defensive driving experience knowledge base to obtain unexpected braking results.

[0322] Step 3: Obtain the acceleration information of the vehicle in front in the adjacent lane from the status information, and use the acceleration information to determine whether the braking behavior is in an effective braking state.

[0323] Step 4: Obtain the vehicle speed information and road speed limit information of the vehicle in front in the adjacent lane from the status information, and use the vehicle speed information and road speed limit information to determine whether the braking behavior is triggered by speeding.

[0324] Step 5: Obtain the relative motion information of the vehicle in the adjacent lane and the vehicle in the same lane from the status information, and use the relative motion information to determine whether the braking behavior is triggered by the risk of collision with the vehicle in the same lane.

[0325] Step 6: If the vehicle in front in the adjacent lane is in an effective braking state and the braking behavior is not triggered by speeding or by a collision risk with the vehicle in front in the same lane, determine that the unexpected braking result is the triggering of defensive driving experience.

[0326] Step 7: In the case of a perception-restricted scenario of type II restricted scenario, perform unexpected judgment processing on the steering behavior of the vehicle in front in the adjacent lane based on state information and defensive driving experience knowledge base to obtain unexpected steering results.

[0327] Step 8: Obtain the lateral movement information of the vehicle in front in the adjacent lane from the status information, and use the lateral movement information to determine whether to initiate a steering action.

[0328] Step 9: Obtain the turn signal status information of the vehicle in front in the adjacent lane from the status information.

[0329] Step 10: Obtain the relative motion information of the vehicle in the adjacent lane and the vehicle in the same lane from the status information, and use the relative motion information to determine whether there is a risk of collision.

[0330] Step 11: If the vehicle in front in the adjacent lane initiates a turning action without activating the turn signal, or if the vehicle in front in the adjacent lane initiates a turning action, activates the turn signal, and turns due to a risk of collision with the vehicle in front in the same lane, determine that the unexpected turning result triggers defensive driving experience.

[0331] Step 12: If the unexpected braking result or unexpected steering result indicates the triggering of defensive driving experience, and the target vehicle meets the preset risk conditions, determine that the target vehicle has triggered defensive driving experience; the preset risk conditions include any of the following: there is a collision risk between the target vehicle and the vehicle in front in the adjacent lane; there is a potential collision risk between the target vehicle and a vulnerable road user in the area obscured by the vehicle in front in the adjacent lane.

[0332] Step 13: Based on the state information of the target vehicle, calculate the first distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration.

[0333] Step 14: Based on the first distance and the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane, determine whether there is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

[0334] Step 15: Determine the occlusion area based on the state information, which is jointly defined by the outline of the vehicle in front in the adjacent lane, the position of the target vehicle, and the lane boundary.

[0335] Step 16: Within the obscured area, determine whether there is a potential collision on the current trajectory of the target vehicle based on the preset speed range, location distribution, and status information of vulnerable road users.

[0336] Step 17: Given that the processing result represents the target vehicle triggering defensive driving experience, the current scene parameters are input into the pre-trained phantom obstacle generation model according to the type characteristics of the perception-limited scene to obtain the phantom obstacle.

[0337] Step 18: Based on the phantom obstacles, a defensive safety trajectory is generated using a preset trajectory planning and control algorithm, and the target vehicle is controlled to travel along the defensive safety trajectory.

[0338] Step 19: If, based on perception information, it is determined that the target vehicle is not in a perception-limited scenario, or is in a perception-limited scenario but has not triggered defensive driving experience, a normal safety trajectory is generated, and the target vehicle is controlled to drive along the normal safety trajectory.

[0339] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0340] Based on the same inventive concept, this application also provides an autonomous driving safety decision-making device for integrating human defensive driving experience in perception-limited scenarios, used to implement the aforementioned autonomous driving safety decision-making method integrating human defensive driving experience in perception-limited scenarios. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the autonomous driving safety decision-making device integrating human defensive driving experience in perception-limited scenarios provided below can be found in the limitations of the autonomous driving safety decision-making method integrating human defensive driving experience in perception-limited scenarios described above, and will not be repeated here.

