Vehicle collision control method and vehicle

Abstract

The application provides a vehicle collision control method and a vehicle. The method is applied to a vehicle and includes: obtaining a plurality of candidate strategies, the candidate strategies including braking parameters and steering parameters of the vehicle; predicting injury scores of each candidate strategy when the vehicle collides with a front target; determining a target strategy from the plurality of candidate strategies according to the injury scores of the candidate strategies, and controlling the vehicle to perform at least one of braking and steering actions based on the target strategy. The method can actively select a collision mode that minimizes collision injury by cooperatively controlling braking and steering when the collision is inevitable, thereby minimizing the collision injury of the passengers and the collision target.

Description

Technical Field

[0001] This application relates to the field of vehicles, and more specifically, to a vehicle collision control method and a vehicle in the field of vehicles. Background Technology

[0002] The decision-making logic of current mainstream Automatic Emergency Braking (AEB) systems is mainly based on simple threshold judgments such as Time-To-Collision (TTC). That is, when the sensor detects a target ahead, the system calculates the TTC and determines whether it is below a preset threshold; if not, no intervention is taken; if so, maximum braking force is triggered to implement emergency braking in order to reduce the relative collision speed to a minimum or even zero. However, in complex traffic scenarios, if a collision is unavoidable, using only full braking, while reducing the relative collision speed, may still result in serious injury to occupants due to the collision. Summary of the Invention

[0003] This application provides a vehicle collision control method and a vehicle. When a collision is unavoidable, the method can actively select a collision mode that minimizes collision damage by coordinating braking and steering, thereby minimizing collision damage to occupants and the collision target.

[0004] In a first aspect, a vehicle collision control method is provided, the method comprising: acquiring multiple candidate strategies, the candidate strategies including vehicle braking parameters and steering parameters; predicting the damage score of each candidate strategy when the vehicle collides with a target in front; determining a target strategy from the multiple candidate strategies based on the damage scores of the candidate strategies; and controlling the vehicle to perform at least one braking action and steering action based on the target strategy.

[0005] The above technical solution enables quantitative assessment and optimization based on the injury score at the time of collision, breaking through the limitations of traditional automatic emergency braking that relies solely on full braking and ignores the collision pattern. In scenarios where collision is unavoidable, it actively avoids collision patterns with high overlap and high injury risk, selects the optimal avoidance trajectory with the lowest risk of occupant injury, thereby significantly improving collision compatibility while reducing relative speed and minimizing the serious injury caused to occupants by collision accidents.

[0006] In conjunction with the first aspect, in some possible implementations, the damage score of each candidate strategy is calculated when the vehicle is predicted to collide with the target ahead, including: predicting candidate collision parameters between the vehicle and the target ahead based on the candidate strategy; and calculating the damage score of each candidate strategy based on the candidate collision parameters.

[0007] The above technical solutions can transform the predicted contact state between the vehicle and the target under different braking and steering combinations into intuitive injury risk indicators, thereby providing a safety assessment basis for the decision-making system. This ensures that the final selected target strategy can not only avoid or mitigate collisions, but also specifically avoid high-risk collision patterns that lead to serious injuries, significantly improving the scientific nature and safety of emergency avoidance decisions.

[0008] In conjunction with the first aspect, in some possible implementations, predicting candidate collision parameters between the vehicle and the target ahead based on candidate strategies includes: acquiring the historical trajectory and perception data of the target ahead; generating a first predicted trajectory of the target ahead based on the historical trajectory and perception data; generating a second predicted trajectory of the vehicle based on each candidate strategy; and generating candidate collision parameters based on the first and second predicted trajectories.

[0009] The above technical solution can simulate the relative motion process between the vehicle and the target under different avoidance operations, thereby quantifying and calculating key candidate collision parameters in advance, effectively overcoming the limitation of traditional methods that rely solely on instantaneous state estimation.

[0010] In conjunction with the first aspect, in some possible implementations, generating candidate collision parameters based on the first predicted trajectory and the second predicted trajectory includes: determining the kinematic features and collision geometry features of the vehicle and the target ahead at the time of the collision based on the first predicted trajectory and the second predicted trajectory, wherein the kinematic features include vehicle motion features and target motion features, and the collision geometry features include vehicle geometry features and target geometry features; and generating candidate collision parameters based on the kinematic features and collision geometry features.

[0011] The above technical solution can transform abstract trajectory prediction into specific and high-dimensional collision feature descriptions to capture key physical elements that determine the degree of injury (i.e., vehicle motion features and target motion features). It can comprehensively reconstruct the physical state at the moment of collision from both kinematic and collision geometry dimensions, thereby providing physical semantic input for injury score calculation and significantly improving the accuracy and robustness of occupant injury risk prediction in complex collision scenarios.

[0012] In conjunction with the first aspect, in some possible implementations, the damage score of each candidate strategy is calculated based on the candidate collision parameters, including: extracting kinematic features and collision geometry features from the candidate collision parameters; obtaining the first attribute information of the vehicle and the second attribute information of the target in front; and calculating the damage score of each candidate strategy based on the kinematic features, collision geometry features, first attribute information and second attribute information.

[0013] The above technical solution can deeply couple kinematic and collision geometry features with first and second attribute information (including compatibility features such as mass and stiffness), breaking through the limitation of traditional algorithms that rely solely on relative velocity for single-dimensional risk assessment. It can quantify the energy absorption characteristics and risks of different vehicle-target combinations under specific collision modes, and considers the calculation mechanism of vehicle-target compatibility differences. This significantly improves the physical realism and scene adaptability of injury scoring, ensuring that the system can select the truly optimal avoidance strategy for specific target characteristics (such as pedestrians, two-wheeled vehicles, or vehicles with different stiffness).

[0014] In conjunction with the first aspect, in some possible implementations, the damage score of each candidate strategy is calculated based on kinematic features, collision geometry features, first attribute information, and second attribute information. This includes: generating vehicle collision parameters and target collision parameters based on kinematic features and collision geometry features, where the vehicle collision parameters include vehicle motion features and vehicle geometry features, and the target collision parameters include target motion features and target geometry features; calculating the vehicle's damage score based on the first attribute information and vehicle collision parameters; calculating the damage score of the target ahead based on the second attribute information and target collision parameters; and calculating the damage score of the candidate strategy based on the vehicle's damage score and the target's damage score.

[0015] The above technical solutions enable the establishment of dedicated risk mapping models for the protection of occupants of the vehicle (based on vehicle attributes and its own collision parameters) and the protection of vulnerable road users or other vehicles (based on target attributes and target collision parameters). This effectively solves the pain point that traditional single scoring mechanisms cannot take into account the safety interests and compatibility differences of both parties, and ensures that the finally selected candidate strategy can minimize overall harm under complex safety constraints.

[0016] In conjunction with the first aspect, in some possible implementations, before acquiring multiple candidate strategies, the following steps are also included: acquiring the historical trajectory and perception data of the target ahead; generating a first predicted trajectory of the target ahead based on the historical trajectory and perception data, and generating a vehicle driving trajectory based on the current state of the vehicle; determining the first collision position and the first collision time between the vehicle and the target obstacle based on the first predicted trajectory and the vehicle driving trajectory, and calculating the first relative distance between the current position of the vehicle and the first collision position; calculating the braking completion time of the vehicle based on the maximum braking deceleration of the vehicle; if the braking completion time is after the first collision time, then acquiring multiple candidate strategies; if the braking completion time is before the first collision time, then directly triggering braking.

