Collision pre-mitigation control method and device, vehicle and medium

Abstract

The invention relates to the technical field of vehicle safety control, and discloses a collision pre-mitigation control method and device, a vehicle and a medium, and the method comprises the steps: obtaining vehicle state data, environment perception data and in-vehicle passenger data of a target vehicle; judging whether the target vehicle has a collision risk or not based on the vehicle state data and the environment sensing data, and determining a current collision type when the target vehicle has the collision risk; a personalized passenger model matched with the current passenger state is constructed based on the passenger data in the vehicle; based on the state data of the vehicle, the current collision type and the personalized passenger model, collision pre-relieving control instructions corresponding to a plurality of vehicle execution systems for the target vehicle are generated, and the vehicle execution systems at least comprise a safety belt restraint system, a seat and airbag system, an active suspension system and a driving brake system; and each vehicle execution system is controlled to execute corresponding actions according to the collision pre-mitigation control instruction, so that effective collision pre-mitigation control is realized, and the safety of passengers is greatly guaranteed.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety control technology, specifically to a collision pre-mitigation control method, device, vehicle, and medium. Background Technology

[0002] With the continuous growth of car ownership and the increasing complexity of the traffic environment, road traffic safety remains a core issue of social concern. Currently, vehicle safety technology has evolved from passive safety to active safety. Traditional passive vehicle safety systems, such as airbags and seatbelt pretensioners, primarily function after a collision. Their triggering conditions and protection strength are mostly set with fixed parameters, making dynamic adjustments impossible based on specific collision scenarios, collision intensity, and individual occupant characteristics. This approach may result in insufficient protection under certain conditions, or even secondary injuries to occupants due to improper force application.

[0003] In the field of active safety, existing Forward Collision Warning (FCW) and Autonomous Emergency Braking (AEB) systems primarily focus on avoiding collisions. However, for collisions that cannot be completely avoided (i.e., "pre-collision" or "unavoidable collision" scenarios), as well as low-intensity, recoverable minor collisions that have already occurred, the system's response strategies remain simplistic and limited. For example, while full braking by an AEB system reduces vehicle speed, it may also alter the vehicle's attitude and the occupants' seating positions. Traditional restraint systems have not been optimized for this new dynamic state, resulting in suboptimal protection.

[0004] Furthermore, modern vehicle control systems, such as suspension, drive, braking, and occupant restraint systems, are typically managed independently by different electronic control units, lacking deep coordination under critical conditions. In collision hazards, if these systems could be coordinated and optimized through active suspension adjustments, axle load transfer, and other means to improve the vehicle's collision posture and energy absorption path, and linked with the occupant restraint system, theoretically, overall vehicle safety performance could be significantly improved. However, achieving such real-time coordinated control across multiple systems, multiple objectives, and strong coupling faces significant challenges, including complex models and difficulties in decision optimization.

[0005] In recent years, the rapid development of artificial intelligence has led to its widespread application in many fields. For example, reinforcement learning (RL) possesses powerful sequential decision-making and online optimization capabilities, demonstrating great potential in solving complex dynamic system control problems. Meanwhile, the development of vehicle-to-everything (V2X) technology, such as vehicle-to-everything (V2X) technology, enables vehicles to acquire beyond-line-of-sight perception capabilities, such as precise information about other vehicles and pedestrians on the road ahead, as well as traffic conditions provided by roadside units. This allows for the aggregation of massive amounts of scene data through cloud platforms for model training. However, how to deeply integrate the intelligent decision-making capabilities of artificial intelligence, the real-time information interaction capabilities of V2X, and the centralized computing and optimization capabilities of the cloud, and apply them to collaborative control scenarios before and during minor collisions to form a self-learning, continuously evolving adaptive safety protection system, remains a pressing technical challenge in this field. Summary of the Invention

[0006] This invention provides a collision pre-mitigation control method, device, vehicle, and medium to address the problem mentioned in the above-mentioned technical background that the prior art is unable to achieve vehicle-cloud coordination and adaptive, multi-system linkage integrated pre-mitigation control before a vehicle collision or after a minor collision, thus seriously affecting occupant safety.

[0007] In a first aspect, the present invention provides a collision pre-mitigation control method, the method comprising: Acquire the target vehicle's own status data, environmental perception data, and occupant data; Based on the vehicle's status data and environmental perception data, determine whether the target vehicle poses a collision risk, and if the target vehicle poses a collision risk, determine the current collision type; A personalized occupant model is built based on in-vehicle occupant data to adapt to the current occupant status; Based on the vehicle status data, the current collision type, and the personalized occupant model, collision pre-mitigation control commands are generated for multiple vehicle execution systems corresponding to the target vehicle. The vehicle execution systems include at least the seat belt restraint system, the seat and airbag system, the active suspension system, and the drive braking system. The system controls each vehicle's execution system to perform corresponding actions based on the collision pre-mitigation control command.

[0008] This invention pre-judgments collision risks and types based on vehicle status and environmental perception data, enabling proactive pre-collision intervention. This helps reduce the severity of collisions and minimizes injury to occupants. It is also adaptable to complex scenarios involving different vehicle driving conditions, occupant combinations, and collision types. Through safety linkage with seatbelt restraint systems, seat and airbag systems, active suspension systems, and drive and braking systems, it generates corresponding collision pre-mitigation control commands for each vehicle's execution system and controls each system to perform corresponding actions. This achieves proactive and comprehensive safety protection against vehicle collisions, effectively improving collision pre-mitigation control efficiency before or after a minor collision, significantly ensuring the safety of vehicle occupants.

[0009] In one optional implementation, the in-vehicle occupant data includes at least the occupant's identity information, weight data, age information, seatbelt wearing status, current sitting posture data, and the relative distance between the occupant and in-vehicle components; based on the in-vehicle occupant data, a personalized occupant model adapted to the current occupant status is constructed, including: Extract the current occupant's identity information from the vehicle occupant data; Determine whether a personalized passenger model exists that matches the identity information; If a personalized occupant model that matches the identity information exists, the personalized occupant model is obtained, and the personalized occupant model is updated using the current occupant's weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and the vehicle interior components. If no personalized occupant model matches the identity information, a biomechanical model representing the current occupant's state is constructed based on the occupant's identity information, weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and in-vehicle components, thus obtaining a personalized occupant model.

[0010] This invention uses identity information as an index to prioritize the reuse of existing personalized occupant models, significantly reducing the computational cost of repetitive modeling. For occupants without historical models, a new model is created in real time based on multi-dimensional occupant data. For historical personalized occupant models with matching data, the model is updated using currently collected data such as weight, posture, and seatbelt wearing status. This avoids discrepancies between the model and reality caused by changes in occupant posture or temporary occupant replacement, ensuring that the model is always synchronized with the current occupant status. While ensuring the timeliness of protection, this greatly improves the accuracy of modeling, thus providing data support for subsequent collision pre-mitigation control.

[0011] In one optional implementation, based on the vehicle's state data, the current collision type, and a personalized occupant model, multiple collision pre-mitigation control commands corresponding to the target vehicle's vehicle execution systems are generated, including: The current occupant status is determined based on a personalized occupant model; The vehicle-cloud collaborative decision-making model is acquired, and the vehicle's status data, current collision type, and current occupant status are input into the vehicle-cloud collaborative decision-making model. The model outputs a first control command for the seat belt restraint system, a second control command for the seat and airbag system, a third control command for the active suspension system, and a fourth control command for the drive and braking system. The vehicle-cloud collaborative decision-making model is obtained by iteratively training and updating the reinforcement learning model using the vehicle's historical control data.

[0012] This invention uses vehicle status data, current collision type, and personalized occupant status as joint inputs to ensure that the output control commands not only conform to the real-time operating conditions of the vehicle but also match the individual protection needs of the occupants, effectively improving the pertinence of protective actions. The vehicle-cloud collaborative decision-making model is built based on reinforcement learning and is continuously updated through the accumulation of historical data. It can autonomously learn the optimal control logic in different scenarios, and thus generate exclusive and high-precision control commands for vehicle execution systems such as seat belt restraint, seats and airbags, active suspension, and drive braking to achieve comprehensive and effective vehicle collision pre-mitigation control.

