Intelligent vehicle limit safety system under extreme working condition

Abstract

The invention discloses an intelligent vehicle limit safety system under an extreme working condition, and the system comprises a vehicle sensing system which supports a limit safety function and is used for collecting environment information of a vehicle under the extreme working condition and transmitting the environment information to an automatic driving system and a high-level auxiliary driving system in real time; the limit safety function decision-making module of the automatic driving system and the high-level auxiliary driving system is used for acquiring environment information, input information based on a data-driven reinforcement learning strategy and feedback vehicle state and execution information in an unknown environment; and the limit safety function execution module is used for acquiring the decision information and outputting the vehicle state and execution information. Operation including but not limited to dynamic critical stability or extreme motion capability can be carried out under the extreme working condition, so that the motion potential of each system of the intelligent vehicle is fully utilized, and the emergency processing capability and backup level of the intelligent vehicle under the extreme working condition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile active safety, and in particular to an extreme safety system for intelligent vehicles under extreme working conditions. Background Art

[0003] Extreme operating conditions usually refer to critical environments with a risk of collision or conditions related to a serious deterioration in the vehicle's own condition. Relevant scenarios include sudden forward accident scenarios, high-speed passing through sharp turns, and critical vehicle instability. Under such operating conditions, the vehicle's reaction speed and decision-making intelligence level are particularly critical. The control idea of ​​the active safety system currently installed on vehicles is mainly to limit the vehicle's driving state to a linear, steady-state range to avoid vehicle instability when the wheels reach their dynamic limits. From the perspective of vehicle controllability, the current active safety functions are too conservative and cannot fully utilize the vehicle's dynamic control potential under extreme conditions.

[0004] At present, the design of safety functions for extreme working conditions is still in its early stages. Active safety functions at the product level, such as automatic emergency braking, target relatively single scenarios and have relatively simple control strategies. Functional design at the research level is relatively fragmented, with limited utilization of the vehicle's dynamic control potential. There is also a lack of a unified framework to integrate methods that use different technologies, which limits the synergy of vehicle execution systems in actual applications.

[0005] At present, the theoretical method for extreme safety functions in extreme scenarios has not been established, and the triggering standards, decision-making planning and control methods for extreme safety functions are not systematic. For the triggering standards, there is no unified triggering system that comprehensively considers the needs of collision avoidance, cornering movement, and dynamic stability, resulting in inconsistencies and limitations in the application of triggering standards in different extreme situations. For decision-making planning, there is a lack of an evaluation system for decision-making in a wide range of extreme scenarios. For control methods, there is a lack of a unified framework to integrate the advantages of various technical routes, which makes the synergy of current technologies in practical applications limited.

[0006] In response to the above problems, there is an urgent need to establish an extreme safety function, architecture and system with unified triggering standards, a complete evaluation system and a collaborative control solution. Summary of the invention

[0007] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0008] To this end, the present invention proposes an extreme safety system for intelligent vehicles under extreme working conditions, which performs operations including but not limited to those involving critical dynamic stability or extreme motion capabilities under extreme working conditions, so as to fully utilize the motion potential of various systems of the intelligent vehicle and enhance the emergency handling capability and backup level of the intelligent vehicle under extreme working conditions.

[0009] To achieve the above objectives, the present invention provides, on one hand, an extreme safety system for an intelligent vehicle under extreme working conditions, comprising:

[0010] The vehicle sensor system that supports extreme safety functions is used to collect environmental information under extreme vehicle conditions and send the environmental information to the autonomous driving system and high-level driver assistance system in real time via Ethernet or CAN communication;

[0011] The automatic driving system and the high-level auxiliary driving system include an extreme safety function decision module and an extreme safety function execution module, wherein;

[0012] The extreme safety function decision module includes an extreme collision avoidance judgment module, an extreme cornering judgment module, an extreme vehicle condition recognition module, an extreme state recognition module and a safety boundary monitoring module; the extreme safety function decision module is used to obtain environmental information and receive input information based on a data-driven reinforcement learning strategy in an unknown environment as well as vehicle status and execution information fed back by an extreme safety function execution module;

[0013] The extreme safety function execution module is used to obtain the decision information sent by the extreme safety function decision module and output the vehicle status and execution information.