[0017] The above technical solution enables on-demand triggering and tiered response of emergency avoidance strategies by introducing a collision time-based prediction mechanism. First, it integrates perception data and historical trajectories to accurately predict the collision time-space point (first collision location / time) and relative distance, and calculates the vehicle's stopping time under extreme braking. Then, it logically separates the braking completion time with the collision time. If the vehicle can come to a complete stop before the collision, a single braking command is executed directly, avoiding unnecessary steering intervention and ensuring the simplicity and stability of control. Only when simple braking cannot avoid the collision (i.e., the braking completion time is later than the collision time) is a complex multi-candidate strategy generation process initiated to seek steering avoidance or joint control solutions. This significantly reduces the computational load and risk of misoperation under non-extreme conditions, and ensures that high-order obstacle avoidance algorithms can be invoked in a timely manner during truly critical moments, thus achieving an optimal balance between driving safety, control smoothness, and computational efficiency.

[0018] In conjunction with the first aspect, in some possible implementations, the target strategy is determined from multiple candidate strategies based on the damage score of the candidate strategies, including: extracting the vehicle's braking and steering parameters from the candidate strategies; determining the vehicle's stability score and path deviation score for executing the candidate strategies based on the braking and steering parameters; calculating the comprehensive score of the candidate strategies based on the damage score, stability score, and path deviation score; and determining the target strategy from multiple candidate strategies based on the comprehensive score.

[0019] The above technical solutions can introduce stability scores and path deviation scores derived from braking and steering parameters, quantify the risk of vehicle dynamic instability and lane keeping ability during strategy execution, and ensure that the final selected target strategy can not only reduce collision damage in theory, but also has high executability and robustness at the physical level. It effectively avoids the risk of vehicle skidding, loss of control or secondary accidents caused by aggressive avoidance, and significantly improves the overall safety under extreme conditions.

[0020] In conjunction with the first aspect, some possible implementations may include, before determining the target strategy from multiple candidate strategies based on the comprehensive score, the following steps may be taken: determining the vehicle's control cost score based on braking parameters and steering parameters; and correcting the comprehensive score based on the control cost score.

[0021] The above technical solution can effectively prevent the system from outputting extreme commands such as sudden braking or sharp steering in pursuit of theoretically maximum damage reduction by controlling the cost score. This not only significantly improves the comfort of occupants during the avoidance process and avoids secondary injuries to occupants caused by excessive vehicle dynamic response, but also ensures that the generated control commands are strictly within the physical response limits of the vehicle actuators and the tire adhesion boundary, thereby guaranteeing the smooth implementation and high reliability of the target strategy in the real physical world.

[0022] In a second aspect, a vehicle is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the vehicle collision control method of the first aspect.

[0023] Thirdly, a vehicle collision control device is provided, comprising: an acquisition module for acquiring multiple candidate strategies, the candidate strategies including vehicle braking parameters and steering parameters; a prediction module for predicting the damage score of each candidate strategy when the vehicle collides with a target in front; and an execution module for determining a target strategy from the multiple candidate strategies based on the damage scores of the candidate strategies, and controlling the vehicle to perform at least one braking action and a steering action based on the target strategy.

[0024] Fourthly, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the vehicle collision control method of the first aspect.

[0025] Fifthly, a computer-readable storage medium is provided having a computer program or instructions stored thereon, which, when executed, implement the vehicle collision control method of the first aspect. Attached Figure Description

[0026] Figure 1 This is a schematic diagram illustrating the working principle of an automatic emergency braking system in related technologies; Figure 2 This is a diagram of the vehicle collision control system architecture according to an embodiment of this application; Figure 3 This is a schematic flowchart of a vehicle collision control method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the vehicle collision control device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the vehicle structure provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions in this application will be clearly and thoroughly described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in the text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0028] Hereinafter, the terms "first" and "second" are for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features.

[0029] The working principle of the automatic emergency braking system in the traditional method is as Figure 1 shown. Its decision logic is based on simple thresholds such as the time to collision (TTC). When the TTC is below the threshold, the system triggers braking. The core objective is to reduce the relative collision speed to the lowest or zero.

[0030] The fundamental limitation of this method is that in complex traffic scenarios (such as when other vehicles suddenly cut in or the driver's reaction is insufficient), the system may judge that a collision is inevitable. At this time, the single full-power braking strategy of the traditional automatic emergency braking system, although reducing the relative collision speed, may result in a frontal, high-overlap collision pattern (i.e., the collision momentum index CMI, where CMI is defined as CMI = ΔV / Vr, where ΔV is the speed change experienced by the host vehicle in the collision and Vr is the magnitude of the relative speed vector between the host vehicle and the target in front at the moment of collision). This pattern poses a very high risk to the occupants, especially those suffering from severe injuries (MAIS 3+ <Maximum Abbreviated Injury Scale 3 or greater, the highest concise injury level 3 and above, often simply referred to as serious or severe injuries>).

[0031] Next, the application scenarios or system architectures of the embodiments of the present application will be described. Refer to Figure 2 , generally, the embodiments of the present application can be implemented through Figure 2 a vehicle collision control system. The system includes a sensing layer, a decision-making layer, and an execution layer. Among them, the sensing layer is responsible for collecting information on the internal and external environments of the vehicle and its own state, and is composed of sensors such as cameras, millimeter-wave radars, and lidar; the decision-making layer contains multiple functional modules, which can be divided into four major stages according to the processing flow: fusion prediction, collision risk assessment, decision-making judgment, and optimized decision-making; the execution layer receives the instructions from the decision-making layer and coordinates each actuator through the chassis domain controller to precisely complete the specified vehicle control actions.

[0032] Figure 3 is a schematic flowchart of a vehicle collision control method provided by an embodiment of the present application.

[0033] Exemplarily, as Figure 3 shown, the method includes the following steps: In step S101, multiple candidate strategies are obtained. The candidate strategies include the braking parameters and steering parameters of the vehicle.

[0034] The candidate strategies are generated by a multi-strategy generator. Each strategy contains a specific target braking deceleration (not a maximum value) and a small, stable target steering angle or steering curvature. All strategies are generated under the constraint of ensuring that the vehicle's lateral stability is not lost and does not deviate significantly from the lane boundary. The braking parameters refer to the command variables of the candidate strategies regarding longitudinal control, specifically the target braking deceleration. The steering parameters refer to the command variables of the candidate strategies regarding lateral control, specifically the target steering angle or steering curvature.

[0035] It is understood that, when the vehicle collision is predicted to be unavoidable, the embodiments of this application no longer simply perform full braking. Instead, under the premise of ensuring that the vehicle does not lose control and does not deviate from the lane safety boundary, multiple candidate strategies containing different combinations of braking deceleration and small steering angles are generated. This lays the foundation for selecting the optimal solution that minimizes the overall damage risk from all candidate strategies, and realizes the technical upgrade from simply avoiding collisions to intelligently selecting the collision mode with the least damage.