[0013] In one alternative implementation, controlling each vehicle actuator to perform corresponding actions according to collision pre-mitigation control commands includes: The seat belt restraint system controls the pretension and limiting force of the seat belt in stages according to the first control command. The pretension and limiting force of the seat belt are at least related to the occupant's weight data; wherein the corresponding values ​​of the pretension and limiting force are positively correlated with the size of the weight data.

[0014] This invention improves adaptability to different occupant types by classifying and controlling the pretension and limiting force of seat belts, thereby helping to meet the needs of various scenarios such as family cars and shared cars. Furthermore, by directly and positively linking the pretension and limiting force of seat belts with occupant weight, it avoids excessive restraint on light occupants such as children and slender individuals, while ensuring sufficient restraint strength for heavy occupants. This prevents excessive body displacement during a collision, balancing the effectiveness of protection and the safety of use, and to a certain extent improving the efficiency of collision pre-mitigation, thereby ensuring occupant safety.

[0015] In one alternative implementation, controlling each vehicle actuator to perform corresponding actions according to collision pre-mitigation control commands includes: The control seat and airbag system coordinately regulate the seat angle, as well as the triggering timing and inflation intensity of the airbag, according to the second control command; wherein, the regulation of the seat angle is related to the current collision type; the regulation of the triggering timing and inflation intensity of the airbag is related to at least one of the following: occupant identity information, weight data, age information, seat belt wearing status, current sitting posture data, relative distance between the occupant and in-vehicle components, and the current collision type.

[0016] This invention uses a second control command to coordinate the control of the seat and airbag, which can significantly improve the overall protection effect. At the same time, the seat angle adjustment is designed to be directly related to the current collision type. It can adjust the occupant's sitting posture in advance according to the force characteristics of different collision scenarios such as frontal, side, and rear-end collisions. This reduces the impact displacement of the occupant's body during the collision and provides the optimal angle for airbag contact, thereby reducing personal injury caused by local pressure. Furthermore, the airbag control is designed to integrate multi-dimensional data such as the occupant's identity, weight, sitting posture, and collision type, which can accurately match the protection tolerance of different occupants and greatly ensure the personal safety of the occupants.

[0017] In one alternative implementation, controlling each vehicle actuator to perform corresponding actions according to collision pre-mitigation control commands includes: The active suspension system adjusts the height and stiffness of the suspension according to the third control command, and the adjustment of the height and stiffness of the suspension is related to the current collision type.

[0018] This invention precisely adjusts the suspension height and stiffness to address the impact characteristics of different collision types, such as frontal, side, and rear-end collisions. It can actively shape the vehicle body's stress posture, thereby maximizing the protective potential of the vehicle body structure, effectively reducing the risk of passenger compartment deformation, and greatly ensuring the personal safety of occupants.

[0019] In one alternative implementation, controlling each vehicle actuator to perform corresponding actions according to collision pre-mitigation control commands includes: The control drive braking system coordinates the load distribution of drive and braking according to the fourth control command, and the coordinated control of load distribution is associated with the current collision type.

[0020] This invention addresses the force characteristics of different collision types, such as frontal, side, and rear-end collisions. By coordinating the control of driving torque and braking force, it actively transfers the load on the front and rear axles or left and right axles of the vehicle, which can actively optimize the load distribution, allowing the energy-absorbing structure of the vehicle body to play its full role. This reduces the intrusion into the passenger compartment, greatly reduces the impact intensity of the collision, and ensures the safety of the occupants.

[0021] In a second aspect, the present invention provides a collision pre-mitigation control device, the device comprising: The data acquisition module is used to acquire the target vehicle's own status data, environmental perception data, and occupant data; The collision detection module is used to determine whether there is a collision risk to the target vehicle based on the vehicle's status data and environmental perception data, and to determine the current collision type when there is a collision risk to the target vehicle. The state modeling module is used to build personalized occupant models that adapt to the current occupant state based on in-vehicle occupant data. The instruction generation module is used to generate collision pre-mitigation control instructions for multiple vehicle execution systems of the target vehicle based on the vehicle status data, the current collision type and the personalized occupant model. The vehicle execution systems include at least the seat belt restraint system, the seat and airbag system, the active suspension system and the drive braking system. The action execution module is used to control the various vehicle execution systems to perform corresponding actions according to the collision pre-mitigation control commands.

[0022] The collision pre-mitigation control device of this invention utilizes vehicle status and environmental perception data to predict collision risks and types in advance, enabling proactive pre-collision intervention. This helps reduce the severity of collision damage and minimizes injury to occupants. Simultaneously, it adapts to complex scenarios involving different vehicle driving conditions, occupant combinations, and collision types. Through safety linkage with the seatbelt restraint system, seat and airbag system, active suspension system, and drive braking system, it generates corresponding collision pre-mitigation control commands for each vehicle execution system and controls each system to perform corresponding actions. This achieves proactive and comprehensive safety protection against vehicle collisions, effectively improving the efficiency of pre-collision mitigation control before or after a minor collision, significantly ensuring the personal safety of vehicle occupants.

[0023] Thirdly, the present invention provides a vehicle, the vehicle including a controller, the controller including a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the collision pre-mitigation control method of the first aspect or any corresponding embodiment described above.

[0024] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the collision pre-mitigation control method of the first aspect or any corresponding embodiment thereof.

[0025] The collision pre-mitigation control method and device provided by this invention judges the collision risk and type based on the vehicle's state and environmental perception data, which helps to reduce the degree of collision damage and realizes proactive pre-collision intervention. By constructing a personalized occupant model adapted to the current occupant state, it can tailor the most suitable seat belt restraint force, airbag deployment parameters, etc. for different occupants, effectively protecting them while minimizing secondary injuries caused by improper deployment of protective devices, thus achieving truly personalized occupant protection. At the same time, it can be adapted to various scenarios and achieves proactive and comprehensive protection against vehicle collisions through safety linkage with various vehicle execution systems. It can effectively improve the efficiency of vehicle collision pre-mitigation control before a collision or after a minor collision, greatly ensuring the personal safety of occupants. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a schematic flowchart of a first type of collision pre-mitigation control method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a second process of the collision pre-mitigation control method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a vehicle-cloud coordinated adaptive collision pre-mitigation control system based on reinforcement learning; Figure 4 This is a flowchart illustrating the vehicle-cloud coordinated adaptive collision pre-mitigation control method based on reinforcement learning. Figure 5 This is a schematic diagram of the structure of reinforcement learning; Figure 6 This is a schematic diagram illustrating differentiated control of seat belts, airbags, and load transfer for different occupants; Figure 7 This is a structural block diagram of a collision pre-mitigation control device according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the hardware structure of a vehicle according to an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] According to an embodiment of the present invention, a collision pre-mitigation control method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] This embodiment provides a collision pre-mitigation control method. Figure 1 This is a schematic flowchart of a first type of collision pre-mitigation control method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps: Step S101: Obtain the target vehicle's own status data, environmental perception data, and occupant data.

[0031] It should be noted that the vehicle status data in this embodiment refers to the vehicle's own operating parameters, structural attitude, and system operating conditions. These can be collected in real time by onboard sensors and controllers, such as using an onboard inertial measurement unit (IMU), wheel speed sensors, and steering angle sensors to collect data on vehicle speed, acceleration, yaw angle, pitch angle, steering angle, and wheel speed to reflect the vehicle's real-time driving attitude. Environmental perception data refers to the road conditions, weather, obstacles, and the status of traffic participants around the vehicle. This can be achieved through onboard cameras, such as forward / side cameras and surround-view cameras, to identify lane lines, traffic lights, pedestrians, non-motorized vehicles, and obstacles (such as fallen rocks and potholes), and to determine the relative position and movement trend of obstacles (such as the vehicle in front slowing down or a pedestrian crossing the road); or through millimeter-wave radar, cameras, weather stations, etc., on roadside equipment to obtain the road surface conditions of the road section ahead (such as icing or construction areas). In-vehicle occupant data refers to occupant identity, physiological characteristics, spatial posture, and interaction status. This data can be collected through dedicated in-cabin sensors and intelligent recognition technologies. For example, facial recognition can be performed using in-cabin cameras to determine occupant identity categories, or pressure distribution sensors and weight sensors integrated into the seats can be used to collect occupant weight (e.g., inferring weight range from pressure distribution) and body shape characteristics (e.g., whether obese / slender). In-cabin wide-angle cameras and millimeter-wave radar can be used to monitor occupant posture in real time (e.g., whether leaning forward or against the door) and the spatial coordinates of key body parts (e.g., head, torso), combined with 3D scanning technology to scan the relative distances between occupants and in-vehicle components (e.g., distance from head to steering wheel, torso to door). Alternatively, seatbelt buckle sensors can be used to detect whether occupants are wearing seatbelts. The above is only an example and can be adjusted according to actual needs.

