Electric vehicle obstacle avoidance control method and system

Abstract

The invention relates to the technical field of automobile active obstacle avoidance, and provides an electric automobile obstacle avoidance control method and system, and the method comprises the steps: obtaining obstacle data, and generating a vector model of an obstacle; projecting the obstacle to the relative motion direction of the current vehicle based on the vector model, and judging whether scraping and collision occur relative to the operation parameters of the current vehicle; when the scraping and collision probability exceeds a preset risk value, generating an obstacle avoidance control instruction deviating from obstacle projection; wherein the obstacle avoidance control instruction comprises a steering angle automatic response instruction of the current vehicle under the current path curvature and a self-feedback torque enhancement instruction based on road surface friction. Through vector model projection and relative motion direction analysis, dynamic evaluation of obstacle collision probability and cooperative control of steering angle automatic response and torque enhancement, path correction and power output are considered, different road conditions (such as ice and snow / wet and slippery road surfaces) are automatically adapted, and active intervention is performed before a preset risk threshold value is triggered.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive obstacle avoidance control, and particularly relates to an electric vehicle obstacle avoidance control method and system thereof. Background Art

[0002] With the rapid development of intelligent driving technology, the active obstacle avoidance system of electric vehicles has become a core module for ensuring driving safety. Traditional obstacle avoidance technologies mainly rely on fixed-threshold collision detection, such as the AEB automatic emergency braking system, and simple path planning algorithms, such as the artificial potential field method. The core logic of traditional solutions is to detect the distance of obstacles through radar or cameras, and trigger braking or steering commands when the distance is lower than a preset safety value.

[0003] However, traditional technologies usually represent obstacles using static bounding boxes and cannot accurately characterize the spatio-temporal evolution characteristics of moving obstacles, such as crossing pedestrians and lane-changing vehicles. For example, in a curved road scenario, the yaw motion of the vehicle itself causes the relative position of the obstacle to continuously change. Since traditional methods do not introduce a vehicle motion state compensation mechanism, the projection trajectory error of the obstacle exceeds 1.2 meters, which is extremely likely to cause misjudgment. Mainstream algorithms, such as the TTC time-to-collision model, only calculate the risk value based on the linear relationship between relative speed and distance, ignoring vehicle dynamics characteristics, such as the influence of tire cornering stiffness and load transfer on the actual obstacle avoidance ability. Experiments show that in an emergency obstacle avoidance on a low-adhesion road surface, traditional methods do not consider the tire grip limit, and the actual braking distance is much higher than the theoretical value. For some automobile manufacturers, the actual braking distance is more than 40% higher than the advertised value, resulting in a sharp increase in the collision risk.

[0004] In the existing patent CN202411192597.4, digital twin technology is adopted to achieve obstacle avoidance of vehicles. It mainly uses the trajectory prediction of obstacles and the motion state of the vehicle. Based on the speed, distance, and trajectory of the obstacles, it determines how to avoid the obstacles. However, it does not consider the road surface friction condition and the torque condition of the vehicle. Therefore, there is also an actual braking distance, which often exceeds the vehicle's own avoidance calculation. In a scenario where a collision may occur, during avoidance, because the actual braking distance exceeds the vehicle's own calculation result, a collision will occur. Although existing automobile manufacturers have set a relatively large braking distance in case of a collision to eliminate the deficiency of the mismatch between the calculation result and the vehicle and road surface friction state, in extreme scenarios, such as rainy and snowy weather, collisions are still likely to occur;

[0005] In the existing patent CN202210459117.0, it is proposed to use ultrasonic waves to achieve obstacle avoidance. This is a theoretical technology that can also achieve obstacle avoidance prediction. However, it is similar to the lidar technology in the existing technology, and even has a lower effect than the lidar technology because the ultrasonic technology is related to the temperature, air pressure, and air density of the air environment, with a large error. Although theoretically this is a perfect technology, the existing technical conditions cannot achieve the theoretical effect. Summary of the Invention

[0006] The present application proposes an obstacle avoidance control method and system for electric vehicles. By fusing multi-source sensors to construct an obstacle vector model, collision determination is realized. When a collision is likely to occur, automatic braking is used to reduce the speed for obstacle avoidance or torque enhancement is used to instantaneously increase the speed for obstacle avoidance.

[0007] In a first aspect, an obstacle avoidance control method for an electric vehicle according to the present invention includes:

[0008] Obtain obstacle data and generate a vector model of the obstacle;

[0009] Based on the vector model, project the obstacle onto the relative motion direction of the current vehicle, and determine whether a scrape will occur with respect to the operating parameters of the current vehicle;

[0010] When the scrape probability exceeds a preset risk value, generate an obstacle avoidance control instruction deviating from the obstacle projection; wherein, the obstacle avoidance control instruction includes an automatic response instruction for the steering angle of the current vehicle under the current path curvature and a self-feedback torque enhancement instruction based on the road surface friction.

[0011] In this embodiment, the present application constructs an obstacle vector model through multi-source sensor fusion, performs dynamic compensation in combination with the vehicle motion state, and then conducts a coupling analysis of the vehicle's real-time dynamic parameters and the obstacle's motion characteristics to establish a probabilistic scrape determination model, which associates the risk value calculation with the actual obstacle avoidance ability of the vehicle. Furthermore, by combining the path curvature constraint steering control with the road surface friction adaptive torque enhancement, the drive wheel torque distribution is synchronously optimized during steering avoidance, reducing the obstacle avoidance trajectory tracking error and avoiding control failure caused by actuator response lag.

[0012] Combined with the first aspect, the vector model is constituted by the spatio-temporal evolution of the dynamic characteristic parameters of the three-dimensional contour model of the obstacle generated after multi-sensor fusion within the current vehicle; wherein, the three-dimensional contour model is a spatio-temporal extended model of the obstacle within the relevant contact distance containing the current vehicle and the obstacle generated after dynamically compensating the vehicle's real-time motion state. When performing dynamic compensation of the motion state, within the relevant contact distance, the current vehicle performs vector marking based on the vehicle's real-time motion state and the obstacle's motion state, and uses the vector marking as the extended range marking of the spatio-temporal extended model.