[0341] In one exemplary embodiment, such as Figure 11 As shown, an autonomous driving safety decision-making device that integrates human defensive driving experience in perception-limited scenarios is provided, comprising: an acquisition module 501, a processing module 502, an input module 503, and a generation module 504, wherein:

[0342] The acquisition module 501 is used to acquire the status information of the target vehicle and the perception information of the surrounding traffic environment;

[0343] The processing module 502 is used to perform defensive driving experience judgment processing on the current driving scenario of the target vehicle based on state information and a preset defensive driving experience knowledge base when it is determined that the target vehicle is in a perception-limited scenario according to the perception information, and to obtain the processing result.

[0344] The input module 503 is used to input the current scene parameters into a pre-trained phantom obstacle generation model to obtain phantom obstacles, based on the type characteristics of the perception-limited scene, when the processing result characterizes the target vehicle's defensive driving experience.

[0345] The generation module 504 is used to generate a defensive safety trajectory based on the phantom obstacles using a preset trajectory planning and control algorithm, and to control the target vehicle to drive along the defensive safety trajectory.

[0346] In an exemplary embodiment, the perception-restricted scenario includes a first type of restricted scenario formed by an unexpected braking by a vehicle in the adjacent lane and a second type of restricted scenario formed by an unexpected steering by a vehicle in the adjacent lane. The processing results include unexpected braking results and unexpected steering results. Specifically, the processing module 502 is used to perform unexpected judgment processing on the braking behavior of the vehicle in the adjacent lane based on state information and a defensive driving experience knowledge base when the perception-restricted scenario is the first type of restricted scenario, to obtain an unexpected braking result; and to perform unexpected judgment processing on the steering behavior of the vehicle in the adjacent lane based on state information and a defensive driving experience knowledge base when the perception-restricted scenario is the second type of restricted scenario, to obtain an unexpected steering result.

[0347] In an exemplary embodiment, the autonomous driving safety decision-making device that integrates human defensive driving experience in the above-mentioned perception-limited scenario is further used to determine that the target vehicle has triggered defensive driving experience when the unexpected braking result or unexpected steering result indicates that defensive driving experience has been triggered, and the target vehicle meets the preset risk conditions.

[0348] In one exemplary embodiment, the aforementioned preset risk conditions include any of the following conditions:

[0349] The target vehicle is at risk of colliding with the vehicle in front in the adjacent lane.

[0350] There is a potential collision risk between the target vehicle and a vulnerable road user within the obstruction area created by the vehicle in front in the adjacent lane.

[0351] In an exemplary embodiment, the autonomous driving safety decision-making device that integrates human defensive driving experience in the above-mentioned perception-limited scenario is further used to calculate the first distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration based on the state information of the target vehicle.

[0352] Based on the first distance and the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane, determine whether there is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

[0353] In an exemplary embodiment, the autonomous driving safety decision-making device that integrates human defensive driving experience in the above-mentioned perception-limited scenario is also used to determine the occlusion area defined by the outline of the vehicle in front in the adjacent lane, the position of the target vehicle, and the lane boundary based on state information.

[0354] Within the obscured area, the potential for collision is determined based on the speed range, location distribution, and status information of the vulnerable road users.

[0355] In an exemplary embodiment, the autonomous driving safety decision-making device that integrates human defensive driving experience in the above-mentioned perception-limited scenario is further used to obtain the acceleration information of the vehicle in front in the adjacent lane from the state information, and use the acceleration information to determine whether the braking behavior is in an effective braking state.

[0356] The vehicle speed information and road speed limit information of the vehicle in front in the adjacent lane are obtained from the status information, and the vehicle speed information and road speed limit information are used to determine whether the braking behavior is triggered by speeding.