[0014] The extreme safety system for an intelligent vehicle under extreme working conditions according to the embodiment of the present invention may also have the following additional technical features:

[0015] In one embodiment of the present invention, the vehicle sensing system supporting the extreme safety function includes a camera, a millimeter wave radar and a combined inertial navigation IMU.

[0016] In one embodiment of the present invention, the extreme safety function decision module is further used to:

[0017] The target trajectory is solved by an extreme safety decision solver based on the evaluation results and using an iterative optimization or dynamic optimization method; wherein the evaluation results include the characteristics of the extreme scenario, and the evaluation results are obtained by evaluating the risk avoidance response of the vehicle in the extreme scenario in combination with the direct risk of the scenario, the expected risk of the scenario, the risk avoidance level of the operation, and the degree of the limit of the operation;

[0018] All target trajectories of decision solutions matching the current extreme scenario are passed to the extreme safety function execution module.

[0019] In one embodiment of the present invention, the system is also used to fuse hierarchical decision control and data-driven decision control using a fusion control mechanism; wherein,

[0020] The fusion control mechanism is designed for situations where hierarchical decision control and data-driven decision control have the same objectives: deploying decisions and generating trajectories in a data generation environment, replacing the trajectory information output by hierarchical decision making with trajectories; using a model-based trajectory tracker to generate closed-loop inputs Directly generate feed-forward input using data-driven The two are added or weighted to generate the final input to be executed:

[0021]

[0022] The fusion control mechanism is aimed at the situation where the hierarchical decision control and data-driven decision control have different goals: by comparing the goal gap between data-driven decision and hierarchical decision, the data-driven strategy directly generates action information a Data , track the hierarchical decision target trajectory to obtain action information a model ,Based on the target difference, the final input to be executed is generated, and the generation methods include neural network fusion and weighted fusion:

[0023] a t =Diff target (a Data ,a model ).

[0024] In one embodiment of the present invention, in response to the execution input output by the fusion control mechanism, the four-wheel drive and rear-wheel drive are combined in driving to control the drive slip and expand the vehicle's extreme driving control boundary; the friction braking and motor braking are combined in braking to provide backup extreme braking capacity guarantee; the active steering and differential steering are combined in steering to provide maximum posture adjustment capability for the extreme safety function; each vehicle actuator executes the target input and performs extreme safety maneuvers.

[0025] In one embodiment of the present invention, the environmental information is a sharp bend environment, and the goal of the extreme safety decision module is to ensure comprehensive safety of extreme drift bends; the evaluation indicators related to extreme bend performance and safety include immediate evaluation and final evaluation: the immediate bend performance indicators include: the evaluation item r used to track the suboptimal trajectory of the scene p :

[0026]

[0027] Among them, k pl and k pv is a negative parameter; l is the lateral deviation of the vehicle relative to the center of the curve in the Frenet coordinate system, v is the current vehicle speed; l ref (s) is the lateral displacement of the suboptimal trajectory of the formation at the current bend center; is the maximum vehicle speed under the current stability constraint.

[0028] In one embodiment of the present invention, the instant evaluation item r for rewarding high sideslip angles is β , encouraging the vehicle to turn with a higher center of mass slip angle:

[0029]

[0030] Among them, k β is a negative number, v x is the longitudinal speed in the vehicle coordinate system, v y is the lateral speed in the vehicle coordinate system, β is the sideslip angle of the vehicle’s center of mass;

[0031] The final reward item r is used to reward the comprehensive performance of the ultimate safety function t , encourages safe and fast cornering:

[0032] r t =(1-χ)·k t1 +k t2 ·χ·(t f -t ref )

[0033] Where χ represents the parameter of the final state of the vehicle: χ = 1 means that the vehicle has safely completed the turning task, χ = 0 means an unsafe event has occurred, t f It represents the time required for the vehicle to reach the final state, t ref represents the total travel time of the pre-optimized trajectory; the constant k t1 and k t2 are negative and positive values, respectively, used to penalize unsafe terminal states and reward the shortest possible extreme turning time.