[0013] In this embodiment, the present application extends the decision-making time through multi-sensor fusion to obtain better decisions. Conventionally, lidar and visual recognition are complemented with each other to eliminate the errors of a single sensor. For the relative position perception deviation caused by the yaw movement of the vehicle (typical scenario: passing vehicles on a curve), the spatial coordinates of the obstacle are corrected in real time to improve the projection trajectory prediction accuracy. The dynamic characteristics of the obstacle and the motion state of the vehicle are jointly modeled to improve the margin of decision-making time during the interaction process.

[0014] Combined with the first aspect, the obstacle projection includes configuring a dynamic safety corridor based on the current vehicle.

[0015] Among them, the dynamic safety corridor is constructed based on the origin of the current vehicle in the lane coordinate system. The dynamic safety corridor is configured with a collision limitation mechanism, and the collision limitation mechanism is based on the coupling coefficient between the projection area under the real-time motion state parameters of the vehicle and the vehicle envelope.

[0016] The width of the dynamic safety corridor is dynamically adjusted according to the road surface adhesion coefficient and the vehicle yaw moment margin, and the coupling coefficient is used to calculate the scraping probability by weighting.

[0017] In this embodiment, a dynamic safety corridor is constructed between the obstacle projection and the vehicle itself, which combines the adhesion of the road surface and the ability of the yaw moment, avoiding the vehicle from adhering to overly conservative and adventurous behaviors, but automatically increasing the lateral safety distance and establishing a direct collision mapping based on physical characteristics to more accurately determine the collision probability.

[0018] Combined with the first aspect, projecting the obstacle to the relative motion direction of the current vehicle based on the vector model further includes:

[0019] Configuring an offset determination mechanism based on virtual piloting, and the offset determination mechanism is used to generate a planning target point for the vehicle path in the reverse direction of the obstacle projection direction;

[0020] The offset determination mechanism is constrained by the steering gradient of the current path curvature change rate, and the correction amplitude of the steering angle is restricted not to be greater than a preset amplitude value;

[0021] The constraint condition is associated with the planning target point of the vehicle path and dynamically coupled with the vehicle heading angle deviation.

[0022] In this embodiment, the present application sets an offset determination mechanism to limit the correction amplitude of the steering angle, thereby reducing the speed peak of the steering wheel and improving driving comfort. And when avoiding obstacles, through the planning target point and the constraint condition, emergency obstacle avoidance can be achieved continuously and smoothly, and at the same time, the steering assist torque can be automatically activated to accelerate the steering wheel return speed.

[0023] Regarding the first aspect, the determination of whether a collision will occur with respect to the operating parameters of the current vehicle includes:

[0024] Obtain the real-time motion parameters of the vehicle and construct a dynamic collision model based on the obstacle projection; wherein, the collision model includes the probability collision area of the current vehicle relative to the safety profile in the moving direction;

[0025] According to the dynamic collision model, determine the collision probability field based on the current vehicle trajectory;

[0026] Through the probability collision field, determine the spatio-temporal overlap risk value of the current vehicle in the probability collision area, and output the real-time corrected collision probability.

[0027] In this embodiment, the present application constructs a dynamic collision model through obstacle projection, thereby determining the probability collision area of the vehicle in the moving direction. Furthermore, the collision probability can be judged and corrected in real time by means of spatio-temporal overlap.

[0028] Regarding the first aspect, the preset risk value is based on a multi-level risk threshold division, and each level of threshold in the multi-level risk threshold corresponds to a different obstacle avoidance mode; wherein, the trigger boundary of the obstacle avoidance mode corresponding to each level of threshold in the multi-level risk threshold is determined by the steering wheel operation tendency parameter of the current vehicle.

[0029] In this embodiment, the present application adopts progressive risk avoidance, corresponding to progressive emergency braking triggering, so there is no abrupt emergency braking triggering, thereby improving driving comfort. The steering wheel operation tendency determines the driver's own style. When dynamically adjusting the threshold boundary, it is more in line with the driver's driving habits and can also improve safety.

[0030] Regarding the first aspect, the generation of an obstacle avoidance control command deviating from the obstacle projection includes:

[0031] Based on the real-time collision probability and the multi-level risk threshold, determine the target obstacle avoidance mode and the obstacle avoidance control command corresponding to the target obstacle avoidance mode;

[0032] The vehicle braking command is based on the correlation between the probability collision coefficient of the probability collision area and the vehicle yaw rate. When it is positively correlated, the vehicle braking command is an acceleration braking command based on self-feedback torque enhancement;

[0033] When it is negatively correlated, the vehicle braking command is a braking response delay command based on the steering angle automatic response command.

[0034] In this embodiment, when the present application generates the obstacle avoidance control command, through the correlation between the probability collision coefficient and the vehicle yaw rate, that is, the fusion of the environmental perception control command and the closed-loop self-adaptation, it avoids the problems of overshoot and overcompensation in the traditional technical solutions.

[0035] In combination with the first aspect, the generation of the obstacle avoidance control instruction deviating from the obstacle projection further includes:

[0036] Construct a first Kalman prediction model based on the motion acceleration of the obstacle, and generate a trajectory cloud map of the obstacle within the prediction time period; wherein, the resolution of the trajectory cloud map is adaptively adjusted according to the sensor sampling frequency and the vehicle speed, and the superposition analysis result is used to correct the confidence weight of the vector model;

[0037] Determine the avoidance space and the response instruction of the obstacle avoidance control instruction corresponding to the avoidance space according to the trajectory cloud map and the motion braking area of the current vehicle.

[0038] In this embodiment, the present application realizes the modeling of the obstacle trajectory through the first Kalman prediction model, determines the obstacle trajectory within a period of time in the future, and then realizes the error limited within the reaction time through the trajectory cloud map formed by the modeling, and identifies the conflict point in advance, thereby improving the success rate of obstacle avoidance.

[0039] In combination with the first aspect, the generation of the obstacle avoidance control instruction deviating from the obstacle projection further includes:

[0040] Establish a steering-torque coupling control domain, and define a control parameter priority matrix under different risk levels;

[0041] Use the model predictive control algorithm to constrain and decouple the steering angle change rate and the torque response rate;

[0042] Wherein, the dimension of the priority matrix includes the vehicle lateral acceleration threshold value and the drive motor thermal load state parameter.