[0032] In this embodiment, after the above data is collected, preprocessing such as spatiotemporal synchronization and noise reduction is performed to ensure data consistency.

[0033] Step S102: Based on the vehicle status data and environmental perception data, determine whether the target vehicle has a collision risk, and if the target vehicle has a collision risk, determine the current collision type.

[0034] In this embodiment, determining whether a target vehicle poses a collision risk can be achieved by calculating core risk parameters, such as collision time (the theoretical collision time is calculated based on the relative distance and relative speed between the vehicle and the obstacle), collision probability (combined with the obstacle's trajectory, such as whether a pedestrian crosses the road or the vehicle changes lanes, and the vehicle's braking / steering response capability, such as the longer braking distance at higher speeds, the probability of a collision is output by an algorithm model), and equivalent collision speed (the estimated impact speed at the time of a collision to reflect the severity of the risk). A dynamic risk threshold is also set (which can be dynamically adjusted based on the road surface adhesion coefficient and weather conditions, such as a shorter collision time threshold in rainy or snowy weather). If the collision time is less than the set threshold, the collision probability is greater than a preset ratio, and the equivalent collision speed exceeds the safety threshold, then a collision risk is determined to exist. Furthermore, the collision type of this application can be determined based on the contact direction of the obstacle and the relative motion relationship between the vehicle and the obstacle, such as frontal collision (environmental perception data shows that the obstacle is located within a certain angle range directly in front of / obliquely in front of the vehicle, and the relative motion direction is the same or opposite, the vehicle's steering angle is close to 0°, and there is no obvious steering avoidance action), side collision (the obstacle is located within a certain angle range to the left / right of the vehicle, and the relative motion direction is perpendicular or obliquely perpendicular to the vehicle's driving direction, such as vehicles approaching from the side at an intersection, non-motorized vehicles crossing, etc., and the side radar / camera detects that the obstacle is rapidly approaching the side of the vehicle), and rear-end collision (the obstacle is located within a certain angle range directly behind / obliquely behind the vehicle, the relative motion direction is the same as the vehicle, and the relative speed of the obstacle is significantly higher than that of the vehicle, such as the rear vehicle not slowing down for a rear-end collision, and the rearward sensor continuously detects the approaching obstacle).

[0035] Step S103: Construct a personalized occupant model adapted to the current occupant status based on the occupant data.

[0036] In this embodiment, the personalized occupant model is a dynamic biomechanical model that is constructed based on the multi-dimensional real-time data of the current occupants in the vehicle and can accurately represent their physiological characteristics, spatial posture and protection adaptation requirements.

[0037] Step S104: Based on the vehicle status data, the current collision type, and the personalized occupant model, generate collision pre-mitigation control commands for multiple vehicle execution systems corresponding to the target vehicle. The vehicle execution systems include at least the seat belt restraint system, the seat and airbag system, the active suspension system, and the drive braking system.

[0038] In this embodiment, the specific functions of the seat belt restraint system, seat and airbag system, active suspension system, and drive braking system can be adapted to be understood by referring to the well-known content in the art.

[0039] Step S105: Control each vehicle's execution system to perform corresponding actions according to the collision pre-mitigation control command.

[0040] It should be noted that, in this embodiment, when the vehicle central controller receives a collision pre-mitigation control command, it first decomposes the core parameters in the command (such as pretension force level, seat angle adjustment amount, suspension stiffness value, braking force distribution ratio, etc.) and converts them into hardware control signals (such as current, voltage, pulse signals) that can be recognized by each execution system to avoid incompatibility between the command and the actuator; then, it sets the action priority according to the collision type and risk level (such as the timing of seat belt pretensioning, seat angle adjustment, and airbag preparation in a frontal collision to avoid action conflicts); in high-risk collisions (such as high-speed frontal collisions), the restraint system (seat belt and airbag) is prioritized to act, while in low-risk collisions, the suspension and drive braking systems can be coordinated synchronously; finally, the action execution results are collected in real time through the built-in feedback sensors of each execution system (such as seat belt pretensioning travel sensor, seat angle sensor, suspension displacement sensor, etc.); if the action is detected to have not achieved the target (such as the seat angle not being adjusted in place), the controller immediately issues a supplementary adjustment command to ensure execution accuracy.

[0041] The collision pre-mitigation control method provided in this embodiment can predict the collision risk and type in advance based on the vehicle's status and environmental perception data, enabling proactive pre-collision intervention. This helps reduce the severity of collision damage and minimizes injury to occupants. Simultaneously, it adapts to complex scenarios involving different vehicle driving conditions, occupant combinations, and collision types. Through safety linkage with the seatbelt restraint system, seat and airbag system, active suspension system, and drive braking system, it generates corresponding collision pre-mitigation control commands for each vehicle execution system and controls each system to perform corresponding actions. This achieves proactive and comprehensive safety protection against vehicle collisions, effectively improving the efficiency of pre-collision mitigation control before or after a minor collision, greatly ensuring the personal safety of vehicle occupants.

[0042] This embodiment provides a collision pre-mitigation control method. Figure 2 This is a schematic diagram of a second process of the collision pre-mitigation control method according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process includes the following steps: Step S201: Obtain the target vehicle's own status data, environmental perception data, and occupant data. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0043] Step S202: Based on the vehicle's status data and environmental perception data, determine whether the target vehicle poses a collision risk, and if the target vehicle poses a collision risk, determine the current collision type. For details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0044] Step S203: Construct a personalized occupant model adapted to the current occupant status based on the occupant data.

[0045] In this embodiment, the occupant data includes at least the occupant's identity information, weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and the vehicle interior components. It should be noted that the identity information in this embodiment is used to distinguish the categories of occupant protection needs, such as adults, children, and the elderly, or can be further refined into ordinary adults, pregnant women, obese adults; infants, school-age children, etc.; age information indicates the actual age or age range of the occupant, such as 0-3 years old, 4-12 years old, 60 years old and above; seat belt wearing status indicates whether the occupant has properly fastened the seat belt and the real-time tension status of the seat belt, including whether it is worn, not worn, or improperly worn (such as the seat belt not being engaged in the buckle, the webbing being loose), etc.; current sitting posture data indicates the real-time body posture of the occupant in a collision risk scenario, including the body tilt angle (such as leaning forward, leaning back, leaning to the side), head position (such as looking down, looking up, leaning to one side), the fit of the torso to the seat, and the position of the limbs; the relative distance between the occupant and the in-vehicle components indicates the real-time straight-line distance between the occupant's key protective parts (such as the head, chest, and sides of the torso) and the core protective components of the vehicle (such as the steering wheel, dashboard, and seat headrests).

[0046] Specifically, step S203 includes: Step S2031: Extract the current occupant's identity information from the occupant data.

[0047] Step S2032: Determine whether a personalized passenger model that matches the identity information exists.

[0048] In this embodiment, matching can be performed based on the current occupant's identity information from a personalized occupant model containing multiple occupants that is pre-stored in a database.

[0049] Step S2033: If a personalized occupant model that matches the identity information exists, the personalized occupant model is obtained, and the personalized occupant model is updated using the current occupant's weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and the vehicle interior components.

[0050] In this embodiment, the method for updating the personalized occupant model can be adaptively determined according to actual needs. The personalized occupant model needs to record relevant content representing the latest occupant status.