[0034] In one embodiment of the present invention, the extreme safety decision solver adopts an end-to-end iterative solver based on reinforcement learning, and the output is directly action information; the iterative method is a Critic and Actor deep neural network, and the Critic network parameters are trained according to minimizing the time difference loss function; the Actor network parameters are trained by maximizing the value function.

[0035] In one embodiment of the present invention, the data acquisition environment used in the extreme safety decision solver is the Carsim platform; the control strategy based on reinforcement learning is deployed to the corresponding scene in the data acquisition environment to obtain the optimal trajectory T in the current scene. p , the trajectory is used as the target trajectory information of the extreme safety function execution module:

[0036] T p ={S i |simulate(P v ,V c ,S ini,π),i=1,2,3,…,n}

[0037] Where S represents the vehicle state along the estimated trajectory, S ini Indicates the vehicle status when entering a curve.

[0038] In one embodiment of the present invention, since the estimated target trajectory is used to generate feedback input, it is necessary to set T p Convert to Cartesian coordinates Tracking Trajectory Using a Proportional-Integral-Derivative Controller Get supplementary input a PID ; The end-to-end actions output by reinforcement learning will serve as important feedforward references and are recorded as a RL ; Hierarchical decision making has the same goal as data-driven decision making, and uses direct addition to obtain execution input:

[0039] a t =a RL +a PID .

[0040] The extreme safety system for intelligent vehicles under extreme working conditions in the embodiment of the present invention integrates traditional friction braking and modern motor braking technology in terms of the braking system, which not only provides precise braking force, but also increases the redundancy of braking to ensure high safety under extreme braking conditions. In terms of steering technology, this embodiment combines active steering system and differential steering technology, which greatly enhances the vehicle's maneuverability and stability in emergency situations and provides the maximum posture adjustment capability for the extreme safety function.

[0041] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0043] Figure 1 is a structural diagram of an extreme safety system for an intelligent vehicle under extreme working conditions according to an embodiment of the present invention;

[0044] Figure 2 is a deployment diagram according to an embodiment of the present invention;

[0045] Figure 3 It is a deployment comparison trajectory diagram according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0048] The following describes an extreme safety system for an intelligent vehicle under extreme working conditions according to an embodiment of the present invention with reference to the accompanying drawings.

[0049] Figure 1 The structure diagram of the extreme safety system of the intelligent vehicle under extreme working conditions of the present invention is as follows: Figure 1 As shown, including:

[0050] The vehicle sensor system 10 supporting the extreme safety function is used to collect environmental information of the vehicle under extreme working conditions and send the environmental information to the automatic driving system and the high-level auxiliary driving system 20 in real time through Ethernet or CAN communication;

[0051] The automatic driving system and the high-level auxiliary driving system 20 include an extreme safety function decision module 21 and an extreme safety function execution module 22, wherein;

[0052] The extreme safety function decision module 21 includes an extreme collision avoidance judgment module, an extreme cornering judgment module, an extreme vehicle condition recognition module, an extreme state recognition module and a safety boundary monitoring module; the extreme safety function decision module is used to obtain environmental information and receive input information based on a data-driven reinforcement learning strategy in an unknown environment as well as vehicle status and execution information fed back by the extreme safety function execution module 22;

[0053] The extreme safety function execution module 22 is used to obtain the decision information sent by the extreme safety function decision module 21 and output the vehicle status and execution information.