[0043] In this embodiment, the present application improves the conflict resolution efficiency of multiple control objectives by constructing a priority matrix for the integrated control of steering and torque, realizes collaborative constraints, and prevents the electrode overheating fault caused by continuous obstacle avoidance.

[0044] On the other hand, the present application proposes an electric vehicle obstacle avoidance control system, including:

[0045] Vector model building module: Obtain obstacle data and generate a vector model of the obstacle;

[0046] Collision determination module: Project the obstacle onto the relative motion direction of the current vehicle based on the vector model, and determine whether a scrape will occur relative to the running parameters of the current vehicle;

[0047] Obstacle avoidance control module: Used to generate an obstacle avoidance control instruction deviating from the obstacle projection when the scrape probability exceeds a preset risk value; wherein, the obstacle avoidance control instruction includes an automatic steering angle response instruction of the current vehicle under the current path curvature and a self-feedback torque enhancement instruction based on the road surface friction.

[0048] In this embodiment, an obstacle vector model is constructed through multi-source sensor fusion, and dynamic compensation is performed in combination with the vehicle motion state. Through the coupled analysis of the vehicle's real-time dynamic parameters and the motion characteristics of obstacles, a probabilistic scraping collision determination model is established, which associates the risk value calculation with the vehicle's actual obstacle avoidance ability. Furthermore, the path curvature constraint steering control is combined with the road surface friction adaptive torque enhancement to synchronously optimize the drive wheel torque distribution during steering avoidance, reduce the obstacle avoidance trajectory tracking error, and avoid control failure caused by actuator response lag.

[0049] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification and the accompanying drawings.

[0050] The technical solutions of the present invention will be further described in detail below through the accompanying drawings and embodiments. Description of the Drawings

[0051] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0052] Figure 1 is a flowchart of a method for obstacle avoidance control of an electric vehicle in an embodiment of the present invention;

[0053] Figure 2 is a process diagram of the generation process of a vector model in an embodiment of the present invention;

[0054] Figure 3 is a process diagram of the safety configuration of obstacle projection in an embodiment of the present invention;

[0055] Figure 4 is a schematic diagram of path dynamic planning based on a virtual leading point in an embodiment of the present invention;

[0056] Figure 5 is a process diagram of constructing a dynamic probability collision field in an embodiment of the present invention;

[0057] Figure 6 is a decision-making process diagram of a multi-level obstacle avoidance decision-making for realizing guidance in human-machine co-driving in an embodiment of the present invention;

[0058] Figure 7 is a process diagram of dynamic decision-making of risk and control coupling in an embodiment of the present invention;

[0059] Figure 8 is a process diagram of dynamic trajectory prediction and avoidance space optimization in an embodiment of the present invention;

[0060] Figure 9 This is the process diagram for constructing multi-constraint coupling in the embodiments of the present invention;

[0061] Figure 10 This is the system composition diagram of an obstacle avoidance system for an electric vehicle in the embodiments of the present invention. Detailed implementation manners

[0062] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0063] In the prior art, the obstacle avoidance methods for current traditional fuel vehicles and new energy vehicles are all limited by the static and discrete design concepts. However, the driving environment is dynamically complex. In driving, in dynamic complex scenarios, many defects will be exposed. Traditional solutions mostly adopt a collision detection mechanism with a fixed safety distance threshold and a single sensor, and adopt a linear control logic of "perception - braking".

[0064] For example, when an obstacle vehicle or pedestrian enters the preset distance range, the vehicle will directly trigger the maximum braking force, ignoring the dynamic coupling effects of multi-dimensional parameters such as the vehicle motion state, road surface friction coefficient, and driver operation intention. A rigid strategy is adopted, and in scenarios such as high-speed lane change and slippery road surface, problems such as an increase in false trigger rate and control instruction conflict are likely to occur, and even vehicle instability may be caused.

[0065] In actual implementation, traditional methods also rely on binary collision judgment (collision or not), and do not perform quantitative evaluation of risk probability, making it difficult to achieve smooth control of human-machine collaboration. The obstacle avoidance failure rate is as high as... in complex urban scenarios.

[0066] Based on the defects of traditional technologies, the present application constructs a trinity obstacle avoidance control technology framework of dynamic perception, probability evaluation, and coupling control.

[0067] During perception, to prevent the emergence of traditional single-sensor data islands, the present application constructs a vector model of obstacles through the spatio-temporal synchronous fusion of data from multi-source heterogeneous sensors, including but not limited to lidar, millimeter-wave radar, vision and other sensing devices. For example, in view of the high missed detection rate of traditional vision systems in backlight environments, the present application adopts a cross-modal fusion algorithm of lidar point cloud and vision semantics to improve the obstacle recognition accuracy.

[0068] At the decision-making level, the present application constructs a joint calculation model of a dynamic safety corridor and a probability collision field. Based on the real-time yaw moment margin of the vehicle and the road surface adhesion coefficient, the width of the safety corridor is dynamically adjusted, and the collision probability field is calculated through a spatio-temporal grid algorithm to quantify the risk level. During the actual test process, in the scenario of a target vehicle suddenly changing lanes, the risk prediction accuracy is improved compared with traditional bounding box detection.

[0069] At the execution level, the present application constructs a design steering-torque coupling control domain to avoid command conflicts caused by traditional discrete control. For example, when avoiding obstacles on a wet and slippery curve, the vehicle automatically triggers a torque step enhancement strategy according to the dynamic correlation between the yaw angular velocity and the collision coefficient, and at the same time restricts the change rate of the steering angle, improving the yaw stability of the vehicle.

[0070] Embodiment 1:

[0071] The present application proposes an obstacle avoidance control method for electric vehicles. Refer to Figure 1 : The implementation process of the method of the present application is as follows:

[0072] First, the present application obtains the three-dimensional contour data of the obstacle through multi-sensor fusion and constructs an obstacle vector model. Then, the dynamic projection algorithm is used to map the obstacle contour to the vehicle motion coordinate system, and the scraping probability is calculated in combination with the real-time motion parameters (speed, yaw angle) of the vehicle. Finally, when the probability exceeds a preset threshold, a composite obstacle avoidance command including a steering angle response and torque enhancement is generated.