[0357] Obtain the relative motion information between the vehicle in the adjacent lane and the vehicle in the same lane from the status information, and use the relative motion information to determine whether the braking behavior is triggered by a collision risk with the vehicle in the same lane.

[0358] If the vehicle in front in the adjacent lane is in a state of effective braking that has not ended, and the braking behavior is not triggered by speeding or by a risk of collision with the vehicle in front in the same lane, the unexpected braking result is determined to trigger defensive driving experience.

[0359] In an exemplary embodiment, the autonomous driving safety decision-making device that integrates human defensive driving experience in the above-mentioned perception-limited scenario is further used to obtain the lateral movement information of the vehicle in front in the adjacent lane from the state information, and use the lateral movement information to determine whether to initiate a steering action.

[0360] Obtain the turn signal status information of the vehicle in front in the adjacent lane from the status information;

[0361] Obtain the relative motion information of the vehicle in the adjacent lane and the vehicle in the same lane from the status information, and use the relative motion information to determine whether there is a risk of collision;

[0362] When a vehicle in front in an adjacent lane initiates a turning action without activating its turn signal, or when a vehicle in front in an adjacent lane initiates a turning action with its turn signal activated and turns due to a risk of collision with the vehicle in front in the same lane, the unexpected turning result is determined to trigger defensive driving experience.

[0363] In an exemplary embodiment, the autonomous driving safety decision-making device that integrates human defensive driving experience in the above-mentioned perception-limited scenario is further used to generate a normal safety trajectory and control the target vehicle to drive according to the normal safety trajectory when it is determined based on perception information that the target vehicle is not in a perception-limited scenario, or is in a perception-limited scenario but has not triggered defensive driving experience.

[0364] The various modules in the autonomous driving safety decision-making device that integrates human defensive driving experience in the aforementioned perception-limited scenarios can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0365] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0366] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0367] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0368] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0369] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0370] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for autonomous driving safety decision-making that integrates human defensive driving experience in perception-limited scenarios, characterized in that, The method includes: Acquire the target vehicle's status information and the surrounding traffic environment's perception information; When it is determined that the target vehicle is in a perception-limited scenario based on the perception information, defensive driving experience judgment processing is performed on the current driving scenario of the target vehicle based on the state information and a preset defensive driving experience knowledge base to obtain the processing result. When the processing result characterizes the target vehicle triggering defensive driving experience, the current scene parameters are input into a pre-trained phantom obstacle generation model according to the type characteristics of the perception-limited scene to obtain phantom obstacles; A defensive safety trajectory is generated based on the phantom obstacles using a preset trajectory planning and control algorithm, and the target vehicle is controlled to travel along the defensive safety trajectory.

2. The method according to claim 1, characterized in that, The perception-limited scenarios include a first type of limited scenario formed by an unexpected braking by a vehicle in the adjacent lane, and a second type of limited scenario formed by an unexpected steering by a vehicle in the adjacent lane; the processing results include unexpected braking results and unexpected steering results. When the processing result indicates that the target vehicle has triggered defensive driving experience, based on the state information and a preset defensive driving experience knowledge base, the current driving scenario of the target vehicle is subjected to defensive driving experience judgment processing to obtain a processing result, including: When the perception-limited scenario is the first type of limited scenario, the braking behavior of the vehicle in front in the adjacent lane is judged and processed unexpectedly based on the state information and the defensive driving experience knowledge base to obtain the unexpected braking result. When the perception-limited scenario is the second type of limited scenario, the unexpected steering behavior of the vehicle in front in the adjacent lane is judged and processed based on the state information and the defensive driving experience knowledge base to obtain the unexpected steering result.

3. The method according to claim 2, characterized in that, The method further includes: If the unexpected braking result or the unexpected steering result indicates that a defensive driving experience has been triggered, and the target vehicle meets the preset risk conditions, then the target vehicle is determined to have triggered a defensive driving experience.