[0054] To achieve the above-mentioned objectives, the extreme safety system for intelligent vehicles under extreme working conditions of the present invention adopts the following technical solutions: extreme safety functions integrated in the advanced driver assistance system (ADAS) or automatic driving system of intelligent vehicles; modular extreme safety function decision and control architecture, including decision intervention models for extreme scenarios, extreme safety function optimizers and vehicle response evaluation systems; extreme operation execution methods that integrate the advantages of various solution technologies, including input fusion control mechanisms and cross-system collaborative input execution mechanisms.

[0055] For example, it is proposed that the vehicle sensing system supporting the extreme safety function includes a single or multiple sensors or sensor combinations: camera, millimeter wave radar and combined inertial navigation. Given that extreme working conditions often occur at high speeds, other sensors that can be used to perceive obstacles at high speeds should also be included. The system sends environmental information to the extreme safety function decision and execution module in real time via Ethernet or CAN communication.

[0056] Exemplarily, an extreme safety function integrated in an advanced driver assistance system (ADAS) or an automatic driving system (ADS) of an intelligent vehicle is proposed, including but not limited to the use of operations involving near-stability of dynamics, such as drift operations. This function can be integrated into the active safety module of the advanced driver assistance system (ADAS) or the automatic driving system (ADS) of the vehicle, and when the extreme safety function intervenes, its own priority is increased, and the vehicle stability function or related active safety functions including but not limited to the anti-lock braking system or the electronic stability control are taken over or shielded.

[0057] For example, an extreme safety function intervention judgment module is proposed. Given that extreme conditions account for a relatively low proportion of driving scenarios, only this intervention judgment module runs full-time, monitoring the extreme collision avoidance requirements, cornering extreme movement requirements, vehicle extreme conditions, and dynamic stability of driving conditions in real time, and comprehensively judging whether to trigger the extreme safety function.

[0058] For example, an evaluation system for intelligent vehicle risk avoidance response in extreme scenarios is constructed. Based on the characteristics of extreme scenarios, the risk of the scenario is directly evaluated, the risk of the scenario is expected, the risk avoidance level of the operation is evaluated, and the degree of the operation limit is combined to evaluate the risk avoidance response of the vehicle in extreme scenarios.

[0059] Exemplarily, an extreme safety decision solver is constructed. Combined with the extreme scenario vehicle risk avoidance response evaluation system established above, methods including but not limited to iterative optimization and dynamic optimization are used to solve the target trajectory or end-to-end input information. It is worth noting that the decision module of the present invention is compatible with various model-based or data-driven solution methods. All decision solution outputs matching the current extreme scenario are passed to the extreme safety function execution module.

[0060] Exemplarily, an extreme operation fusion mechanism is proposed that integrates the advantages of various solution technologies. Hierarchical decision-making usually has strong environmental adaptability, and usually outputs target trajectory information. Data-driven decision control usually has better global adaptability, and usually outputs directly action information. The present invention adopts a fusion control mechanism to combine the advantages of these two technical routes.

[0061] Fusion control mechanism, for the case where hierarchical decision and data-driven decision have the same goal: deploy the decision in the data generation environment and generate a trajectory, and use this trajectory to replace the trajectory information output by the hierarchical decision. Furthermore, a model-based trajectory tracker is used to solve and generate a closed-loop input Directly generate feed-forward input using data-driven The two are added or weighted to generate the final input to be executed.

[0062]

[0063] Fusion control mechanism, for the situation where hierarchical decision-making and data-driven decision-making have different goals: Compare the goal gap between data-driven decision-making and hierarchical decision-making, and the data-driven strategy directly generates action information a Data , track the hierarchical decision target trajectory to obtain action information a model ,Considering the target difference, the final input to be executed is generated, and the generation methods include but are not limited to neural network fusion, weighted fusion, etc.