[0051] Step S2034: If there is no personalized occupant model that matches the identity information, then a biomechanical model representing the current occupant's state is constructed based on the current occupant's identity information, weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and the vehicle interior components, thus obtaining a personalized occupant model.

[0052] In this embodiment, the occupant data is first preprocessed with standardized format and noise reduction correction. For example, age information is converted into specific numerical values ​​or age range codes, such as 0-3 years = 0.3; weight data is standardized to kilograms (kg); and sensor noise (such as jitter deviation in sitting posture data) is removed using a filtering algorithm, and outliers are corrected (e.g., when a child's weight is recorded as 80kg, it is corrected to a reasonable range based on identity information). Then, the occupant data is converted into biomechanical indicators, specifically based on human collision biomechanical theory (such as head injury criteria, chest compression limits), establishing the mapping relationship between the above six types of data and the core parameters of the model, forming the input and output logic of the model. For example, identity information and age information are mapped to body tolerance coefficients (impact tolerance thresholds for the head, chest, and neck), bone strength coefficients, and soft tissue cushioning capacity coefficients. The mapping rule is that for children (identity = 2), the head tolerance threshold is reduced by 30%, and the bone... The strength coefficient is 0.6; for the elderly (identity = 3), the neck tolerance threshold is reduced by 40%. Finally, a framework of basic biomechanical templates and personalized parameter correction is adopted to construct a two-dimensional model of physiological characteristics superimposed on spatial state. For example, three types of core basic templates are preset (including adults, children, and the elderly). The templates contain standard human biomechanical parameters (such as the bone distribution corresponding to the standard adult height of 175cm and the soft tissue cushioning characteristics corresponding to the standard child weight of 30kg). The basic templates are constructed based on national standards for human collision protection and industry experimental data (such as dummy collision test data). The core parameters mapped above are used to make targeted adjustments to the basic templates to form a dedicated model. For example, the tolerance parameters of the basic template are corrected by identity, age, and weight; the spatial adaptation parameters of the basic template are corrected by sitting posture, relative distance, and seat belt status; finally, a biomechanical model containing two types of key parameters is formed, which directly serves subsequent protection decisions.

[0053] In this embodiment of the invention, identity information is used as an index to prioritize the reuse of existing personalized occupant models, which can significantly reduce the computational cost of repeated modeling. For occupants without historical models, a new model is created in real time based on multi-dimensional occupant data. For historical personalized occupant models with matching data, the model is still updated using currently collected data such as weight, sitting posture, and seat belt wearing status. This avoids discrepancies between the model and reality caused by changes in occupant posture or temporary occupant replacement, ensuring that the model is always synchronized with the current occupant status. While ensuring the timeliness of protection, this greatly improves the accuracy of modeling, thereby providing data support for subsequent collision pre-mitigation control.

[0054] Step S204: Based on the vehicle status data, the current collision type and the personalized occupant model, generate collision pre-mitigation control commands corresponding to multiple vehicle execution systems for the target vehicle. The vehicle execution systems include at least the seat belt restraint system, the seat and airbag system, the active suspension system and the drive braking system.

[0055] Specifically, in step S204 above, based on the vehicle's state data, the current collision type, and the personalized occupant model, multiple collision pre-mitigation control commands corresponding to the vehicle execution system for the target vehicle are generated, including: Step S2041: Determine the current occupant status based on the personalized occupant model.

[0056] In this embodiment, the current occupant status is a comprehensive representation built based on multi-dimensional real-time data, which can fully reflect the occupant's physiological adaptation characteristics, spatial dynamic posture, and safety-related status. It accurately captures "who the occupant is, what posture they are currently in, and whether they have basic safety conditions," providing a direct basis for the personalized adaptation of subsequent protection strategies.

[0057] Step S2042: Obtain the vehicle-cloud collaborative decision-making model, and input the vehicle status data, current collision type and current occupant status into the vehicle-cloud collaborative decision-making model, and output the first control command for the seat belt restraint system, the second control command for the seat and airbag system, the third control command for the active suspension system and the fourth control command for the drive braking system; wherein, the vehicle-cloud collaborative decision-making model is obtained by iteratively training and updating the reinforcement learning model using the vehicle's historical control data.

[0058] In this embodiment, the vehicle-cloud collaborative decision-making model is obtained through cloud-based training and construction, vehicle-side deployment and invocation, and continuous iterative updates, aiming to obtain an intelligent decision-making model that is "adaptable to multiple scenarios, supports real-time decision-making, and can be dynamically optimized".

[0059] In this embodiment of the invention, the vehicle's status data, the current collision type, and the personalized occupant status are used as joint inputs to ensure that the output control commands not only conform to the vehicle's real-time operating conditions but also match the individual protection needs of the occupants, effectively improving the pertinence of the protective actions. The vehicle-cloud collaborative decision-making model is built based on reinforcement learning and is continuously updated through the accumulation of historical data. It can autonomously learn the optimal control logic in different scenarios, and thus generate exclusive and high-precision control commands for vehicle execution systems such as seat belt restraint, seats and airbags, active suspension, and drive braking, so as to achieve comprehensive and effective vehicle collision pre-mitigation control.

[0060] Step S205: Control each vehicle's execution system to perform corresponding actions according to the collision pre-mitigation control command.

[0061] It should be noted that the vehicle execution system in this embodiment consists of four types, so the above step S205 includes: controlling the seat belt restraint system to perform graded control on the pretension force and limiting force of the seat belt according to the first control command, wherein the pretension force and limiting force of the seat belt are at least related to the occupant's weight data; wherein the corresponding values ​​of the pretension force and limiting force are positively correlated with the size of the weight data.

[0062] It should be noted that the pretensioning force of the seat belt refers to the tension generated by the seat belt pretensioner actively contracting the webbing after collision risk identification (pre-collision condition) or in the early stage of collision. In essence, it is an active restraint force that "eliminates the slack of the webbing in advance and quickly fixes the occupant's body in the best protective position of the seat." The force limiting force refers to the critical force that the force limiting device automatically "releases pressure" or "slips" when the restraint force of the seat belt webbing on the occupant reaches the preset maximum value during the collision, allowing the webbing to be slowly pulled out. In essence, it is a passive protection mechanism that "limits the upper limit of the restraint force to avoid exceeding the occupant's body tolerance."

[0063] In this embodiment, the corresponding values ​​of preload and limiting force are related not only to body weight, but also to the type and size of the occupant.

[0064] In this embodiment of the invention, by classifying and controlling the pretension and limiting force of the seat belt, the adaptability to different occupant types can be improved, thereby helping to meet the usage needs of multiple scenarios such as family cars and shared cars. Furthermore, by directly and positively linking the pretension and limiting force of the seat belt with the occupant's weight, it can avoid excessive restraint on light occupants such as children and those with small body sizes, while ensuring sufficient restraint strength for heavy occupants. This can prevent excessive body displacement during a collision, balancing the effectiveness of protection and the safety of use, and to a certain extent improving the efficiency of collision pre-mitigation, thereby ensuring occupant safety.

[0065] In this embodiment, step S205 includes: controlling the seat and airbag system to coordinately regulate the seat angle, the airbag triggering timing and inflation intensity according to the second control command; wherein, the regulation of the seat angle is related to the current collision type; the regulation of the airbag triggering timing and inflation intensity is related to at least one of the following: occupant identity information, weight data, age information, seat belt wearing status, current sitting posture data, relative distance between the occupant and in-vehicle components, and the current collision type.

[0066] It should be noted that the seat angle refers to the angle between the seat back and the seat cushion. It is adjusted to the "optimal force-bearing posture" of the occupant's body based on the type of collision and the occupant's sitting posture, creating conditions for subsequent airbag protection and seat belt restraint. The airbag deployment timing refers to the time from the identification of collision risk or the initial stage of a collision to the airbag ignition and deployment. It is to ensure that the airbag fully deploys when the "occupant's body is about to contact the airbag and is in the airbag's optimal protection area," avoiding premature or delayed deployment that could lead to protection failure. The inflation intensity refers to the internal pressure or inflation volume of the airbag when it deploys. Through graded control, personalized buffering can be achieved for different occupants and different scenarios, which can prevent the impact force of the airbag deployment from exceeding the occupant's body tolerance.