[0036] In this embodiment of the application, before obtaining multiple candidate strategies, the method further includes: obtaining the historical trajectory and perception data of the target ahead; generating a first predicted trajectory of the target ahead based on the historical trajectory and perception data, and generating a vehicle driving trajectory based on the current state of the vehicle; determining the first collision position and the first collision time between the vehicle and the target obstacle based on the first predicted trajectory and the vehicle driving trajectory, and calculating the first relative distance between the current position of the vehicle and the first collision position; calculating the braking completion time of the vehicle based on the maximum braking deceleration of the vehicle; if the braking completion time is after the first collision time, then obtaining multiple candidate strategies; if the braking completion time is before the first collision time, then directly triggering braking.

[0037] Historical trajectory and perception data refer to the position sequence, velocity changes, and acceleration information of a target object (such as a pedestrian, two-wheeled vehicle, or vehicle of different stiffness) over a past period, collected in real time by onboard sensors (such as radar and cameras). The historical trajectory reflects the target's motion trend, while the perception data provides the target's current state and physical attributes. Both serve as the basic input for predicting future behavior. The first predicted trajectory refers to the most likely driving path of the target in the future, deduced using kinematic models or behavior prediction algorithms based on the historical trajectory and current perception data of the target. It is a key basis for determining whether a collision will occur and the nature of the collision. The trajectory prediction model can process historical trajectories and perceived data to obtain the first predicted trajectory. This model includes a feature extraction layer, a feature encoding layer, and a trajectory decoding layer. The feature extraction layer maps low-dimensional historical trajectories and perceived data inputs to a high-dimensional space and calculates local relative relationships, extracting rich feature embeddings containing high-frequency details and semantic information from the original data. The feature encoding layer utilizes a cross-attention mechanism to decompose and fuse the global relationships between the target and the map, historical time series, and interactions between targets, enhancing the model's spatiotemporal perception capability in complex traffic scenarios. The trajectory decoding layer generates multiple candidate trajectories and their confidence scores to output the final prediction result, for example: The trajectory prediction model is built based on QcNet (Query-Centric Trajectory Prediction). It adopts query-centric encoding to establish invariant local features that are independent of the global context for each scene element. The training process is as follows: the feature extraction layer provides the necessary spatial relationship information by introducing the relative spatiotemporal position embedding between two elements. The input includes all information of traffic participants and surrounding environment information, specifically including target information such as position coordinates, lateral / longitudinal velocity / acceleration, heading angle, and orientation angle, as well as map information such as lane segments and lane lines. The model calculates feature embeddings (such as relative distance, relative angle, etc.) in the local coordinate system of each element, and uses Fourier features to map low-dimensional inputs to high-dimensional space to extract high-frequency details and semantic information from time-series data. The feature encoding layer uses a cross-attention mechanism for feature fusion, decomposing complex global interactions into three easier-to-process modules: capturing the relationship between the target and the map through Agent-Map-Attention; obtaining the target's features in the historical time dimension through the Agent_Time_Attention module; and extracting the interaction information between target vehicles through Agent_Agnet_Attention. During attention calculation, the model also explicitly adds original feature information to enhance the model's ability to perceive spatiotemporal information. The trajectory decoding layer adopts a query-based decoder and is divided into two stages. The first stage is a recurrent, anchorless trajectory proposal. Using a set of learnable query vectors, the trajectory interacts with the scene context through cross-attention computation similar to DETR (DEtection Transformer, a Transformer-based object detection model). This stage outputs three preliminary trajectory proposals as dynamic anchors for the second stage. The second stage is anchor-based trajectory refinement. The output trajectory of the first stage is encoded using a GRU (Gated Recurrent Unit) network, and its output is used as a new anchor-based query. These queries are again cross-attention computed with the scene encoding. Finally, the output is the offset relative to the proposed trajectory and the confidence score of each of the three trajectories. The trajectory with the highest confidence score is used as the first predicted trajectory.

[0038] The vehicle's trajectory is determined by the vehicle's current state or its current planned trajectory. Based on the first predicted trajectory and the vehicle's trajectory, the first collision position and the first collision time between the vehicle and the target obstacle are determined through trajectory intersection analysis. Trajectory intersection analysis is a spatiotemporal geometric calculation method used to simultaneously examine the predicted paths of two moving objects over a period of time in the future. This analysis not only focuses on the spatial path overlap (i.e., whether two lines intersect), but also strictly combines the time dimension to determine whether the two objects will arrive at the overlap point at the same time, thereby distinguishing between safe scenarios where paths intersect but times are staggered and dangerous scenarios where time and space coincide.

[0039] It is understood that, before generating multiple candidate control strategies, this application embodiment first generates a first predicted trajectory using the historical trajectory and perception data of the target ahead, and simulates the vehicle's driving trajectory in conjunction with the current state of the vehicle, thereby calculating the first collision position, the first collision time, and the first relative distance between the two without intervention. Subsequently, the system calculates the theoretical braking completion time based on the vehicle's maximum braking deceleration and makes a key comparison with the first collision time. If the braking completion time is earlier than the collision time, it means that braking alone can avoid the accident, and braking is directly triggered; conversely, if the braking completion time is later than the collision time, it means that braking alone is insufficient to avoid the accident. If a collision cannot be avoided, a more complex emergency mechanism will be immediately activated to acquire and evaluate multiple candidate strategies to find a better avoidance solution. This mechanism realizes graded response and precise triggering of avoidance strategies, and can quickly determine whether simple braking is sufficient to resolve the crisis. If it is sufficient, it will brake directly, avoiding the risk of loss of control and passenger discomfort caused by unnecessary steering operations. If the braking is insufficient, it will immediately and seamlessly switch to the multi-candidate strategy selection mode to ensure that steering and other means can be called in extreme conditions. This not only reduces the computational load and the probability of erroneous actions, but also makes the vehicle's emergency response more decisive, smooth and in line with human driving intuition while ensuring safety.

[0040] In step S102, the damage score of each candidate strategy is predicted when the vehicle collides with the target in front.

[0041] It is understood that, when a collision is unavoidable, the embodiments of this application simulate the critical situation of a collision between a vehicle and a target ahead for each generated candidate strategy, calculate and evaluate the damage score that may be caused at the moment of collision, thereby providing a basis for the decision-making system to select the optimal avoidance path with the least risk.

[0042] In this embodiment of the application, predicting the damage score of each candidate strategy when a collision occurs between a vehicle and a target ahead includes: predicting candidate collision parameters between the vehicle and the target ahead based on the candidate strategies; and calculating the damage score of each candidate strategy based on the candidate collision parameters.

[0043] Among them, candidate collision parameters refer to the feature vectors derived from the specific physical interaction data under the premise that the vehicle executes a certain candidate strategy and eventually collides with the target in front. These parameters usually include key indicators such as the predicted relative speed at the collision point, the predicted change in the vehicle's speed, the predicted change in the target vehicle's speed, and the equivalent energy speed.