[0064] a t =Diff target (a Data ,a model )

[0065] Exemplarily, cross-system coordinated extreme action execution is proposed. For the execution input output by the fusion control mechanism, the four-wheel drive and rear-wheel drive are combined in driving to accurately control the drive slip and expand the vehicle's extreme driving control boundary; the friction brake and motor brake are combined in braking to provide accurate and backup extreme braking capacity guarantee; the active steering and differential steering are combined in steering to provide the maximum posture adjustment capability for the extreme safety function. Furthermore, each actuator of the vehicle executes the target input and performs extreme safety maneuvers. The executed maneuvers and the resulting scene evolution will be transmitted to the risk avoidance response evaluation system for iterative / dynamic safety decision-making.

[0066] Specifically, the extreme safety system for intelligent vehicles under extreme working conditions proposed by the present invention can improve the emergency response capability under extreme working conditions, such as Figure 1 As shown,

[0067] Module 10 provides the architecture of the vehicle sensor system that supports extreme safety functions, which consists of cameras, millimeter-wave radars, and combined inertial navigation IMUs. It transmits data with the autonomous driving system and high-level driver assistance systems through Ethernet and CAN FD integration.

[0068] Module 21 provides the architecture of the extreme safety function decision module, which is composed of multiple modules such as intervention judgment, comprehensive evaluation and optimization solution. Module 21 obtains environmental information from module 10 and vehicle status and execution information from module 22.

[0069] Module 22 provides an architecture for the execution of extreme safety functions, which consists of multiple modules including data processing, input fusion and collaborative execution. Module 22 obtains decision information from module 21. In this embodiment, module 21 receives input information based on a data-driven reinforcement learning strategy in an unknown environment.

[0070] In this embodiment, module 10 detects a sharp curve environment with a diameter of 11m and sends relevant information to the extreme safety function decision module. The extreme cornering judgment module gives an activation flag and transmits it to the extreme safety decision module. The extreme safety decision module triggers the extreme safety function based on the extreme cornering conditions, vehicle speed and driver intention, with the goal of ensuring comprehensive safety of extreme drift cornering.

[0071] After the extreme safety decision module intervenes, the vehicle stability function and other active safety functions such as collision avoidance are blocked.

[0072] It is understandable that in this embodiment, only the items related to the embodiment scenario in the evaluation system are listed. Specifically, only the safety and performance evaluation indicators of the extreme cornering scenario are listed.

[0073] In this embodiment, the evaluation indexes related to extreme cornering performance and safety include immediate evaluation and final evaluation. The immediate cornering performance index includes: evaluation items r for tracking the suboptimal trajectory of the scene p :

[0074]

[0075] Among them, k pl and k pv is a negative parameter; l is the lateral deviation of the vehicle relative to the center of the curve in the Frenet coordinate system, v is the current vehicle speed; l ref (s) is the lateral displacement of the suboptimal trajectory of the formation at the current bend center; is the maximum vehicle speed under a specific stability constraint under the current attachment. This reward encourages driving along the pre-optimized path at the maximum speed allowed by the stability constraint, thereby improving the exploration efficiency during the iterative optimization process.

[0076] The instant evaluation item r used to reward high sideslip angles β , encouraging the vehicle to turn with a higher center of mass slip angle:

[0077]

[0078] Among them, k β is a negative number, v x is the longitudinal speed in the vehicle coordinate system, v y is the lateral speed in the vehicle coordinate system, and β is the sideslip angle of the vehicle's center of mass.

[0079] The final reward item r is used to reward the comprehensive performance of the ultimate safety function t , encourages safe and fast cornering:

[0080] r t =(1-χ)·k t1 +k t2 ·χ·(t f -t ref )

[0081] Here, χ represents the parameter of the final state of the vehicle: χ = 1 means that the vehicle has safely completed the turning task, while χ = 0 means that an unsafe event has occurred, such as a collision with the track boundary or a rollover. In addition, t f It represents the time required for the vehicle to reach the final state, t ref represents the total travel time of the pre-optimized trajectory. The constant k t1 and k t2 are negative and positive values, respectively, used to penalize unsafe terminal states and reward the shortest possible extreme turning time.