[0067] In this embodiment, the seat angle adjustment is directly related to the current collision type. It can adjust the occupant's sitting posture in advance according to the force characteristics of different collision scenarios such as frontal, side, and rear-end collisions. For example, in a frontal collision, the seat can be tilted back to disperse the impact force, and in a side collision, the seat can be slightly adjusted to move away from the collision side. This reduces the impact displacement of the occupant's body during the collision and provides the optimal angle for airbag contact, thereby reducing the injury caused by local pressure.

[0068] In this embodiment, airbag control integrates multi-dimensional data such as occupant identity, weight, and seating posture, as well as collision type, rather than solely relying on collision intensity. For example, the airbag inflation intensity is reduced for child occupants, the triggering timing is delayed when the occupant is leaning forward, and the inflation pressure is weakened when the occupant is too close to in-vehicle components. This precisely matches the protection tolerance of different occupants, completely solving the problem of excessive impact or insufficient protection caused by the traditional "one-size-fits-all" approach to airbags. Specifically, in this embodiment, the seat and airbag are coordinated and controlled through a second control command, which can significantly improve the overall protection effect and greatly ensure the personal safety of the occupants.

[0069] In this embodiment, step S205 includes: controlling the active suspension system to adjust the height and stiffness of the suspension according to a third control command, wherein the adjustment of the height and stiffness of the suspension is related to the current collision type.

[0070] It should be noted that suspension height refers to the vertical distance from the center of the wheel to the bottom of the vehicle body, which can be raised or lowered by active actuators (such as air springs or electromagnetic control valves); suspension stiffness is the ability of the suspension to resist deformation. The greater the stiffness, the smaller the deformation of the vehicle body when it is subjected to an impact, and vice versa.

[0071] In this embodiment, focusing on collision pre-mitigation scenarios, suspension height and stiffness are adjusted in conjunction with the collision type. Height adjustment optimizes the vehicle's stress angle, while stiffness adjustment controls deformation and attitude. These two adjustments work together to create the optimal vehicle body state for the protective actions of systems such as seat belts and airbags, minimizing the impact on occupants. For example, suspension height and stiffness are precisely adjusted to address the impact characteristics of different collision types, such as frontal, side, and rear-end collisions. In a frontal collision, the front suspension is raised and stiffness increased, allowing energy-absorbing structures like the front longitudinal beams to absorb the impact at the optimal angle. In a side collision, the suspension on the collision side is raised and stiffness is strengthened, ensuring that robust components like the sill beams and B-pillars effectively resist lateral intrusion. Specifically, this embodiment actively shapes the vehicle's stress posture, thereby maximizing the protective potential of the vehicle structure, effectively reducing the risk of passenger compartment deformation, and significantly ensuring occupant safety.

[0072] In this embodiment, step S205 includes: controlling the drive braking system to coordinate the load distribution of drive and braking according to the fourth control command, wherein the coordinated control of load distribution is associated with the current collision type.

[0073] It should be noted that the corresponding load distribution for driving and braking is achieved by coordinating the control of driving torque and braking force to actively adjust the load distribution ratio between the front and rear axles and the left and right axles of the vehicle. Essentially, this utilizes dynamic principles to change the load distribution of the vehicle's weight, optimizing the vehicle's stress state and attitude stability for collision mitigation. For example, in a frontal collision, the load is transferred from the rear axle to the front axle, enhancing the stress matching of energy-absorbing structures such as the front longitudinal beams; in a side collision, the load is transferred to the opposite side of the collision, improving the load-bearing capacity of the vehicle's structure on the collision side. Furthermore, load imbalance during a collision can easily lead to abnormal postures such as nose-diving, side-squatting, and tail-swinging of the vehicle. In this embodiment, precise load transfer control can effectively suppress these problems. For example, in a rear-end collision, the load is transferred to the rear axle to prevent excessive body roll that could cause occupants' heads to detach from the headrests; in a side collision, the load is transferred by braking the opposite wheels, reducing lateral displacement of occupants due to body roll and lowering the probability of secondary collisions with interior components. Specifically, this embodiment actively optimizes the load distribution, allowing the vehicle's energy-absorbing structures to fully function, thereby reducing intrusion into the passenger compartment, significantly reducing the impact intensity of the collision, and ensuring occupant safety.

[0074] In one specific embodiment, given the existing challenges of deeply integrating reinforcement learning, vehicle-to-everything (V2X) technology, and cloud computing, designing a comprehensive pre-collision mitigation control system capable of vehicle-cloud coordination, adaptation, and multi-system linkage before and during minor collisions has become a crucial technical challenge in the development of automotive safety technology. This embodiment proposes a reinforcement learning-based vehicle-cloud coordinated adaptive collision pre-mitigation control system and method. This system can achieve precise occupant identification and personalized constraint system adjustment, active vehicle attitude control, and coordinated intervention of the powertrain system under pre-collision or minor collision conditions, thereby forming an integrated, adaptive safety protection network.

[0075] In this embodiment, Figure 3 This is a schematic diagram of a vehicle-cloud coordinated adaptive collision pre-mitigation control system based on reinforcement learning. As shown in the diagram, the system includes: an onboard perception module, a vehicle-to-everything (V2X) communication module, a cloud-based decision support platform, an onboard central decision controller, and multiple actuator modules. The system comprises several modules: an onboard perception module for real-time collection of vehicle status, surrounding environment, and passenger cabin information; a vehicle-to-the-cloud (V2C), vehicle-to-vehicle (V2V), and vehicle-to-infrastructure (V2I) communication module for data interaction between vehicle and cloud (V2C), vehicle and vehicle (V2V), and vehicle and infrastructure (V2I), receiving pre-trained reinforcement learning models and strategies from the cloud-based decision support platform and uploading the vehicle's perception data; a cloud-based decision support platform with a reinforcement learning model training engine and scenario database for offline training and model optimization using massive amounts of data, generating and distributing optimal control strategy models; an onboard central decision controller, the core processing unit with an embedded online reinforcement learning inference engine for rapidly calculating adaptive cooperative control commands based on real-time perception data, V2X information, and strategy models downloaded from the cloud; and an actuator module that receives commands from the onboard central decision controller and executes specific actions, including: a seatbelt adaptive adjustment unit, a seat and airbag cooperative control unit, an active suspension control unit, and a drive and braking coordination control unit.

[0076] It should be noted that the vehicle-mounted perception module in this embodiment includes forward-facing radar, forward and side cameras, and lidar for environmental perception; an inertial measurement unit (IMU) and wheel speed sensors for vehicle status perception; and an in-cabin camera, seat weight sensor, and pressure distribution pad for occupant perception. The vehicle-to-everything (V2X) communication module uses a 4G / 5C cellular communication module and a C-V2X PC5 interface module. The cloud-based decision support platform is deployed on a public cloud or industry private cloud, using a high-performance server cluster for distributed training of the reinforcement learning model. The vehicle-mounted central decision controller uses a high-performance multi-core system-on-a-chip (SoC) compliant with automotive standards, running a real-time operating system (such as AUTOSAR Adaptive), and integrating a reinforcement learning inference engine (such as TensorRT Lite). The seatbelt adaptive adjustment unit in the actuator module uses a multi-stage pyrotechnic pretensioner and an electronically controlled force limiter; the seat and airbag coordination control unit features an integrated seat with electric adjustment functions (backrest and height) and multiple zone airbags; the active suspension control unit uses air suspension or magnetorheological suspension, enabling rapid adjustment of height and stiffness; the drive and braking coordination control unit, through the Vehicle Control Unit (VCU) and Electronic Stability Program (ESP / Electronic Stability Control, ESC), achieves precise control of the motor and braking system.

[0077] In this embodiment, a corresponding control method is proposed based on the above system. Figure 4 This is a flowchart illustrating a vehicle-cloud coordinated adaptive collision pre-mitigation control method based on reinforcement learning. As shown in the diagram, the method includes the following steps: Step S1: Data perception and fusion.