[0073] In actual implementation, obtaining the three-dimensional contour data of the obstacle through multi-sensor fusion is achieved through the heterogeneous data fusion of lidar, millimeter-wave radar, and vision sensors. Based on the spatio-temporal alignment algorithm, the sensor parallax is eliminated to generate a dynamic model including the three-dimensional contour and motion speed vector of the obstacle.

[0074] Compared with the traditional solution, the present method introduces a motion compensation mechanism. When the vehicle accelerates, decelerates or steers, the inertial measurement unit (IMU) is used to correct the coordinate system offset of the vehicle in real time to ensure the spatio-temporal consistency of the projected position of the obstacles faced by the vehicle. Then, when projecting the obstacle vector model onto the vehicle's motion direction, a non-linear coordinate system transformation algorithm is adopted. A virtual motion trajectory prediction model is built through the vehicle's real-time yaw rate and steering wheel angle, which is similar to digital twin. However, in this application, the spatio-temporal overlapping area between the obstacle envelope and the vehicle contour, as well as the friction of the vehicle road surface and the torque of the vehicle itself, are calculated to achieve the linkage of the three. When the collision meter makes a judgment, this application quantifies the scraping risk based on the probability density function, rather than the traditional binary judgment, so it can effectively distinguish between "potential risk" and "emergency risk" scenarios. When the risk value exceeds the preset threshold, this application will synchronously generate a steering angle response and a torque enhancement instruction. Furthermore, the path curvature tracking algorithm is used for steering control to ensure the dynamic matching of the steering angle change rate with the current vehicle speed and road surface adhesion coefficient. A self-feedback mechanism is introduced into the torque enhancement module. By monitoring the motor speed and battery output power of the vehicle in real time, the torque gradient is dynamically adjusted to avoid wheel slip caused by torque mutation.

[0075] In the actual implementation process, for example, on an urban road, when a vehicle using the method of this application is driving at a medium speed and a stationary cone suddenly appears in the right front. The vehicle will identify the cone contour through multi-sensor fusion and calculate the current trajectory of the vehicle through dynamic projection to find out whether there will be a scrape with the obstacle after 1.5 seconds or some other seconds. The probability assessment module outputs a risk value. At this time, the vehicle's on-board system will generate a composite instruction:

[0076] For example: the steering module is slightly adjusted to the left by 2.5° to smoothly deviate from the original path, and at the same time, the drive motor applies progressive torque enhancement to the right rear wheel (to avoid yaw instability caused by unilateral braking).

[0077] The whole process is completed within a very short time, for example, within 200 ms. Then the vehicle bypasses the obstacle along a continuous curvature path, so that the driver only feels a slight change in the steering wheel assist and there will be no abrupt steering or braking impact. Therefore, the dual-modal response of steering and torque in this application can achieve the synergistic effect of "avoiding first and then stabilizing", reducing the lateral acceleration mutation compared with the traditional single braking strategy. The self-feedback mechanism of this application ensures the compatibility between the control instruction and the driver's operation intention. When the driver actively steers, the on-board system automatically reduces the intervention intensity to avoid competing for control.

[0078] Embodiment 2:

[0079] The present application proposes a generation method of a vector model. Refer to Figure 2Using a spatio-temporal evolution algorithm, this application fuses the radial velocity data of millimeter-wave radar with the semantic information of vision to generate a three-dimensional contour model with motion compensation. Furthermore, an extended Kalman filter is used to vector-label the relative motion states of the vehicle and obstacles, constructing a spatio-temporal extended model.

[0080] In actual implementation, a three-dimensional contour model is constructed through the fusion of three-dimensional data from lidar, millimeter-wave radar, and vision, breaking through the limitations of the traditional single-sensor perspective. The lidar in this application provides a high-precision point cloud contour, while the millimeter-wave radar collects the radial motion velocity, and the vision device of the vehicle recognizes semantic attributes. For example: vehicles, pedestrians.

[0081] The spatio-temporal alignment module of this application uses a quaternion interpolation algorithm to map heterogeneous data from multiple sensors to a unified spatio-temporal coordinate system.

[0082] For example, when the vehicle is turning, the inertial measurement unit (IMU) of the vehicle outputs the yaw angular velocity in real time, dynamically correcting the timestamp deviation of each sensor data, which can avoid coordinate misalignment caused by changes in the vehicle body posture.

[0083] During the three-dimensional contour modeling process of this application, a two-way motion compensation mechanism for the vehicle and obstacles is introduced, preferably using a Kalman filter (EKF) to perform synchronous prediction to determine the vehicle's own motion trajectory (acceleration, yaw angle), as well as the motion trend of obstacles, such as: velocity vector, heading angle.

[0084] Through the spatial transformation of space-time, the obstacle contour is transformed from the global coordinate system to the relative motion coordinate system centered on the vehicle.

[0085] For example, when the vehicle approaches a laterally moving obstacle along a curved trajectory, the vehicle system compensates in real time for the perspective shift caused by the vehicle's steering, thus ensuring that the projected position of the obstacle contour in the relative coordinate system is consistent with the real physical space.

[0086] If within the critical contact distance range, vector field marking will be performed on the obstacle surface. The vector field marking includes a direction vector for pointing to the vehicle's motion direction and an intensity vector for determining the property proportional to the relative velocity.

[0087] The spatio-temporal extended model of this application analyzes the vector field distribution pattern through a convolutional neural network (CNN) to generate the dynamic influence range of the obstacle.

[0088] For example, when it comes to a stationary vehicle with its door being opened, the vector marking of this application shows a high-density distribution in the direction of the door unfolding. The spatio-temporal extended model predicts the growth trend of the door opening angle over time based on this, and then delimits the risk area in advance.

[0089] In actual implementation, the present application introduces the scenario of a highway. When the user's vehicle is traveling at a relatively high speed, if a truck in the front suddenly deviates into the user's lane. The lidar of the vehicle-mounted computer of the present application captures the contour point cloud of the truck, the millimeter-wave radar detects its lateral speed, and the vision system identifies the turn signal of the truck. Then, the spatio-temporal alignment module fuses multi-source data, dynamically compensates for the coordinate offset caused by the acceleration of the vehicle itself, and generates a three-dimensional contour model of the truck. The motion state analysis module will predict how many seconds it will take for the truck to invade the user's lane and what the invasion width is, and a vector marker forms a high-density field on the right side of the truck. The spatio-temporal expansion model delimits the risk area based on these, and then triggers the obstacle avoidance system to generate a progressive steering command to the left in advance, and automatically adjusts the torque distribution ratio according to the road surface humidity to ensure that the vehicle smoothly deviates from the risk area.