4. The method according to claim 3, characterized in that, The preset risk conditions include any of the following conditions: The target vehicle is at risk of collision with the vehicle in front in the adjacent lane; The target vehicle poses a potential collision risk with vulnerable road users within the obstruction area formed by the obstruction of the vehicle in front in the adjacent lane.

5. The method according to claim 4, characterized in that, The process of determining whether there is a collision risk between the target vehicle and the vehicle in front in the adjacent lane includes: Based on the state information of the target vehicle, calculate the first distance required for the target vehicle to decelerate to a stop at a preset comfort deceleration. Based on the first distance and the current longitudinal distance between the target vehicle and the vehicle in front in the adjacent lane, it is determined whether there is a risk of collision between the target vehicle and the vehicle in front in the adjacent lane.

6. The method according to claim 4, characterized in that, The process of determining whether there is a potential collision risk between the target vehicle and vulnerable road users within the obstructed area includes: Based on the state information, a masking area is determined by the outline of the vehicle in front of the adjacent lane, the position of the target vehicle, and the lane boundary. Within the obstructed area, the potential for collision is determined based on the preset speed range, location distribution, and status information of vulnerable road users.

7. The method according to claim 2, characterized in that, The method further includes: The acceleration information of the vehicle in front in the adjacent lane is obtained from the status information, and the acceleration information is used to determine whether the braking behavior is in an effective braking state. The vehicle speed information and road speed limit information of the vehicle in front in the adjacent lane are obtained from the status information, and the vehicle speed information and road speed limit information are used to determine whether the braking behavior is triggered by speeding. The relative motion information between the vehicle in the adjacent lane and the vehicle in the same lane is obtained from the status information, and the relative motion information is used to determine whether the braking behavior is triggered by a collision risk with the vehicle in the same lane. If the vehicle in front in the adjacent lane is in a state of effective braking that has not ended, and the braking behavior is not triggered by speeding or by a collision risk with the vehicle in front in the same lane, the unexpected braking result is determined to trigger defensive driving experience.

8. The method according to claim 2, characterized in that, The method further includes: The lateral movement information of the vehicle in front in the adjacent lane is obtained from the status information, and the lateral movement information is used to determine whether to initiate a steering action. Obtain the turn signal status information of the vehicle in front in the adjacent lane from the status information; The relative motion information between the vehicle in the adjacent lane and the vehicle in the same lane is obtained from the status information, and the relative motion information is used to determine whether there is a risk of collision. If the vehicle in front in the adjacent lane initiates a turning action without activating its turn signal, or if the vehicle in front in the adjacent lane initiates a turning action, activates its turn signal, and turns due to a risk of collision with the vehicle in front in the same lane, the unexpected turning result is determined to trigger defensive driving experience.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: If, based on the perceived information, it is determined that the target vehicle is not in a perception-limited scenario, or is in a perception-limited scenario but has not triggered defensive driving experience, a normal safety trajectory is generated, and the target vehicle is controlled to drive according to the normal safety trajectory.

10. An autonomous driving safety decision-making device that integrates human defensive driving experience in perception-limited scenarios, characterized in that, The device includes: The acquisition module is used to acquire the status information of the target vehicle and the perception information of the surrounding traffic environment; The processing module is used to perform defensive driving experience judgment processing on the current driving scenario of the target vehicle based on the state information and a preset defensive driving experience knowledge base when it is determined that the target vehicle is in a perception-limited scenario according to the perception information, and to obtain the processing result. The input module is used to input the current scene parameters into a pre-trained phantom obstacle generation model to obtain phantom obstacles, based on the type characteristics of the perception-limited scene, when the processing result characterizes the target vehicle triggering defensive driving experience; The generation module is used to generate a defensive safety trajectory based on the phantom obstacles using a preset trajectory planning and control algorithm, and to control the target vehicle to travel along the defensive safety trajectory.

Images