[0078] In this embodiment, the vehicle's status data, surrounding environment dynamic data, and occupant biometrics and posture data are continuously acquired through the vehicle-mounted perception module; at the same time, V2X information is received through the vehicle-to-everything (V2X) communication module; and the data from all sources are synchronized in time and fused spatially to construct a unified, real-time updated environmental situational awareness map.

[0079] Step S2: Collision Risk and Scene Recognition.

[0080] In this embodiment, based on the environmental situation awareness map constructed in step S1, a preset algorithm or lightweight neural network is used to determine whether the current situation is a pre-collision risk condition or whether a minor collision has occurred. The identified content includes: collision type (such as frontal, side, or rear-end collision), expected collision point, time to collision (TTC), collision equivalent velocity (Delta-V), and the probability of collision inevitability.

[0081] Step S3: Personalized occupant status modeling.

[0082] In this embodiment, in response to the identified risks, the system initiates a high-precision occupant identification process, including: 1. Using in-cabin cameras and seat pressure distribution sensors, different occupants (such as the driver and the front passenger being an adult and a child respectively) are distinguished through facial recognition and body shape analysis, and pre-stored or cloud-synchronized occupant profiles are retrieved.

[0083] 2. Real-time monitoring of occupant posture, head and chest displacement, and whether seat belts are fastened, etc., to construct a dynamic biomechanical model for each occupant.

[0084] Step S4: Adaptive cooperative control decision based on reinforcement learning.

[0085] In this embodiment, this is a core step. Figure 5 This is a schematic diagram of reinforcement learning. Specifically, the reinforcement learning inference engine in the vehicle's central decision controller is activated. Its state space S includes: vehicle state (speed, acceleration, yaw angle, pitch angle), environmental state (relative position and speed of obstacles, road curvature), and occupant state (identity, weight, seating posture, distance from the steering wheel / dashboard). The action space A includes: A1: Multi-level adjustment command for seat belt pretension and force limiter at various positions; A2: Commands for triggering the airbags (side airbags, chest and abdominal airbags, etc.) and inflation pressure, as well as fine-tuning commands for seat back angle and seat cushion height. A3: Command to adjust the height and stiffness (soft / medium / hard) of the active suspension; A4: Coordinated control commands for drive motor torque and wheel braking pressure to actively trigger axle load transfer.

[0086] In this embodiment, the reward function R is carefully designed with the core objective of minimizing occupant injury indicators, such as the Head Injury Criterion (HIC), chest compression, and neck torque, while also considering vehicle stability, control energy consumption, and post-collision vehicle controllability. Based on the current state S, the controller selects the action combination A that yields the maximum long-term cumulative reward in the current state, according to the optimal strategy π*(A|S) downloaded from the cloud and updated periodically.

[0087] Step S5: Multi-system collaborative execution and control.

[0088] In this embodiment, the optimal action instruction A calculated in step S4 is decomposed and sent to each actuator module: 1. Differentiated Seat Belt Adjustment Control: The seat belt pretensioner is triggered differently based on the occupant's identity and body type (e.g., weight). For lighter occupants (e.g., children or women), a lower first-level pretension force and a lower force limiter threshold are used to prevent excessive restraint from causing chest and abdominal injuries; for heavier occupants, a higher pretension force and force limiter value are used to ensure effective restraint. Figure 6 This is a schematic diagram illustrating differentiated control of seat belts, airbags, and load transfer for different occupants. Please refer to the relevant content in the diagram for an adaptive understanding.

[0089] 2. Seat airbag and angle adjustment control: Based on the occupant's identity, seating posture, and collision direction, the deployment sequence and inflation intensity of the corresponding airbags are controlled. For example, for shorter occupants, the deployment angle of the steering wheel side airbag is appropriately lowered; before a frontal collision, the seat back is slightly tilted back to allow the occupant to contact the airbag at a better angle, thus dispersing the impact force.

[0090] In this embodiment, upon a collision, the vehicle collects information via sensors. The control unit then precisely controls the airbag deployment sequence and inflation intensity based on factors such as occupant identity, seating position, and collision direction to achieve optimal protection. Specifically: i. Based on occupant identity: The vehicle can identify occupant identity information such as weight and body size through seat pressure sensors and occupant information camera modules. For adults weighing ≥30kg, the airbag will activate normally and deploy at standard inflation strength; for lighter children or smaller occupants, the airbag may inhibit deployment or reduce inflation pressure to avoid injury from the airbag impact. If the occupant is not wearing a seatbelt, the system will increase the airbag deployment threshold, so that the airbag will only trigger under more severe collision conditions, reducing the risk of airbag deployment to unrestrained occupants.

[0091] ii. Based on seating posture: The vehicle can obtain occupant seating posture information through seat position sensors, seat back tilt angle sensors, etc. If the occupant's seating posture is normal, the airbag will deploy according to the normal strategy. If the occupant is in an abnormal seating posture, such as leaning forward or having their head close to the airbag module, the system will delay the airbag deployment time or reduce the inflation intensity to prevent secondary injury to the occupant when the airbag deploys.

[0092] iii. Based on the direction of the collision: The vehicle is equipped with collision sensors in multiple directions, such as front collision sensors and side collision sensors. In a frontal collision, the front airbags will deploy first. If the collision intensity is high, the knee airbags will deploy subsequently. The inflation intensity is adjusted according to the severity of the collision; in high-intensity collisions, the airbags will deploy quickly with a higher inflation intensity, while in low-intensity collisions, the inflation intensity will be reduced. In a side collision, the side airbags and curtain airbags will deploy rapidly to provide side protection for the occupants. The inflation intensity is also adjusted according to the collision force to effectively buffer the side impact.

[0093] 3. Vehicle suspension height and stiffness control: The suspension is adjusted instantaneously before a collision based on the type of impact. For a frontal collision, the front suspension height is actively raised and its stiffness increased, allowing the front longitudinal beams to participate in energy absorption at the optimal angle. Simultaneously, the rear suspension stiffness is reduced to facilitate rear-end sag, suppressing the "nose-diving" phenomenon and maintaining passenger compartment stability. For a side collision, the suspension on the impact side is instantly raised, allowing robust components such as the sill beams to better withstand the impact.

[0094] 4. Axle load transfer control in drive and braking: By precisely controlling the output torque of the drive motor and applying slight braking to the wheels on the non-collision side, a pitch or roll moment is actively generated to achieve axle load transfer. Before a frontal collision, a slight drive torque request (even if the driver's foot is off the accelerator) or rear wheel braking is applied to create a "sit-back" effect, increasing the rear axle load and reducing the front axle load, thereby reducing the risk of frontal collision intrusion and optimizing the operating conditions of the front safety system.

[0095] In this embodiment, based on data sharing, parameters are preset and looked up according to the received collision degree, collision location, vehicle speed, etc., and the parameters are verified and checked after being fed back.

[0096] Step S6: Collaboration between vehicle-side and cloud-side models.

[0097] In this embodiment, this step is essentially online learning and cloud-based model updates. Specifically, after each control event (regardless of whether a collision ultimately occurs), the system uploads the current state, actions, and final result (such as the actual collision intensity, changes in occupant status, etc., which can be inferred through post-event analysis or sensor data) as an experience sample to the cloud-based decision support platform via the vehicle-to-everything (V2X) communication module. The platform continuously trains and optimizes the reinforcement learning model using a massive amount of new samples, and iteratively distributes the improved model to all vehicles in the fleet, enabling continuous evolution of the system's capabilities.

[0098] In one specific embodiment, taking a typical rear-end collision risk scenario on an urban road as an example, see [reference]. Figures 3-6 The aforementioned vehicle-cloud coordinated adaptive collision pre-mitigation control method based on reinforcement learning includes: Step S1: This vehicle is following the vehicle in front in congested traffic. The onboard forward radar and camera detect a sudden emergency braking by the vehicle in front. The IMU shows that this vehicle is decelerating. At the same time, V2V communication receives the emergency braking signal from the vehicle in front, providing an earlier and more accurate warning. The in-cabin camera identifies the driver as a medium-sized adult male with a normal sitting posture.

[0099] Step S2: After data fusion, the system calculates that the TTC is less than 1.5 seconds and the collision probability is greater than 90%, which is determined to be a high-risk rear-end collision pre-collision condition.