[0090] Embodiment 3:

[0091] The present application proposes a safety configuration method for obstacle projection. Refer to Figure 3 , the present application constructs a core space constraint model for a dynamic safety corridor. The space constraint model takes the vehicle's centroid as the origin, and through high-precision map and vision lane line recognition, establishes a lane relative coordinate system.

[0092] The lane relative coordinate system of the present application is different from the traditional fixed coordinate system. The present application introduces a dynamic compensation mechanism for the vehicle's yaw angular velocity and heading deviation. When the vehicle deviates from the center line of the lane, the lane relative coordinate system automatically tilts and corrects, so as to ensure that the projection reference is consistent with the actual motion trend of the vehicle.

[0093] For example, when driving on a curve, the lane relative coordinate system will rotate dynamically along the curvature of the curve, so that the geometric center line of the dynamic safety corridor always coincides with the vehicle's expected path.

[0094] The width of the dynamic safety corridor is determined by a non-linear function of the road surface adhesion coefficient and the yaw moment margin. The vehicle-mounted computer system monitors the tire slip ratio in real time and infers the current road surface adhesion coefficient value; it also calculates the yaw moment margin through the vehicle dynamics model, and the yaw moment margin indicates the ability boundary of the vehicle to maintain stable steering.

[0095] When the road surface adhesion coefficient decreases, such as on a rainy or snowy road surface, and during an emergency lane change at high speed with insufficient yaw margin, the width of the dynamic safety corridor automatically shrinks, thereby forcibly restricting the vehicle's lateral displacement range. Compared with the static defect of the traditional fixed-width corridor, the present application realizes the dynamic balance between safety and passability.

[0096] The present application performs a spatial convolution operation on the obstacle projection contour and the external expansion components such as the rearview mirror of the vehicle envelope to generate a risk coupling coefficient. The risk coupling coefficient consists of three parts: the geometric overlap degree, the proportion of the contact area between the obstacle and the vehicle envelope;

[0097] In this application, regarding the relevance of motion trends, the cosine value of the angle between the obstacle velocity vector and the vehicle's motion direction, and the reciprocal of the time to collision calculated based on the relative velocity through a time approximation factor. Then, by weighted fusion of various parameters, such as the cosine value of the angle between the obstacle velocity vector and the vehicle's motion direction, a normalized scraping probability value in the range of 0-1 is output to accurately quantify the risk level.

[0098] During actual implementation, based on the solution of this application, a wet and slippery curve scenario on a highway was measured. When the vehicle enters the curve at a high speed, at this time, the road adhesion coefficient is significantly reduced due to the influence of rain. The dynamic safety corridor module of the in-vehicle system performs the following actions:

[0099] First, based on the IMU yaw angular velocity and visual lane line detection, an inclined lane coordinate system is established to correct the deviation between the inner and outer sides of the curve.

[0100] Then, according to the tire slip rate, the decrease in the road adhesion coefficient is inferred, and the corresponding reduction in the yaw moment margin due to the increase in centrifugal force is calculated simultaneously, and the corridor width is contracted within a certain proportion of the standard value.

[0101] If it is detected that the guardrail on the outer side of the curve invades the corridor boundary, the coupling coefficient between the guardrail contour and the vehicle envelope will be calculated. Since the vehicle has a tendency to slip outward (high motion trend relevance) and the road adhesion coefficient rises rapidly, the obstacle avoidance system is triggered to generate a progressive steering command towards the inner side of the curve, and then the torque of the inner wheels is slightly increased to enhance the grip. The entire process completes risk avoidance on the premise that the vehicle does not undergo obvious yaw oscillations.

[0102] Embodiment 4:

[0103] This application proposes a path dynamic planning mechanism based on a virtual leading point. Refer to Figure 4 , based on the spatial inverse mapping principle of the obstacle projection contour, a path planning target point is generated on the opposite side of the obstacle intrusion direction.

[0104] The solution of this application is different from the traditional geometric center point selection. This solution introduces a dynamic potential field algorithm: the obstacle contour is regarded as a high potential energy region, the current position of the vehicle is judged as a low potential energy point, and then the optimal path target point is searched along the direction of the potential energy gradient descent.

[0105] For example, when a stationary obstacle appears on the right side of the vehicle, the system generates a series of virtual leading points in the left lane space to form a progressive offset path.

[0106] To avoid the risk of vehicle instability caused by sudden steering, this application will set a dynamic constraint boundary for the path curvature change rate. Calculate the maximum allowable steering gradient value at the current vehicle speed and road adhesion coefficient through the vehicle dynamics model, and use these data as the hard constraint conditions for path optimization.

[0107] In specific implementation, a cubic spline interpolation algorithm is adopted to generate a continuous curvature path to ensure a smooth transition of the steering angle change rate. When the curvature gradient of the planned path exceeds the critical value, the path replanning module is triggered, and a feasible solution will be searched again within the constraint boundary.

[0108] When this application is implemented, the deviation between the planned target point and the vehicle's real-time heading angle is closed-loop coupled. By designing a proportional-derivative (PD) controller, the steering execution rate is dynamically adjusted according to the azimuth angle deviation of the target point.

[0109] For example, when the heading deviation of the vehicle is large, the steering response speed is increased; when the vehicle approaches the target path, the adjustment amplitude is automatically reduced to achieve the progressive convergence effect of "fast following and slow adjustment". At the same time, a hysteresis compensation mechanism is introduced to eliminate the control oscillation caused by the sensor data delay.