[0082] In this embodiment, the extreme safety solver adopts an end-to-end iterative solver based on reinforcement learning, and the output is directly action information. The constructed state space and action space are determined according to the actual situation of the deployed vehicles. In order to use common technologies in the field of reinforcement learning optimization, this embodiment will not be expanded in detail. In this embodiment, the iterative method is the Critic and Actor deep neural network. The Critic network parameters are trained by minimizing the time difference loss function; the Actor network parameters are trained by maximizing the value function. The above is only a specific example given in this embodiment, and the strategies of imitation learning and reinforcement learning are included in this embodiment.

[0083] In this embodiment, the data acquisition environment used in the extreme safety solver is the Carsim platform, in which the vehicle parameters are common parameters of Class-C, and the interfaces of the stability function are all open. This embodiment is also compatible with self-built vehicle dynamics models or other dynamic response data acquisition platforms.

[0084] In this embodiment, the friction coefficient in the real environment is close to but not completely consistent with that of the data acquisition platform, and the inconsistency of tire types is introduced as an interference item.

[0085] In order to achieve precise control of cornering, the control strategy based on reinforcement learning is deployed to the corresponding scene in the data collection environment to obtain the optimal trajectory T in the current scene. p , this trajectory is used as the target trajectory information of the extreme safety function execution module:

[0086] T p ={S i |simulate(Pv ,V c ,S ini ,π),i=1,2,3,…,n}

[0087] Where S represents the vehicle state along the estimated trajectory, S ini represents the vehicle state when entering a curve. Note that since the estimated trajectory is used to generate feedback input, T p Convert to Cartesian coordinates The conversion details are omitted here.

[0088] This embodiment uses a proportional-integral-derivative controller to track the trajectory Get supplementary input a PID Please note that this embodiment includes all other methods of obtaining trajectory tracking input, including but not limited to linear quadratic Gaussian controller and model predictive controller. Specifically, the calculated input will be further processed by the input fusion mechanism.

[0089] In this embodiment, the end-to-end action output by reinforcement learning will be used as an important feedforward reference and recorded as a RL This input usually has good global optimality and worrying environmental adaptability. Specifically, the input will be directly passed to the input fusion mechanism for further processing.

[0090] In this embodiment, since the hierarchical decision and the data-driven decision have the same goal, a direct addition method is used to obtain the execution input:

[0091] a t =a RL +a PID

[0092] In this embodiment, for the input to be executed a output by the fusion control mechanism, t In terms of driving, this embodiment adopts a combination of four-wheel drive and rear-wheel drive to accurately control the slip of the drive wheels, thereby expanding the control boundary of the vehicle under extreme driving conditions. Through this driving method, the vehicle can maintain optimal power output and stability under various complex road conditions. In terms of the braking system, the system integrates traditional friction braking with modern motor braking technology, which not only provides precise braking force, but also increases the redundancy of braking to ensure high safety under extreme braking conditions. In terms of steering technology, this embodiment combines active steering system and differential steering technology, which greatly enhances the vehicle's handling and stability in emergency situations, and provides maximum posture adjustment capabilities for extreme safety functions.

[0093] Furthermore, each actuator of the vehicle performs extreme safety maneuvers according to the target input of the coordination mechanism. These operations and the resulting scenario evolution information will be transmitted to the previously established risk avoidance response evaluation system, which evaluates the vehicle's safety response in real time and makes iterative or dynamic safety decision adjustments. This process ensures that the vehicle can take the most appropriate risk avoidance measures in changing environments and emergencies, maximizing the emergency response capability and backup level of intelligent vehicles under extreme conditions.

[0094] The performance of this embodiment in the extreme cornering scenario is as follows Figure 2 and Figure 3 shown. Figure 3 In the comparison trajectory shown, it can be found that the control effect of the strategy proposed in this embodiment is significantly better than the control effect of the data-driven strategy alone and the control effect of the feedback controller alone. Among the three comparison strategies, only the solution of this embodiment can control the vehicle to pass through a 180-degree U-turn curve with a diameter of 11 meters at a limit state of high side slip angle.