[0044] It is understood that, for each generated candidate strategy, this application embodiment simulates and deduces several specific candidate collision parameters (such as relative speed and impact angle) between the vehicle and the target ahead when a collision occurs after the strategy is executed. Subsequently, using the derived candidate collision parameters, the damage score corresponding to each strategy is obtained through a calculation model, thereby intuitively measuring the potential harm of each strategy in extreme conditions in numerical form, assisting the system in selecting the decision with the lowest risk. By transforming abstract avoidance actions into specific collision parameters and mapping them to intuitive damage scores, the risks of each strategy under extreme conditions can be accurately identified at the decision-making front end. This not only significantly improves the decision-making algorithm's analytical depth and safety boundary control capabilities for complex collision scenarios, but also ensures that the selection of the final target strategy strictly follows the principle of minimizing damage, thereby fundamentally optimizing the active safety performance and decision-making level of the autonomous driving system.

[0045] In this embodiment of the application, predicting candidate collision parameters between a vehicle and a target ahead according to a candidate strategy includes: acquiring the historical trajectory and perception data of the target ahead; generating a first predicted trajectory of the target ahead based on the historical trajectory and perception data; generating a second predicted trajectory of the vehicle according to each candidate strategy; and generating candidate collision parameters based on the first predicted trajectory and the second predicted trajectory.

[0046] The second predicted trajectory refers to the expected driving path and state sequence of the vehicle within the same time period, simulated by the vehicle dynamics model after the vehicle executes a specific candidate strategy. Each candidate strategy will generate a unique second predicted trajectory. The state variables and control inputs of the vehicle dynamics model are: State variables ,in, The longitudinal velocity of the vehicle's center of gravity (m / s); The lateral velocity of the vehicle's center of gravity (m / s); The vehicle's yaw rate (rad / s); The position (m) of the vehicle's center of mass in the world coordinate system; The vehicle heading angle (rad); Indicates transpose; Control input include: The target longitudinal acceleration (m / s²) comes from the braking deceleration command (negative value) in the strategy. The front wheel steering angle (rad) comes from the steering angle command in the strategy.

[0047] The core dynamic equation is Longitudinal movement: ,in, The total mass of the vehicle; This refers to the longitudinal tire force on the front and rear axles; This refers to air resistance. For emergency braking It is approximately linearly related to braking pressure and is limited by the road adhesion coefficient μ. A load-dependent linear tire model is adopted: ,in The front wheel slip ratio, The rear wheel slip ratio is mapped from the target deceleration. , The dynamic axle loads are for the front and rear axles.

[0048] Lateral and sideways:

[0049]

[0050] in, Let be the moment of inertia of the vehicle about its vertical axis; , This represents the distance from the center of mass to the front and rear axles, respectively. , This refers to the lateral tire force on the front and rear axles.

[0051] The vehicle dynamics model can be solved using numerical integration methods to solve the dynamic differential equations. The specific steps are as follows: Input the current state of the vehicle (including parameters such as the longitudinal velocity of the vehicle's center of gravity, the lateral velocity of the vehicle's center of gravity, the yaw rate (rad / s), the position of the vehicle's center of gravity in the world coordinate system, and the vehicle's heading angle) and candidate strategies. Iterate cyclically with a small time step Δt (e.g., 10ms-50ms). At each time step, calculate a series of discrete trajectory points within the next T seconds based on the dynamic equations and the tire model. Predict a second predicted trajectory based on the discrete trajectory points. Generate candidate collision parameters based on the first and second predicted trajectories, which is also achieved through the trajectory cross-analysis described above.

[0052] It is understood that the embodiments of this application collect historical trajectory and perception data of the target ahead, and deduce the first predicted trajectory of the target in the future based on this. At the same time, for each candidate strategy planned by the vehicle, the corresponding second predicted trajectory of the vehicle is simulated and generated. Finally, by performing trajectory cross-analysis on the first predicted trajectory of the target and the second predicted trajectory of the vehicle, the specific candidate collision parameters that may occur if the vehicle is driven according to the strategy are calculated, thereby quantifying and evaluating the details of collision risk under each strategy. By performing trajectory cross-analysis on the predicted trajectory of the target ahead and the vehicle's planned trajectory under each candidate strategy, the abstract risk avoidance decision is transformed into a concrete risk indicator. This not only significantly improves the perception granularity and deduction confidence of the decision-making system in complex dynamic game scenarios, but also provides high-fidelity and quantifiable data support for subsequent strategy selection based on the principle of minimizing damage, effectively enhancing the active safety defense capability of the autonomous driving system under extreme conditions.

[0053] In this embodiment of the application, generating candidate collision parameters based on a first predicted trajectory and a second predicted trajectory includes: determining the kinematic features and collision geometry features of the vehicle and the target ahead at the time of the collision based on the first predicted trajectory and the second predicted trajectory, wherein the kinematic features include vehicle motion features and target motion features, and the collision geometry features include vehicle geometry features and target geometry features; and generating candidate collision parameters based on the kinematic features and collision geometry features.

[0054] The kinematic features describe the dynamic state of the vehicle and the target during the collision. Vehicle kinematic features include the predicted relative velocity at the point of impact, the predicted change in the vehicle's velocity, the equivalent energy velocity, and the main force direction of the vehicle during the collision. Target kinematic features include the predicted relative velocity at the point of impact, the predicted change in the target vehicle's velocity, the equivalent energy velocity, and the main force direction of the target during the collision. The collision geometry features describe the spatial relationship between the vehicle and the target during the collision. Vehicle geometry features include the predicted overlap rate during the collision, the type of vehicle collision area (e.g., full width, left offset, right offset, small overlap), the longitudinal angle between the vehicle and the target in front, and the relative position of the collision point in front of the vehicle (e.g., front left, center, front right). Target geometry features include the predicted overlap rate during the collision, the type of target collision area (e.g., full width, left offset, right offset, etc.). The parameters include small overlap, the angle between the longitudinal axis of the target ahead and the vehicle, and the relative position of the collision point in front of the vehicle (e.g., left front, center, right front). The predicted overlap rate is not predicted by the collision damage risk prediction model, but is calculated through trajectory analysis. Specifically, based on the vehicle's current or planned trajectory and the first predicted trajectory, trajectory cross-analysis is performed to calculate the collision angle and overlap rate of possible future collisions between the two vehicles. Candidate collision parameters are feature vectors assembled from kinematic features and collision geometric features in a fixed format. It is understood that, by comparing and analyzing the first predicted trajectory of the target ahead and the second predicted trajectory of the vehicle, the embodiments of this application accurately extract the detailed state of the two at the moment of the assumed collision, including the kinematic features describing the dynamic motion state and the collision geometric features describing the spatial morphological relationship. Subsequently, these multi-dimensional kinematic and geometric features are used as inputs and fused through physical models or algorithms to finally generate a complete set of candidate collision parameters, thereby providing accurate data support for assessing the severity of the collision under this specific strategy.

[0055] In this embodiment of the application, the damage score of each candidate strategy is calculated based on the candidate collision parameters, including: extracting kinematic features and collision geometry features from the candidate collision parameters; obtaining the first attribute information of the vehicle and the second attribute information of the target in front; and calculating the damage score of each candidate strategy based on the kinematic features, collision geometry features, first attribute information and second attribute information.