[0110] During actual testing, the implementation process of an urban blind search section is introduced. When the vehicle is driving along the center line of the lane, sudden road construction ahead causes one-third of the right lane to be occupied. The system executes the following control process:

[0111] The contour of the cone obstacle is identified through the lidar point cloud, and a series of virtual leading points are generated in the obstacle-free area on the left to form a virtual path that deviates to the left. Based on the current vehicle speed and the dry road surface condition, that is, the road surface adhesion coefficient is high, the maximum allowable steering gradient is calculated as a moderate value to ensure a gentle change in the path curvature. The deviation between the vehicle's initial heading angle and the virtual path is determined, and the PD controller generates a progressive left-turn instruction. The steering angle is adjusted at a constant rate at a suitable angle per second until the heading deviation converges within 1°. During the process, the torque distribution module synchronously increases the torque of the right wheel to suppress the yaw trend caused by steering. There is no obvious sharp turn in direction or vehicle body shaking during the entire obstacle avoidance process, and the passenger's body feeling is kept stable and natural.

[0112] Embodiment 5:

[0113] This application proposes a method for constructing a dynamic probability collision field to achieve risk quantification. Refer to Figure 5 , based on the speed, yaw angle, acceleration corresponding to the vehicle's real-time motion parameters and the projection contour of the obstacle, this application establishes a spatio-temporal two-dimensional collision model. In the spatial dimension of this application, the vehicle safety contour (including the outer expansion envelope) is discretized into multiple sub-regions along the motion direction; in the time dimension of this application, the trajectory distribution of multiple future moments is predicted based on the vehicle kinematic equation. Different from the traditional geometric model, the model established in this solution introduces a probability density function (PDF) to describe the risk weight of each spatio-temporal unit, forming a continuous risk distribution field.

[0114] For example, when the vehicle is accelerating, the model automatically increases the weight coefficient of the spatio-temporal unit in front to reflect the cumulative collision risk caused by the extended braking distance.

[0115] By performing multi-dimensional superposition on the risk weights of spatio-temporal discrete units, there are spatial overlap degree, motion trend correlation, and time approximation factor in this process:

[0116] The spatial overlap degree represents the geometric intersection area between the obstacle contour and the vehicle safety envelope;

[0117] The motion trend correlation represents the cosine value of the angle between the obstacle velocity vector and the vehicle trajectory;

[0118] The time approximation factor represents the reciprocal of the collision time calculated based on the relative velocity;

[0119] This application fuses the above parameters through a fuzzy logic algorithm to generate a normalized probability collision field, which is then presented in the form of a heat map in three-dimensional space. The high-temperature area represents high-risk spatio-temporal units.

[0120] Finally, a convolutional neural network (CNN) is used to extract features from the probability collision field to identify high-risk aggregation areas. Through a sliding window mechanism, the cumulative risk value of the current vehicle trajectory passing through each spatio-temporal unit is calculated. If the risk value exceeds the preset threshold, the system determines that there is a scraping risk and triggers a hierarchical obstacle avoidance strategy.

[0121] Introduce the following vehicle-following scenario on a highway. When suddenly decelerating, the vehicle will be in the adaptive cruise state.

[0122] Execute the following process:

[0123] First, build a dynamic model. Based on the speed of the host vehicle and the deceleration of the leading vehicle, construct a spatio-temporal collision model within the next 3 seconds to predict the intersection area of the safety envelopes of the rear of the leading vehicle and the front of the host vehicle.

[0124] Then, by identifying the semantic information corresponding to the brake light signal of the leading vehicle, increase the risk weight of the spatio-temporal unit in front to determine the risk gradient distribution along the lane.

[0125] When the CNN of the in-vehicle system detects spatio-temporal overlap between the vehicle trajectory and the high-risk area, it will output a scraping probability value. If the probability exceeds the threshold, the in-vehicle system triggers a composite command of longitudinal torque enhancement (increasing braking pressure) and lateral fine-tuning (slightly steering to avoid), and then the vehicle smoothly transitions to a safe following distance.

[0126] Embodiment 6:

[0127] This application proposes a multi-level obstacle avoidance decision-making method for realizing guidance in human-machine co-driving. Refer to Figure 6, equipped with three levels of progressive risk thresholds - warning level, auxiliary level, and emergency level, each level corresponds to a differentiated obstacle avoidance mode. This application is different from the traditional fixed threshold. This solution introduces the steering wheel operation tendency parameter as the input variable for dynamic adjustment of the threshold. The input variable is calculated through multi-dimensional data fusion of steering wheel angular velocity, torque feedback, and grip status (such as capacitive sensing), which will quantify the intensity of the driver's active control intention when driving.

[0128] For example, if the driver continuously applies a steering torque greater than the basic value in a specific direction, the vehicle system determines that he has a clear intention to avoid obstacles, and will automatically increase the trigger thresholds of the auxiliary level and the emergency level, thereby reducing unnecessary vehicle system intervention.

[0129] The obstacle avoidance levels identified by this application include:

[0130] The warning level means that sound and light prompts and slight vibrations of the steering wheel are triggered, but vehicle control is not intervened, and the driver retains full operating rights.

[0131] The auxiliary level means starting steering torque compensation and torque pre-distribution to form a "human-led - machine-assisted" collaborative obstacle avoidance mode.

[0132] The emergency level means that the vehicle control is forcibly taken over and fully automatic obstacle avoidance actions are performed, while the driver is clearly prompted through the HMI interface.

[0133] During the mode switching process of this application, the vehicle system continuously monitors the dynamic changes of the steering wheel operation tendency parameters. If it detects that the driver actively reverses the operation (such as the driver forcibly turns in the emergency avoidance phase), the control weight arbitration mechanism will be immediately activated, and the operation instructions with a larger steering gradient will be executed first.

[0134] In order to avoid conflicts caused by switching of control rights, this application designs a progressive control transition algorithm. If the vehicle system is downgraded from the emergency level to the auxiliary level, the control weight is gradually released through a linear attenuation function, allowing the driver to take over smoothly. At the same time, a hysteresis filter is introduced at the moment of mode switching to eliminate false switching caused by sensor noise.

[0135] Introducing the scenario of a highway, if the vehicle enters the auxiliary obstacle avoidance mode due to an obstacle suddenly appearing in front of it, the vehicle system executes the following decision process:

[0136] The steering wheel sensor of the present application detects that the driver continues to apply a small left turn torque, indicating an intention to evade to the left, but the steering speed is lower than a preset threshold.