[0095] The beneficial effects of the present invention are:

[0096] 1. Improve emergency handling capabilities and backup levels: The present invention integrates advanced dynamic control technology to allow intelligent vehicles to demonstrate superior emergency handling capabilities under extreme conditions. This includes but is not limited to efficient response capabilities in emergency braking, extreme steering or high-speed cornering. In addition, through the backup mechanism of multiple braking and driving systems, the safety and reliability of the vehicle in extreme conditions are improved.

[0097] 2. Optimize decision-making and control accuracy: The modular extreme safety function decision-making and control architecture integrates the decision intervention model and extreme safety function optimizer for extreme scenarios, providing accurate decision support for a wide range of extreme scenarios. This enables the vehicle to make the most appropriate dynamic response in various extreme environments, significantly improving the accuracy and efficiency of control.

[0098] 3. Enhance the vehicle's dynamic performance: Through advanced sensing systems and control algorithms, the present invention enables the vehicle to operate close to the dynamic limit, such as high-speed cornering and emergency obstacle avoidance, making full use of the vehicle's motion potential. This not only enhances the vehicle's performance, but also greatly expands its operating range and safety margins.

[0099] 4. Achieve high integration and system synergy: Through a fusion control mechanism that integrates the advantages of various solution technologies, the present invention achieves cross-system collaborative execution. This synergy not only improves the consistency and coordination of operations, but also optimizes the response speed and efficiency of the entire vehicle system, especially when dealing with complex and changing driving environments.

[0100] According to the extreme safety system for intelligent vehicles under extreme working conditions of an embodiment of the present invention, operations including but not limited to those involving critical dynamic stability or extreme motion capabilities are performed under extreme working conditions to fully utilize the motion potential of various systems of the intelligent vehicle and enhance the emergency handling capability and backup level of the intelligent vehicle under extreme working conditions.

[0101] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0102] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

Claims

1. An extreme safety system for intelligent vehicles under extreme working conditions, characterized in that: include: The vehicle sensor system that supports extreme safety functions is used to collect environmental information under extreme vehicle conditions and send the environmental information to the autonomous driving system and high-level driver assistance system in real time via Ethernet or CAN communication; The automatic driving system and the high-level auxiliary driving system include an extreme safety function decision module and an extreme safety function execution module, wherein; The extreme safety function decision module includes an extreme collision avoidance judgment module, an extreme cornering judgment module, an extreme vehicle condition recognition module, an extreme state recognition module and a safety boundary monitoring module; the extreme safety function decision module is used to obtain environmental information and receive input information based on a data-driven reinforcement learning strategy in an unknown environment as well as vehicle status and execution information fed back by an extreme safety function execution module; The extreme safety function execution module is used to obtain the decision information sent by the extreme safety function decision module and output the vehicle status and execution information.

2. The system according to claim 1, characterized in that The vehicle sensing system that supports extreme safety functions includes cameras, millimeter-wave radars and combined inertial navigation IMU.

3. The system according to claim 1, characterized in that The extreme safety function decision module is also used for: The target trajectory is solved by an extreme safety decision solver based on the evaluation results and using an iterative optimization or dynamic optimization method; wherein the evaluation results include the characteristics of the extreme scenario, and the evaluation results are obtained by evaluating the risk avoidance response of the vehicle in the extreme scenario in combination with the direct risk of the scenario, the expected risk of the scenario, the risk avoidance level of the operation, and the degree of the limit of the operation; All target trajectories of decision solutions matching the current extreme scenario are passed to the extreme safety function execution module.