[0100] Step S3: The system confirms the driver's identity, retrieves the driver's preset weight parameters (e.g., 75kg), and monitors the distance between the driver's head and the headrest in real time.

[0101] Step S4: The reinforcement learning inference engine is activated. State S includes: vehicle speed 50 km / h, vehicle in front speed 5 km / h, relative speed 45 km / h, TTC = 1.2s, driver weight 75 kg, and normal seating posture. Based on the cloud-deployed strategy model optimized for rear-end collision scenarios, the inference engine calculates the optimal action combination A: A1: Trigger medium-level pretensioning of the driver's seatbelt, with the force limiter set to medium.

[0102] A2: Adjust the seat back angle slightly backward by 2 degrees and move the headrest forward appropriately to better support the head tilting backward; increase the inflation pressure of the rear impact protection airbag integrated in the seat back.

[0103] A3: The rear suspension height is lowered by 20mm and the stiffness is increased to "hard" mode to optimize the energy absorption characteristics of the rear collision and suppress "uplift" deformation; the front suspension maintains the original height and the stiffness is adjusted to "soft" to facilitate energy absorption.

[0104] A4: The command drives the motor to output a slight negative torque (maximum energy recovery intensity) and simultaneously applies a slight braking force to the front wheels (without triggering ABS), causing the vehicle to "nod" and shift the load forward, enhancing front wheel grip to assist braking and optimizing the vehicle's pitch attitude.

[0105] Step S5: Each actuator completes the above actions within approximately 50-100 milliseconds. Subsequently, the vehicle's AEB system triggers full braking, but due to the previous axle load transfer, the vehicle's braking efficiency is higher. Ultimately, the vehicle collided with the vehicle in front at a relatively low speed (e.g., 15 km / h), constituting a minor collision. Due to the coordinated adjustments of the seatbelt, seat, and suspension, the impact on the driver's neck from this rear-end collision was significantly mitigated.

[0106] Step S6: The data from this event is recorded and uploaded to the cloud for further optimization of control strategies in rear-end collision scenarios.

[0107] In one specific embodiment, taking a side collision as an example, particularly when the system identifies the occupant on the side of the collision as a child, see [reference]. Figures 3-6 The aforementioned vehicle-cloud coordinated adaptive collision pre-mitigation control method based on reinforcement learning also includes: Step S1: The side radar detected a vehicle approaching at high speed from the side of the intersection, posing a risk of side collision.

[0108] Step S2: The cabin camera and pressure sensor identify that a child is sitting in the left rear seat (through facial and body recognition, or a preset child seat recognition system).

[0109] Steps S3-S5: Seatbelt control (A1): The seatbelt pretension of the child seat (or the child himself) is set to the lowest level, and the force limit is also set to the lowest level to avoid causing pressure injury to the child's delicate body.

[0110] Seat and Airbag Control (A2): Deactivates or reduces the deployment force of the chest airbag on the impact side (left side) to prevent injury to the child from the airbag deployment itself. Simultaneously, the angle of the child seat base can be adjusted (if electrically adjustable) to face it more towards the inside of the vehicle.

[0111] Suspension Control (A3): Instantly commands the left front / left rear suspension to rise to its highest position and set to maximum stiffness, so that the robust door sill beam and B-pillar area are better aligned with the front of the impacting vehicle, maximizing the absorption of collision energy by the chassis structure and reducing direct intrusion into the passenger compartment.

[0112] Axle load transfer control (A4): Apply a brief, moderate braking force to the right wheel, causing the vehicle to tilt instantaneously to the left, further raising the left side of the vehicle body, while generating a lateral acceleration away from the direction of impact, attempting to push the occupants to the right, away from the point of impact.

[0113] In this embodiment, through the above-mentioned coordinated control, child occupants can be provided with maximum protection even in unavoidable side collisions.

[0114] In summary, this embodiment constructs an intelligent, adaptive, and collaborative collision pre-mitigation system by deeply integrating reinforcement learning, vehicle-to-everything (V2X) technology, significantly improving occupant protection under complex and hazardous conditions. Specifically, the reinforcement learning-based vehicle-cloud coordinated adaptive collision pre-mitigation control scheme in this embodiment has the following advantages: 1. Foresight and Adaptability: This embodiment uses reinforcement learning for decision-making, which can comprehensively consider the current and future changes in the state of the vehicle and occupants to make long-term optimal collaborative control decisions, rather than simple rule responses.

[0115] 2. Personalized protection: In this embodiment, through precise occupant identification and status modeling, the restraint systems such as seat belts and airbags are adjusted in a personalized manner, which significantly improves the protection effect, especially for occupants with special body types (such as children and pregnant women).

[0116] 3. Vehicle-wide collaborative optimization: This embodiment breaks down the barriers of traditional systems operating independently, integrating the constraint system, chassis system, and power system control into one. Through suspension adjustment and axle load transfer, it actively shapes a more favorable collision posture and mechanical environment, transforming from "passively bearing" to "actively responding".

[0117] 4. Vehicle-Cloud Collaborative Evolution: The introduction of a cloud platform in this embodiment enables the experience gained by one vehicle to be shared by the entire fleet, greatly accelerating the training and optimization process of reinforcement learning models and enabling the system to cope with rare but dangerous "long tail" scenarios.

[0118] 5. Technological Foresight and Economic Efficiency: This embodiment introduces reusable airbags, solving the problems of traditional airbags being "disposable" and "overkill," allowing the system to intervene in multiple warnings, thus improving system availability and economic efficiency. Pre-collision seat movement is a low-cost, high-efficiency injury mitigation method.

[0119] 6. Wider coverage: This embodiment not only targets high-speed severe collisions, but also provides effective mitigation measures for low-speed minor collisions and unavoidable pre-collision phases, thus improving the coverage and overall performance of the safety system.

[0120] This embodiment also provides a collision pre-mitigation control device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0121] This embodiment provides a collision pre-mitigation control device, such as... Figure 7 As shown, it includes: The data acquisition module 701 is used to acquire the target vehicle's own status data, environmental perception data, and occupant data.

[0122] The collision determination module 702 is used to determine whether there is a collision risk to the target vehicle based on the vehicle's status data and environmental perception data, and to determine the current collision type when there is a collision risk to the target vehicle.

[0123] The state modeling module 703 is used to build a personalized occupant model that adapts to the current occupant state based on in-vehicle occupant data.

[0124] The instruction generation module 704 is used to generate collision pre-mitigation control instructions for multiple vehicle execution systems of the target vehicle based on the vehicle status data, the current collision type and the personalized occupant model. The vehicle execution systems include at least the seat belt restraint system, the seat and airbag system, the active suspension system and the drive braking system.

[0125] The action execution module 705 is used to control each vehicle execution system to perform corresponding actions according to the collision pre-mitigation control command.

[0126] In some alternative implementations, the state modeling module 703 includes: The first modeling submodule is used to extract the identity information of the current occupants from the occupant data.

[0127] The second modeling submodule is used to determine whether a personalized passenger model that matches the identity information exists.

[0128] The third modeling submodule is used to obtain a personalized occupant model if a personalized occupant model that matches the identity information exists, and to update the personalized occupant model using the current occupant's weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and the vehicle interior components.

[0129] The fourth modeling submodule is used to construct a biomechanical model representing the current occupant's state based on the occupant's identity information, weight data, age information, seat belt wearing status, current sitting posture data, and the relative distance between the occupant and in-vehicle components if no personalized occupant model matching the identity information does not exist, thus obtaining a personalized occupant model.

[0130] In some alternative implementations, the instruction generation module 704 includes: The first generation submodule is used to determine the current occupant status based on the personalized occupant model.

[0131] The second generation submodule is used to acquire the vehicle-cloud collaborative decision-making model, and input the vehicle's state data, current collision type and current occupant state into the vehicle-cloud collaborative decision-making model, and output the first control command for the seat belt restraint system, the second control command for the seat and airbag system, the third control command for the active suspension system and the fourth control command for the drive braking system; wherein, the vehicle-cloud collaborative decision-making model is obtained by iteratively training and updating the reinforcement learning model using the vehicle's historical control data.