[0137] Based on the operating tendency parameters, the assistance level threshold will be automatically raised to the medium risk range, allowing the vehicle system to maintain collaborative avoidance mode at a higher risk level.

[0138] The in-vehicle system will enhance the left-turn assist and simultaneously apply progressive torque to the right rear wheel to balance the vehicle body posture. When the risk value breaks through the adjusted threshold due to the continuous approach of an obstacle, the in-vehicle system upgrades to the emergency level, briefly takes over the control to complete a rapid lane change, and finally returns the control to the driver through a progressive transition algorithm.

[0139] Embodiment 7:

[0140] This application proposes a dynamic decision-making method that couples risk and control. Refer to Figure 7 , based on the cross-judgment of the real-time collision probability and multi-level risk thresholds, the in-vehicle system dynamically switches between the three-level modes of warning - cooperation - emergency. Different from the traditional single-mode control, in the driving environment, this application will introduce a driver operation weight factor and dynamically allocate control weights according to the driver's steering input intensity in the cooperation mode.

[0141] For example, when the medium risk threshold is triggered in this application, if the driver continuously applies a clear steering torque, the in-vehicle system only provides torque compensation assistance instead of taking over completely; conversely, if it is detected that the driver has no operation tendency, the system control weight is increased to the dominant position. Yaw-braking correlation decision tree:

[0142] This application will establish a real-time correlation model between the probability collision coefficient and the yaw angular velocity:

[0143] In the positive correlation scenario (ρ > 0): when the yaw angular velocity and the collision risk increase synchronously (such as when the vehicle understeers and causes lateral offset), torque enhancement braking will be triggered. This application forms a virtual differential braking effect through the gradient distribution of motor torque (the torque of the inner wheel decreases and the outer wheel increases) to assist the vehicle in restoring the expected trajectory.

[0144] In the negative correlation scenario (ρ < 0): when the yaw angular velocity and the collision risk change in the opposite direction (such as when oversteering causes the vehicle body to rotate), a steering priority strategy is triggered. This application determines to delay the execution of the braking instruction and first stabilizes the vehicle body posture through steering angle correction to avoid the braking intervention from exacerbating the yaw instability.

[0145] Instruction generation and execution smoothing:

[0146] This application uses the model predictive control (MPC) framework to optimize the timing continuity of control instructions. When generating braking instructions, it will dynamically adjust the torque gradient based on the current vehicle speed and road surface adhesion coefficient to ensure that the change rate of the braking torque matches the vehicle dynamics envelope. The generation of steering instructions introduces a preview-following algorithm to achieve a smooth transition of path tracking through feedforward-feedback composite control.

[0147] In actual implementation, this application introduces the scenario of avoiding pedestrians on a wet slippery curve on a highway or in an urban road, and the in-vehicle system executes the following decision-making process:

[0148] If vision and radar are fused to detect a pedestrian suddenly breaking in, the scraping probability jumps to the high-risk range, triggering the emergency takeover mode.

[0149] Furthermore, through correlation analysis, the vehicle will experience understeer due to low road adhesion (the growth of yaw rate is positively correlated with the collision risk), and the system determines that ρ > 0.

[0150] By applying pulsed torque decay to the left front wheel and torque enhancement to the right rear wheel in the way of this application, a counterclockwise yaw moment is formed to correct the driving direction; meanwhile, based on the MPC framework, the steering angle correction amount is planned to enable the vehicle to avoid pedestrians along a continuous curvature path.

[0151] If the pedestrian leaves the risk area, the vehicle system gradually returns the vehicle control to the driver through torque linear release and steering angle return rate control.

[0152] Example 8:

[0153] This application proposes a decision-making method for dynamic trajectory prediction and avoidance space optimization. Refer to Figure 8 , and first, based on the spatio-temporal evolution characteristics of the obstacle motion acceleration, an extended Kalman filter (EKF) is designed.

[0154] This application is different from the traditional uniform motion model. In actual implementation, this application introduces an acceleration mutation detection module into the state equation, and identifies abnormal motion patterns of obstacles, such as emergency braking and sudden steering, through sliding window variance analysis. The prediction model of this application outputs the obstacle trajectory distributions at multiple future moments, forming a probability density cloud map.

[0155] For example, for a laterally moving bicycle, the model can predict the acceleration change trend and generate a fan-shaped diffusion trajectory cloud map to reflect the possible risk area of direction change.

[0156] This application will dynamically adjust the spatial resolution of the cloud map according to the product relationship between the sensor sampling frequency and the self-measured speed. When the vehicle is driving at high speed, the time-axis resolution is improved to ensure prediction timeliness;

[0157] When there is low speed, the spatial-axis resolution is enhanced to capture the fine motion trend. At the same time, the wavelet transform algorithm is used to perform multi-scale decomposition on the cloud map, and the calculation efficiency is optimized while ensuring the prediction accuracy during the implementation process.

[0158] For example, when the vehicle approaches an intersection, the system automatically increases the lateral spatial resolution to accurately identify the starting acceleration of pedestrians moving laterally. Confidence weight correction and avoidance decision:

[0159] This application performs spatio-temporal superposition analysis on the trajectory cloud map and the prediction results of the vector model by calculating the dynamic confidence weight. The dynamic confidence weight consists of three parts, such as historical prediction consistency, sensor data quality, and motion pattern matching degree;

[0160] The historical prediction consistency represents the average deviation between the predicted trajectories in the past 3 frames and the actual observed positions;

[0161] The sensor data quality represents the radar point cloud density and the visual recognition confidence score;

[0162] The motion pattern matching degree represents the matching index between the current acceleration and the common obstacle behavior library;

[0163] This application dynamically adjusts the update rate and avoidance space boundary of the vector model based on the weight value, and preferentially uses the high-confidence prediction area to generate obstacle avoidance instructions.

[0164] In the scenario of turning right on an urban road, the vehicle detects a pedestrian standing on the right sidewalk. The system executes the following decision-making process:

[0165] Based on the micro-motion characteristics of the pedestrian's standing posture (such as leaning forward), the model predicts that it may step into the lane after 2 seconds and generates a trajectory cloud map extending forward.