4. The system according to claim 3, characterized in that The system is also used to fuse hierarchical decision control and data-driven decision control using a fusion control mechanism; wherein, The fusion control mechanism is designed for situations where hierarchical decision control and data-driven decision control have the same objectives: deploying decisions and generating trajectories in a data generation environment, replacing the trajectory information output by hierarchical decision making with trajectories; using a model-based trajectory tracker to generate closed-loop inputs Directly generate feed-forward input using data-driven The two are added or weighted to generate the final input to be executed: The fusion control mechanism is aimed at the situation where the hierarchical decision control and data-driven decision control have different goals: by comparing the goal gap between data-driven decision and hierarchical decision, the data-driven strategy directly generates action information a Data , track the hierarchical decision target trajectory to obtain action information a model ,Based on the target difference, the final input to be executed is generated, and the generation methods include neural network fusion and weighted fusion: a t =Difft target (a Data ,a model )。 5. The system according to claim 4, characterized in that For the execution input of the fusion control mechanism output, the four-wheel drive and rear-wheel drive are combined in driving to control the drive slip and expand the vehicle's extreme driving control boundary; the friction brake and motor brake are combined in braking to provide backup extreme braking capacity guarantee; the active steering and differential steering are combined in steering to provide the maximum posture adjustment capability for extreme safety functions; Each actuator of the vehicle executes the target input and performs extreme safety maneuvers.

6. The system according to claim 1, characterized in that The environmental information is a sharp bend environment, and the goal of the extreme safety decision module is to ensure comprehensive safety during extreme drift bends; The evaluation indicators related to extreme cornering performance and safety include immediate evaluation and final evaluation: The immediate cornering performance indicators include: the evaluation item r used to track the suboptimal trajectory of the scene p : Among them, k pl and k pv is a negative parameter; l is the lateral deviation of the vehicle relative to the center of the curve in the Frenet coordinate system, v is the current vehicle speed; l ref (s) is the lateral displacement of the suboptimal trajectory of the formation at the current bend center; is the maximum vehicle speed under the current stability constraint.

7. The system according to claim 6, characterized in that The instant evaluation item r used to reward high sideslip angles β , encouraging the vehicle to turn with a higher center of mass slip angle: Among them, k β is a negative number, v x is the longitudinal speed in the vehicle coordinate system, v y is the lateral speed in the vehicle coordinate system, β is the sideslip angle of the vehicle’s center of mass; The final reward item r is used to reward the comprehensive performance of the ultimate safety function t , encourages safe and fast cornering: r t =(1-x)·k t1 +k t2 ·x·(t f -t ref ) Where χ represents the parameter of the final state of the vehicle: χ = 1 means that the vehicle has safely completed the turning task, χ = 0 means an unsafe event has occurred, t f It represents the time required for the vehicle to reach the final state, t ref represents the total travel time of the pre-optimized trajectory; the constant k t1 and k t2 are negative and positive values, respectively, used to penalize unsafe terminal states and reward the shortest possible extreme turning time.

8. The system according to claim 3, characterized in that The extreme safety decision solver adopts an end-to-end iterative solver based on reinforcement learning, and the output is directly action information; the iterative method is a Critic and Actor deep neural network, and the Critic network parameters are trained according to minimizing the time difference loss function; the Actor network parameters are trained by maximizing the value function.

9. The system according to claim 3, characterized in that The data acquisition environment used in the extreme safety decision solver is the Carsim platform; the control strategy based on reinforcement learning is deployed to the corresponding scene in the data acquisition environment to obtain the optimal trajectory T in the current scene. p , the trajectory is used as the target trajectory information of the extreme safety function execution module: T p ={S i |simulate(P v ,V c ,S ini ,π),i=1,2,3,…,n} Where S represents the vehicle state along the estimated trajectory, S ini Indicates the vehicle status when entering a curve.

10. The system according to claim 9, characterized in thatSince the estimated target trajectory is used to generate feedback input, T p Convert to Cartesian coordinates Tracking Trajectory Using a Proportional-Integral-Derivative Controller Get supplementary input a PID ; The end-to-end actions output by reinforcement learning will serve as an important feedforward reference and will be recorded as a RL ; Hierarchical decision making has the same goal as data-driven decision making, and uses direct addition to obtain execution input: a t =a RL +a PID 。

Images