[0056] The vehicle's first attribute information includes vehicle mass, vehicle type (e.g., sedan, truck), and front-end stiffness rating (determined based on vehicle type according to a preset type-stiffness table). The target's second attribute information includes target type (e.g., sedan, truck), target mass (determined based on target type according to a preset type-mass table), and target front-end stiffness rating (determined based on target type according to a preset type-stiffness table). The injury score is a weighted composite probability that a collision may cause injury to vehicle occupants, predicted based on the collision parameters. The scoring threshold is set according to actual needs. Determining the injury scores of the vehicle and the target at the time of collision based on candidate collision parameters can be achieved through a collision injury risk prediction model, which is pre-built according to needs, for example: This study employs XGBoost (eXtreme Gradient Boosting, an open-source distributed gradient boosting tree model that sequentially constructs multiple decision trees, each learning and correcting the residuals of the previous tree) to achieve high prediction accuracy, strong adaptability to tabular data, fast inference speed (microseconds), and low memory footprint. XGBoost is used to construct 150 decision trees, each with a maximum depth of 5, to balance complexity and generalization ability. Kinematic and geometric features at the time of collision are obtained from a real-world accident database as training samples, along with the corresponding injury scores of both vehicles at the time of the collision. For each collision case, the highest injury level (AIS, Abbreviated InjuryScale) of the two vehicles is extracted, and a comprehensive injury severity score S is defined, for example: S = 1.0 * (AIS≥5) + 0.7 * (AIS=4) + 0.4 * (AIS=3) + 0.1 * (AIS=2), where the weights reflect the social costs of different injury levels and are set according to actual needs, without specific limitations here. The continuous score S is converted into a binary classification training label Y through a threshold (e.g., setting the threshold to 0.2): Y=1 represents high risk (severe overall injury), and Y=0 represents relatively low risk, thus obtaining the ground truth label Y. The collision injury risk prediction model is trained using training samples and ground truth labels until the collision injury risk prediction model meets the accuracy requirements. When a target ahead is detected, the kinematic features, collision geometry features, vehicle's first attribute information, and target's second attribute information at the time of the collision are obtained. The kinematic features, collision geometry features, vehicle's first attribute information, and target's second attribute information are assembled into a feature vector and input into the collision injury risk prediction model. The collision injury risk prediction model outputs candidate collision parameters, and the injury scores of the vehicle and the target ahead at the time of the collision are determined based on the candidate collision parameters.

[0057] It is understood that the embodiments of this application deconstruct key kinematic and collision geometric features from candidate collision parameters, while calling the vehicle's first attribute information (such as body stiffness and safety configuration) and the second attribute information of the target in front (such as target type, mass, and vulnerability). Subsequently, these dynamic collision data and static attribute data are input into the collision injury risk prediction model. By comprehensively calculating the collision energy transfer efficiency, structural compatibility, and occupant protection capability, a specific injury score is generated for each candidate strategy, thereby intuitively reflecting the expected severity of injury or death that may result from implementing the strategy, and providing a decision-making basis for finally selecting the least injury strategy.

[0058] In this embodiment, the damage score of each candidate strategy is calculated based on kinematic features, collision geometry features, first attribute information, and second attribute information. This includes: generating vehicle collision parameters and target collision parameters based on kinematic features and collision geometry features, where the vehicle collision parameters include vehicle motion features and vehicle geometry features, and the target collision parameters include target motion features and target geometry features; calculating the vehicle's damage score based on the first attribute information and vehicle collision parameters; calculating the damage score of the target ahead based on the second attribute information and target collision parameters; and calculating the damage score of the candidate strategy based on the vehicle's damage score and the target's damage score.

[0059] Among them, vehicle collision parameters refer to a subset of data describing the state of a vehicle under a specific collision scenario. They consist of vehicle motion characteristics (such as the vehicle's velocity vector and rate of deceleration at the moment of collision) and vehicle geometric characteristics (such as the location of the vehicle's impact point, the deformation area of ​​the vehicle body, and the definition of the impact angle relative to the vehicle's coordinate system). They are used to assess the risk of damage to the vehicle's structure and the force experienced by the occupants. Target collision parameters refer to a subset of data describing the state of a target ahead under a specific collision scenario. They consist of target motion characteristics (such as the target's relative velocity and mass inertia effect) and target geometric characteristics (such as the impact point of the target, the target's own structural stiffness, or vulnerable areas). They are used to assess the degree of damage to the target object and the risk of injury to its occupants (if it is a vehicle) or the target itself (if it is a pedestrian). The vehicle injury score is a quantitative indicator calculated based on primary attribute information (the vehicle's safety features, body strength, etc.) and vehicle collision parameters. It specifically reflects the severity of injuries (such as the probability of serious injury and AIS level) that the driver and passengers of the vehicle may suffer in the collision. The forward target injury score is a quantitative indicator calculated based on secondary attribute information (target type, mass, vulnerability, etc.) and target collision parameters. It specifically reflects the severity of injuries that forward targets (such as their occupants or pedestrians) may suffer in the collision. The candidate strategy injury score is the final comprehensive risk assessment value, which is obtained by weighted fusion of the vehicle injury score and the forward target injury score. It represents the total severity of the overall accident consequences after implementing the candidate strategy.

[0060] It is understood that the embodiments of this application decompose general kinematic and geometric features into specific vehicle collision parameters and target collision parameters; then, the vehicle injury score specific to the occupants of the vehicle is calculated by combining the first attribute information of the vehicle, and the front target injury score specific to the front target is calculated by combining the second attribute information of the target and the target collision parameters; finally, the two independent injury scores are weighted to generate the final injury score of the candidate strategy, thereby realizing the independent assessment and overall quantification of the potential injury risk of both parties in the accident, and ensuring that the safety of the vehicle and others can be taken into account when making decisions.

[0061] In step S103, the target strategy is determined from multiple candidate strategies according to the damage score of the candidate strategies, and at least one of the braking action and the steering action is controlled based on the target strategy to execute on the vehicle.

[0062] Among them, the braking action refers to the control behavior of the vehicle applying braking force through a braking system (such as an AEB system) to generate a negative acceleration of the vehicle to reduce the vehicle speed or completely stop, aiming to reduce the collision energy or avoid collisions; the steering action refers to the control behavior of the vehicle changing the front wheel angle through a steering system (such as an EPS <Electric Power Steering, electric power steering system>) to adjust the vehicle driving trajectory to avoid obstacles or change the collision angle.

[0063] It can be understood that after obtaining the damage scores of all candidate strategies in the embodiment of the present application, the damage scores corresponding to all candidate strategies are compared, and following the principle of minimizing damage, the plan with the lightest expected consequence is selected as the target strategy; subsequently, the underlying control system of the vehicle immediately executes at least one of the braking action and the steering action, or a coordinated combination of both, according to the specific instructions of the target strategy, so as to implement risk avoidance at the physical level and strive to minimize the potential losses of traffic accidents.

[0064] In the embodiment of the present application, determining the target strategy from multiple candidate strategies according to the damage score of the candidate strategies includes: extracting the braking parameters and steering parameters of the vehicle in the candidate strategies; determining the stability score and path deviation score of the vehicle executing the candidate strategy according to the braking parameters and steering parameters; calculating the comprehensive score of the candidate strategy according to the damage score, stability score and path deviation score of the candidate strategy, and determining the target strategy from multiple candidate strategies according to the comprehensive score.