[0132] In some alternative implementations, the action execution module 705 includes: The first execution submodule is used to control the seat belt restraint system to perform graded control of the pretension and limiting force of the seat belt according to the first control command. The pretension and limiting force of the seat belt are at least related to the occupant's weight data; wherein the corresponding values ​​of the pretension and limiting force are positively correlated with the size of the weight data.

[0133] The second execution submodule is used to control the seat and airbag system to coordinate the adjustment of the seat angle, as well as the triggering timing and inflation intensity of the airbag, according to the second control command; wherein, the adjustment of the seat angle is related to the current collision type; the adjustment of the triggering timing and inflation intensity of the airbag is related to at least one of the following: occupant identity information, weight data, age information, seat belt wearing status, current sitting posture data, relative distance between the occupant and in-vehicle components, and the current collision type.

[0134] The third execution submodule is used to control the active suspension system to adjust the height and stiffness of the suspension according to the third control command. The adjustment of the height and stiffness of the suspension is related to the current collision type.

[0135] The fourth execution submodule is used to control the drive and braking system to coordinate the load distribution of drive and braking according to the fourth control command. The coordinated control of load distribution is associated with the current collision type.

[0136] The collision pre-mitigation control device provided in this embodiment of the invention can execute the collision pre-mitigation control method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the various modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0137] The collision pre-mitigation control device in this embodiment of the invention judges the collision risk and type based on the vehicle's status and environmental perception data, which helps to reduce the degree of collision damage and realizes proactive pre-collision intervention. By constructing a personalized occupant model adapted to the current occupant status, it can tailor the most suitable seat belt restraint force, airbag deployment parameters, etc. for different occupants, effectively protecting them while minimizing secondary injuries caused by improper deployment of protective devices, thus achieving truly personalized occupant protection. At the same time, it can be adapted to various scenarios and, through safety linkage with various vehicle execution systems, achieves proactive and comprehensive protection against vehicle collisions. It can effectively improve the efficiency of pre-mitigation control of vehicle collisions before or after a minor collision, greatly ensuring the personal safety of occupants.

[0138] Figure 8 This is a schematic diagram of the structure of a vehicle provided in an embodiment of the present invention.

[0139] The following is a detailed reference. Figure 8 The diagram illustrates a structural schematic suitable for implementing a vehicle according to an embodiment of the present invention. The vehicle includes a controller, which may include a processor (e.g., a central processing unit, a graphics processor, etc.) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory ROM 802 or a program loaded from memory 808 into a random access memory RAM 803. The RAM 803 also stores various programs and data required for vehicle operation. The processor 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0140] Typically, the following devices can be connected to I / O interface 805: input devices 806 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; memory devices 808 including, for example, magnetic tape, hard disk, etc.; and communication devices 809. Communication device 809 allows the vehicle to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 Vehicles with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0141] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 809, or installed from a memory 808, or installed from a ROM 802. When the computer program is executed by the processor 801, it performs the functions defined in the collision pre-mitigation control method of the embodiments of the present invention.

[0142] Figure 8 The vehicle shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0143] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium after being downloaded via a network. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the collision pre-mitigation control method shown in the above embodiments is implemented.

[0144] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A pre-crash mitigation control method characterized by, The method comprises: acquiring the in-vehicle state data, the environment perception data and the in-vehicle passenger data of a target vehicle; judging whether the target vehicle has a collision risk based on the in-vehicle state data and the environment perception data, and determining a current collision type when the target vehicle has a collision risk; constructing a personalized passenger model adaptive to the current passenger state based on the in-vehicle passenger data; generating collision pre-relief control instructions corresponding to a plurality of vehicle execution systems of the target vehicle based on the in-vehicle state data, the current collision type and the personalized passenger model, the vehicle execution systems at least including a seat belt restraint system, a seat and airbag system, an active suspension system and a drive braking system; controlling each vehicle execution system to perform corresponding actions according to the collision pre-relief control instructions.

2. The collision pre-mitigation control method according to claim 1, characterized by, The in-vehicle passenger data at least includes the identity information, weight data, age information, seat belt wearing state, current sitting posture data and relative distance of the passenger to the vehicle components of the passenger; the personalized passenger model adaptive to the current passenger state is constructed based on the in-vehicle passenger data, comprising: extracting the identity information of the current passenger in the in-vehicle passenger data; judging whether there is a personalized passenger model matching the identity information; if there is a personalized passenger model matching the identity information, acquiring the personalized passenger model and updating the personalized passenger model by using the weight data, age information, seat belt wearing state, current sitting posture data and relative distance of the passenger to the vehicle components of the current passenger; if there is no personalized passenger model matching the identity information, constructing a biomechanical model representing the current passenger state according to the identity information, weight data, age information, seat belt wearing state, current sitting posture data and relative distance of the passenger to the vehicle components of the current passenger to obtain the personalized passenger model.

3. The collision pre-mitigation control method according to claim 1, wherein The personalized passenger model is generated based on the in-vehicle state data, the current collision type and the personalized passenger model, comprising: determining the current passenger state based on the personalized passenger model; acquiring a vehicle-cloud collaborative decision-making model and inputting the in-vehicle state data, the current collision type and the current passenger state into the vehicle-cloud collaborative decision-making model to output a first control instruction for the seat belt restraint system, a second control instruction for the seat and airbag system, a third control instruction for the active suspension system and a fourth control instruction for the drive braking system; wherein the vehicle-cloud collaborative decision-making model is obtained by iteratively training and updating a reinforcement learning model using historical control data of the vehicle.

4. The collision pre-mitigation control method according to claim 3, wherein The control of each vehicle execution system to perform corresponding actions according to the collision pre-relief control instructions comprises: controlling the seat belt restraint system to perform hierarchical control on the pretightening force and force limiting of the seat belt according to the first control instruction, the pretightening force and force limiting of the seat belt being at least associated with the weight data of the passenger; wherein the corresponding values of the pretightening force and force limiting are positively correlated with the size of the weight data.

5. The collision pre-mitigation control method according to claim 3, wherein The control each vehicle execution system according to the collision pre-relief control instruction executes corresponding actions, including: The seat and airbag system is controlled according to the second control instruction to coordinate the control of the seat angle and the triggering time and inflation strength of the safety airbag; wherein the control of the seat angle is associated with the current collision type; the control of the triggering time and inflation strength of the safety airbag is associated with at least one of the identity information, weight data, age information, seat belt wearing state, current sitting posture data, relative distance between the occupant and the vehicle interior components, and the current collision type of the occupant.

6. The collision pre-mitigation control method according to claim 3, wherein The control each vehicle execution system according to the collision pre-relief control instruction executes corresponding actions, including: The active suspension system is controlled according to the third control instruction to control the height and stiffness of the suspension; the control of the height and stiffness of the suspension is associated with the current collision type.

7. The collision pre-mitigation control method according to claim 3, wherein The control each vehicle execution system according to the collision pre-relief control instruction executes corresponding actions, including: The drive and brake system is controlled according to the fourth control instruction to coordinate the control of the corresponding load distribution of driving and braking; the coordination control of the load distribution is associated with the current collision type.

8. A pre-crash mitigation control device characterized by comprising: The device comprises: A data acquisition module for acquiring the vehicle state data, environment perception data and in-vehicle occupant data of the target vehicle; A collision judgment module for judging whether the target vehicle has a collision risk based on the vehicle state data and the environment perception data, and determining the current collision type when the target vehicle has a collision risk; A state modeling module for constructing a personalized occupant model adapted to the current occupant state based on the in-vehicle occupant data; An instruction generation module for generating collision pre-relief control instructions corresponding to a plurality of vehicle execution systems of the target vehicle based on the vehicle state data, the current collision type and the personalized occupant model, the vehicle execution systems including at least a seat belt restraint system, a seat and airbag system, an active suspension system and a drive and brake system; An action execution module for controlling each vehicle execution system to execute corresponding actions according to the collision pre-relief control instructions.

9. A vehicle characterized by comprising: The vehicle comprises a controller, the controller comprises a memory and a processor, the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the collision pre-relief control method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the collision pre-relief control method of any one of claims 1-7.

Images