[0065] Among them, the path deviation cost can be measured by the integral or maximum value of the lateral offset distance between the second predicted trajectory calculated by the braking parameters and steering parameters and the reference line (such as the lane center line, specifically set according to actual needs); the stability risk is mainly based on the vehicle dynamics state and is measured by the following indicators: calculating the sideslip angle of the center of mass and its change rate through the braking parameters and steering parameters, and if the sideslip angle of the center of mass and its change rate exceed the threshold, it indicates instability; judging whether the yaw rate follows the steering wheel angle through the braking parameters and steering parameters to determine understeer / oversteer.

[0066] Understandably, the embodiments of this application first extract specific braking and steering parameters from each candidate strategy, and calculate the stability score and path deviation score when executing the strategy based on the braking and steering parameters respectively; then, the originally calculated damage score is weighted and fused with the two newly added handling scores to generate a comprehensive score that reflects the overall quality of risk avoidance; finally, it realizes that it no longer only relies on the potential damage size, but also on this comprehensive score that takes into account personnel safety, vehicle stability and path legality, to lock in and output the final target strategy from multiple candidate solutions, ensuring that the vehicle can effectively reduce the consequences of the accident during emergency avoidance, and avoid loss of vehicle control or secondary accidents due to excessive operation.

[0067] In this embodiment of the application, before determining the target strategy from multiple candidate strategies based on the comprehensive score, the method further includes: determining the vehicle's control cost score based on braking parameters and steering parameters; and correcting the comprehensive score based on the control cost score.

[0068] Among them, the control cost score is an indicator used to quantify the degree of control intensity or comfort loss required to execute a specific strategy. This score not only focuses on whether the vehicle can complete the task, but also on whether the task execution process is smooth. It is usually proportional to the abruptness of the action, that is, the more intense the action and the faster the change, the higher the control cost score, that is, the heavier the penalty. The braking smoothness dimension is measured by the rate of change of braking deceleration. If the braking force suddenly increases from 0 to the maximum, the rate of change is extremely high, and the control cost will increase significantly, which will cause the occupants to be thrown forward uncomfortably or even injured. The steering smoothness dimension is measured by the steering angular velocity (the speed at which the steering wheel or wheels turn). If the vehicle needs to complete a large-angle turn in a very short time, the extremely high angular velocity will not only make the occupants feel strong centrifugal force discomfort, but may also exceed the physical response limit of the EPS (Electric Power Steering) actuator.

[0069] The overall score is adjusted based on the control cost score through a comprehensive cost function, which is a tool that quantifies multiple security objectives into a single scalar for objectively comparing the merits of different candidate strategies. For each candidate strategy... Its total cost Defined as:

[0070] in, The predicted injury score measures the severity of injury to both occupants in a predicted collision under this strategy. The path deviation score measures the degree to which the strategy causes the vehicle trajectory to deviate from the predetermined path. The stability score measures the risk of the vehicle losing stability while implementing the strategy. To control the cost factor, the severity of braking and steering actions is measured for smooth control.

[0071] : The adaptive weight coefficients corresponding to each sub-item, satisfying The weighting coefficients are adaptively modified based on the risk: when the collision risk is high and the confidence level is strong, The highest weight is given to minimizing damage. When safety boundaries take precedence: when the vehicle is approaching instability or about to deviate from its lane, the weight is dynamically increased. or To prevent secondary accidents from occurring.

[0072] Optimal Strategy That is, the strategy that minimizes the total cost:

[0073] For example, suppose the scenario is: the car is traveling at 100km / h and the car in front brakes suddenly, making a collision unavoidable. The multi-strategy generator generates three candidate strategies, as shown in Table 1. Table 1 is a table of candidate strategies and their corresponding scores.

[0074] Table 1

[0075] Calculate the total cost:

[0076]

[0077]

[0078] Among them, strategy C has a larger path deviation score D, which is close to the lane boundary; The score is the lowest, therefore, strategy B (forced movement + slight left turn) is the optimal strategy.

[0079] Understandably, this application's embodiments introduce a control cost score before finally locking in the target strategy: First, based on the braking and steering parameters in the candidate strategies, the control cost score is calculated, focusing on the rate of change of braking deceleration (measuring whether braking is abrupt) and steering angular velocity (measuring whether steering is too sharp), to quantify the severity of the action and the potential impact on occupant comfort; then, the original comprehensive score is corrected using this cost score (e.g., deducting corresponding points), thereby penalizing extreme strategies that, while reducing collision damage, require actuators to work under excessive load or cause extreme discomfort to occupants; finally, the target strategy is determined based on this comprehensive score after smoothness correction using the control cost score, ensuring that when the vehicle performs avoidance actions, it can not only effectively reduce accident damage and maintain stability, but also take into account the comfort of occupants and the physical feasibility of actuators, avoiding unbearable violent maneuvers in pursuit of theoretically minimal damage.

[0080] In summary, this application's embodiments acquire multiple candidate strategies, including the vehicle's braking and steering parameters; predict the injury score of each candidate strategy when the vehicle collides with a target ahead; determine a target strategy from the multiple candidate strategies based on the injury scores of the candidate strategies; and control the vehicle to perform at least one braking or steering action based on the target strategy. This method quantitatively evaluates and optimizes based on the injury score at the time of collision, overcoming the limitations of traditional automatic emergency braking that relies solely on full braking and ignores collision patterns. In scenarios where collisions are unavoidable, it actively avoids collision patterns with high overlap and high injury risk, selecting the optimal avoidance trajectory with the lowest risk of occupant injury. This significantly improves collision compatibility while reducing relative speed, minimizing severe injuries to occupants in collision accidents.

[0081] The vehicle collision control method of this application will be further described below through a specific embodiment.

[0082] Assume your vehicle (a mid-size SUV, approximately 2000 kg) is traveling at a constant speed of 100 km / h in the fast lane of a highway. A sedan (the target vehicle, approximately 1500 kg) is traveling at a similar speed in the same lane ahead. A large container truck is traveling parallel to you in the slow lane to the right. Suddenly, the target vehicle brakes suddenly due to an obstacle ahead (such as a fallen object), decelerating drastically. Your driver, distracted, fails to react in time. By the time your vehicle's sensors detect the sudden deceleration of the target vehicle, the distance between the two vehicles is only about 30 meters, and the relative speed difference is rapidly increasing. Even with a traditional AEB system applying full braking force within this extremely short distance, a collision is still unavoidable, and a high-speed rear-end collision could result in severe structural intrusion and occupant injuries.

[0083] Millimeter-wave radar detects the rapidly increasing relative speed of the vehicle ahead, while a camera simultaneously confirms that the vehicle's brake lights are illuminated and its outline is rapidly enlarging. LiDAR provides high-precision distance and outline change data. The vehicle's current speed is 100 km / h, and the trajectory prediction model, based on the target vehicle's historical trajectory (rapid deceleration), the absence of other obstacles ahead, and the highway scenario, predicts its future trajectory as: continuing to decelerate within the current lane until near a stop.

[0084] Based on the current trajectory of the self-vehicle (assuming no driver intervention) and the predicted trajectory of the target vehicle, it is immediately calculated that approximately 1.1 seconds later, the self-vehicle will rear-end the nearly stationary target vehicle ahead at a high speed, with the point of collision located behind the target vehicle.

[0085] The simplified collision model quickly estimates a high predicted speed change based on the predicted relative collision speed (initially calculated to be close to 100 km / h, but dynamically changes over time as the target vehicle decelerates), the mass of the two vehicles, and the geometry of the rear-end collision (the frontal rigid collision area is large).

[0086] The collision injury risk prediction model receives the above parameters, combines the rear-end collision type (the vehicle is the attacker, but its occupants also face a high deceleration impact risk) and the collision compatibility information of the two vehicles, and outputs a quantified initial predicted injury risk value of "high risk" (posing a high risk to the occupants of both vehicles). The calculation is based on dynamically compressed short distances, high relative speeds, and the vehicle's physical braking limits. The judgment result shows that even if the vehicle brakes at its maximum capacity (assuming -1.0g) instantaneously, a collision cannot be avoided within the remaining distance and time. The system determines that "collision is unavoidable" and immediately activates the collision optimization mode.

[0087] The multi-policy generator considers the current situation where there are trucks running parallel to each other on the right and a central guardrail on the left. In addition to the baseline policy of "full braking," it generates a limited but critical set of candidate policies: Strategy A (Baseline Strategy): Full braking (-1.0g), no steering. Predicted outcome: High-speed head-on rear-end collision.

[0088] Strategy B: Full braking (-1.0g) + very slight right turn (0.3 degrees, attempting to utilize the minimal clearance that may exist between the vehicle and the truck, but the main purpose is to slightly alter the collision overlap area).

[0089] Strategy C: Strong braking (-0.8g) + slight, stable left steering (0.8 degrees, causing the vehicle to slightly adjust to the left, intending to create a small-angle offset rear-end collision).

[0090] (Note: All strategies are subject to constraints. Right turn strategies must not encroach on the truck's safety boundary; left turn strategies must not be excessive and result in scraping against the central guardrail.)

[0091] Simulation evaluation of all strategies: Strategy A: High-speed head-on rear-end collision, large overlap area, collision force is directly and completely transmitted through the front longitudinal beam, which places extremely high demands on the energy absorption structure of the vehicle's front compartment and poses a high risk of intrusion into the rear of the target vehicle.

[0092] Strategy B: Under full braking, a slight right turn is unlikely to substantially change the vehicle's position before a collision, and the risk and cost of encroaching on the truck lane increase sharply. Therefore, this strategy was quickly downgraded due to stability and path risk.

[0093] Strategy C: The vehicle decelerates sharply while smoothly veering to the left. Simulation predictions indicate that upon collision, the vehicular vehicle will collide with the target vehicle's right rear corner at an angle, with approximately 40% of the overlap area of ​​its front bumper on the left side. Simultaneously, due to the slightly prolonged approach time caused by steering and the fact that braking did not reach its limit, the final relative collision speed is slightly higher than under Strategy A, but the collision geometry is fundamentally altered.

[0094] Injury Risk Assessment: The collision injury risk prediction model assesses new collision patterns (offset collisions). For the vehicle itself: The collision force is mainly transmitted through the left front longitudinal beam (the right longitudinal beam can still participate in some energy absorption), avoiding the high acceleration impact risk in a head-on collision. The model assesses that the risk of injury to the vehicle's occupants has decreased from "high risk" to "medium risk".

[0095] For the target vehicle: An oblique collision reduces the concentrated force acting directly behind the rear axle of the target vehicle, which may reduce the extent to which the trunk is "deformed by impact" and intrudes into the rear passenger compartment, and also reduces the risk of injury to the target vehicle occupants.

[0096] Stability risks: Strategy C has a very small steering angle, and under strong deceleration, ESP can easily maintain the vehicle's attitude, with no risk of rollover or fishtailing.

[0097] Path deviation cost: A slight leftward adjustment causes the vehicle to approach but not collide with the central guardrail, resulting in a controllable path deviation cost.

[0098] Optimal Strategy Selection: The optimal strategy selector performs comprehensive calculations. Strategy A (frontal collision) has the highest risk of injury. Strategy B is excluded due to its high risk. Although Strategy C is slightly inferior in absolute speed, it minimizes the sum of the overall risk of injury to both vehicles by cleverly changing the collision mode. Therefore, the system selects Strategy C as the optimal collision mitigation strategy.

[0099] The chassis domain controller instantly sends a command to the ESP to apply a strong deceleration of 0.8g (not the limit value), and at the same time requests the EPS to stably execute a left steering angle of 0.8 degrees, thereby optimizing the collision strategy.

[0100] Next, the vehicle collision control device proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0101] Figure 4 This is a schematic diagram of a vehicle collision control device according to an embodiment of this application.

[0102] like Figure 4 As shown, the vehicle collision control device includes: an acquisition module 201, a prediction module 202, and an execution module 203.

[0103] The acquisition module 201 is used to acquire multiple candidate strategies, including the vehicle's braking parameters and steering parameters; the prediction module 202 is used to predict the damage score of each candidate strategy when the vehicle collides with the target in front; the execution module 203 is used to determine the target strategy from multiple candidate strategies based on the damage score of the candidate strategies, and control the vehicle to perform at least one braking action and steering action based on the target strategy.

[0104] In this embodiment of the application, the prediction module 202 is further configured to: predict candidate collision parameters between the vehicle and the target ahead based on the candidate strategies; and calculate the damage score for each candidate strategy based on the candidate collision parameters.

[0105] In this embodiment, the prediction module 202 is further configured to: acquire the historical trajectory and perception data of the target ahead; generate a first predicted trajectory of the target ahead based on the historical trajectory and perception data; generate a second predicted trajectory of the vehicle based on each candidate strategy; and generate candidate collision parameters based on the first and second predicted trajectories.

[0106] In this embodiment of the application, the prediction module 202 is further configured to: determine the kinematic features and collision geometry features of the vehicle and the target ahead at the time of the collision based on the first prediction trajectory and the second prediction trajectory, wherein the kinematic features include vehicle motion features and target motion features, and the collision geometry features include vehicle geometry features and target geometry features; and generate candidate collision parameters based on the kinematic features and collision geometry features.

[0107] In this embodiment, the prediction module 202 is further configured to: extract kinematic features and collision geometry features from candidate collision parameters; obtain the first attribute information of the vehicle and the second attribute information of the target in front; and calculate the damage score of each candidate strategy based on the kinematic features, collision geometry features, first attribute information and second attribute information.

[0108] In this embodiment, the prediction module 202 is further configured to: generate vehicle collision parameters and target collision parameters based on kinematic features and collision geometry features, wherein the vehicle collision parameters include vehicle motion features and vehicle geometry features, and the target collision parameters include target motion features and target geometry features; calculate the vehicle's damage score based on first attribute information and vehicle collision parameters; calculate the damage score of the target ahead based on second attribute information and target collision parameters; and calculate the damage score of the candidate strategy based on the vehicle's damage score and the target ahead's damage score.

[0109] In this embodiment, the acquisition module 201 is further configured to: acquire the historical trajectory and perception data of the target ahead before acquiring multiple candidate strategies; generate a first predicted trajectory of the target ahead based on the historical trajectory and perception data, and generate a vehicle driving trajectory based on the current state of the vehicle; determine the first collision position and the first collision time between the vehicle and the target obstacle based on the first predicted trajectory and the vehicle driving trajectory, and calculate the first relative distance between the current position of the vehicle and the first collision position; calculate the braking completion time of the vehicle based on the maximum braking deceleration of the vehicle, and if the braking completion time is after the first collision time, acquire multiple candidate strategies; if the braking completion time is before the first collision time, directly trigger braking.

[0110] In this embodiment, the execution module 203 is further configured to: extract the vehicle's braking parameters and steering parameters from the candidate strategies; determine the stability score and path deviation score of the vehicle executing the candidate strategies based on the braking parameters and steering parameters; calculate the comprehensive score of the candidate strategies based on the damage score, stability score, and path deviation score of the candidate strategies; and determine the target strategy from multiple candidate strategies based on the comprehensive score.

[0111] In this embodiment, the execution module 203 is further configured to: determine the vehicle's control cost score based on braking parameters and steering parameters before determining the target strategy from multiple candidate strategies based on the comprehensive score; and correct the comprehensive score based on the control cost score.

[0112] The vehicle collision control device proposed in this application acquires multiple candidate strategies, including the vehicle's braking and steering parameters; predicts the injury score of each candidate strategy when the vehicle collides with a target ahead; determines a target strategy from the multiple candidate strategies based on the injury scores of the candidate strategies; and controls the vehicle to perform at least one braking or steering action based on the target strategy. This method quantitatively evaluates and optimizes based on the injury score at the time of collision, overcoming the limitations of traditional automatic emergency braking that relies solely on full braking and ignores the collision pattern. In scenarios where collision is unavoidable, it actively avoids collision patterns with high overlap and high injury risk, selecting the optimal avoidance trajectory with the lowest risk of occupant injury. This significantly improves collision compatibility while reducing relative speed, minimizing severe injuries to occupants in collision accidents.

[0113] It should be noted that the foregoing explanation of the vehicle collision control method embodiment also applies to the vehicle collision control device of this embodiment, and will not be repeated here.

[0114] This application also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described vehicle collision control method.

[0115] Figure 5 A schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle may include: The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0116] When the processor 302 executes the program, it implements the drift method provided in the above embodiments.

[0117] Furthermore, the vehicle also includes: Communication interface 303 is used for communication between memory 301 and processor 302.

[0118] The memory 301 is used to store computer programs that can run on the processor 302.

[0119] The memory 301 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0120] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0121] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0122] Processor 302 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0123] The vehicle in this application embodiment may also include the aforementioned electronic equipment.

[0124] This application also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed, implements the vehicle collision control method described above.

[0125] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0126] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0127] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A vehicle collision control method, characterized in that, include: Multiple candidate strategies are obtained, including vehicle braking parameters and steering parameters; Predict the damage score for each candidate strategy when the vehicle collides with the target ahead; A target strategy is determined from a plurality of candidate strategies based on the damage score of the candidate strategies, and the vehicle is controlled to perform at least one of braking and steering actions based on the target strategy.

2. The vehicle collision control method according to claim 1, characterized in that, The prediction of the damage score for each candidate strategy when the vehicle collides with the target ahead includes: Predict candidate collision parameters between the vehicle and the target ahead based on the candidate strategy; The damage score for each candidate strategy is calculated based on the candidate collision parameters.

3. The vehicle collision control method according to claim 2, characterized in that, The step of predicting candidate collision parameters between the vehicle and the target ahead based on the candidate strategy includes: Acquire the historical trajectory and perception data of the target ahead; A first predicted trajectory of the target ahead is generated based on the historical trajectory and the sensing data; A second predicted trajectory for the vehicle is generated based on each of the candidate strategies; The candidate collision parameters are generated based on the first predicted trajectory and the second predicted trajectory.

4. The vehicle collision control method according to claim 3, characterized in that, The step of generating the candidate collision parameters based on the first predicted trajectory and the second predicted trajectory includes: Based on the first predicted trajectory and the second predicted trajectory, the kinematic features and collision geometry features of the vehicle and the target ahead at the time of the collision are determined, wherein the kinematic features include vehicle motion features and target motion features, and the collision geometry features include vehicle geometry features and target geometry features; The candidate collision parameters are generated based on the kinematic features and the collision geometry features.

5. The vehicle collision control method according to claim 4, characterized in that, The step of calculating the damage score for each candidate strategy based on the candidate collision parameters includes: Extract the kinematic features and collision geometric features from the candidate collision parameters; Obtain the first attribute information of the vehicle and the second attribute information of the target ahead; The damage score for each candidate strategy is calculated based on the kinematic features, the collision geometry features, the first attribute information, and the second attribute information.

6. The vehicle collision control method according to claim 5, characterized in that, The step of calculating the damage score for each candidate strategy based on the kinematic features, the collision geometry features, the first attribute information, and the second attribute information includes: Vehicle collision parameters and target collision parameters are generated based on the kinematic features and the collision geometry features. The vehicle collision parameters include the vehicle kinematic features and the vehicle geometry features, and the target collision parameters include the target kinematic features and the target geometry features. The damage score of the vehicle is calculated based on the first attribute information and the vehicle collision parameters; The damage score of the target in front is calculated based on the second attribute information and the target collision parameters; The damage score of the candidate strategy is calculated based on the damage score of the vehicle and the damage score of the target in front.

7. The vehicle collision control method according to any one of claims 1-6, characterized in that, Before obtaining multiple candidate strategies, the following is also included: Acquire the historical trajectory and perception data of the target ahead; generate a first predicted trajectory of the target ahead based on the historical trajectory and perception data; generate the vehicle's driving trajectory based on the current state of the vehicle. Based on the first predicted trajectory and the vehicle's driving trajectory, determine the first collision position and the first collision time between the vehicle and the target obstacle, and calculate the first relative distance between the vehicle's current position and the first collision position; The braking completion time of the vehicle is calculated based on the vehicle's maximum braking deceleration. If the braking completion time is after the first collision time, multiple candidate strategies are obtained; if the braking completion time is before the first collision time, braking is triggered directly.

8. The vehicle collision control method according to claim 1, characterized in that, The step of determining the target strategy from multiple candidate strategies based on the damage score of the candidate strategies includes: Extract the vehicle's braking and steering parameters from the candidate strategies; Based on the braking parameters and the steering parameters, determine the stability score and path deviation score of the vehicle executing the candidate strategy; Based on the damage score, stability score, and path deviation score of the candidate strategies, a comprehensive score is calculated for each candidate strategy, and a target strategy is determined from multiple candidate strategies based on the comprehensive score.

9. The vehicle collision control method according to claim 8, characterized in that, Before determining the target strategy from the multiple candidate strategies based on the comprehensive score, the method further includes: The control cost score of the vehicle is determined based on the braking parameters and the steering parameters; The overall score is adjusted based on the control cost score.

10. A vehicle, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the vehicle collision control method according to any one of claims 1-9.

Images