[0166] Since the autonomous vehicle is in a low-speed turning state (<15 km / h), the system improves the cloud map spatial resolution to the centimeter level to capture the tiny displacement trend of the pedestrian's toe direction.

[0167] The vision system identifies the angle between the pedestrian's gaze direction and the lane, combines historical data to judge its intention to cross the road, and raises the confidence weight of the corresponding area to a high level.

[0168] Generate a horseshoe-shaped avoidance path in the area where the pedestrian may intrude, trigger the steering system to fine-tune to the left in advance and reduce the vehicle speed to ensure a safe distance margin.

[0169] Embodiment 9:

[0170] This application proposes how to construct a multi-constraint coupling method to solve the problem of suppressing the oversteering risk and heat load under complex working conditions. Refer to Figure 9 , based on the risk level division (low, medium, high), define a three-dimensional control parameter priority matrix, and the dimensions include:

[0171] Risk level axis: reflecting the urgency of the current obstacle avoidance requirement

[0172] Lateral acceleration threshold axis: constraining the dynamic stability boundary of the vehicle during steering

[0173] Motor heat load axis: monitoring the temperature state of the drive motor to prevent overheating and derating

[0174] The matrix cell value is the control weight coefficient, representing the resource allocation ratio between steering and torque control. For example, when performing continuous obstacle avoidance in a high-temperature environment, the system automatically increases the weight coefficient of the motor's thermal load and limits the torque output gradient to protect the hardware.

[0175] Adopt the model predictive control (MPC) framework to solve the optimal solution set of the steering angle change rate and torque response rate within the rolling time domain.

[0176] Different from traditional PID control, this solution introduces a multi-objective cost function to simultaneously optimize the following indicators:

[0177] Path tracking accuracy: The lateral deviation between the actual vehicle trajectory and the planned path;

[0178] Energy efficiency: The comprehensive energy consumption of the motor torque output;

[0179] Thermal load accumulation: The balance between the motor temperature rise rate and the heat dissipation capacity;

[0180] Use the Lagrange multiplier method to dynamically balance conflicting objectives and generate decoupled control instructions that meet multiple constraint conditions.

[0181] According to the real-time risk level and hardware status, adjust the constraint boundaries of MPC:

[0182] Lateral acceleration threshold: In a low-adhesion road surface, shrink the allowable range of acceleration to suppress the risk of side slip;

[0183] Torque response rate: When the motor temperature approaches the critical value, reduce the torque change gradient to prevent the overheat protection from being triggered;

[0184] Steering angle change rate: In a narrow curve scenario, limit the steering rate to ensure the service life of the steering mechanism;

[0185] In the scenario of continuous obstacle avoidance on the highway, the vehicle needs to complete two lane changes within a short time to avoid multiple target obstacles ahead:

[0186] The system identifies that the current risk level is moderate, the motor temperature is in the normal range, assigns the highest weight (0.6) to path tracking accuracy, the second highest weight (0.3) to thermal load, and the lowest weight (0.1) to energy consumption efficiency.

[0187] During the first-stage lane change, the MPC algorithm generates a fast steering angle change instruction while satisfying the lateral acceleration constraint, and simultaneously increases the torque of the outer wheels to enhance the grip; during the second-stage lane change, as the motor temperature rises to the warning range, the system automatically reduces the torque response rate and preferentially completes path tracking through steering correction.

[0188] The steering actuator completes double-S trajectory tracking at a smooth angular rate, and the drive motor outputs stepped torque within the thermal load limit to ensure no power interruption or steering impact during the obstacle avoidance process.

[0189] Embodiment 10:

[0190] This application proposes an obstacle avoidance system for electric vehicles, specifically constructing an integrated perception-decision-execution hierarchical system. Refer to Figure 10 , the vector model building module generates a dynamic vector model of obstacles through spatio-temporal alignment and complementary fusion of heterogeneous sensors, such as lidar, millimeter-wave radar, and vision. The lidar provides a high-precision three-dimensional point cloud contour, the millimeter-wave radar captures the radial velocity, and the vision system identifies semantic attributes, such as vehicles and pedestrians. The module embeds a motion compensation algorithm, based on the yaw angular velocity and wheel speed signals of the IMU, to eliminate the coordinate offset caused by the vehicle's own motion. For example, when the vehicle brakes emergently, the module corrects the tilt of the sensor coordinate system in real time to ensure the spatio-temporal consistency of the obstacle position projection.

[0191] The collision determination module uses a non-linear projection algorithm to map the vector model to the vehicle's relative motion coordinate system. Through spatio-temporal grid-based discrete analysis, it calculates the spatio-temporal overlap risk value between the vehicle's envelope and the obstacle's contour. A fuzzy logic algorithm is introduced to comprehensively consider dynamic parameters such as the road surface friction coefficient and yaw moment margin, and outputs a normalized scraping probability. This module breaks through the limitations of traditional binary judgment and supports continuous quantitative evaluation of risk levels.

[0192] The obstacle avoidance control module of this application generates composite control instructions based on the risk level:

[0193] The steering angle response adopts a preview-follow algorithm to dynamically adjust the steering angle while ensuring the continuity of the path curvature, avoiding yaw instability caused by sharp turns.

[0194] The torque enhancement feedback dynamically distributes the drive wheel torque according to the real-time μ value, and assists the steering control through the virtual differential effect. For example, on a low-adhesion road surface, the torque of the wheel on the high-μ value side is increased to suppress the understeering tendency.

[0195] The dual-mode instructions are synchronously transmitted to the actuator through the CAN FD bus to ensure the timing consistency of steering and torque changes.

[0196] In actual implementation, for example, in an urban scenario, when the vehicle is driving along the lane and suddenly encounters a vehicle parked illegally on the right side that partially invades the lane:

[0197] The lidar detects the three-dimensional contour of the vehicle body that has invaded, the millimeter-wave radar captures its stationary state, the vision system identifies the door opening state, and the vector model building module fuses multi-source data to generate an obstacle vector model with door opening degree prediction.

[0198] The collision determination module projects the model in the vehicle's moving direction, calculates the spatio-temporal overlap risk value between the door unfolding trajectory and the envelope line of the vehicle's rearview mirror, and outputs the medium scraping probability.

[0199] The obstacle avoidance control module generates a steering command to turn slightly 3° to the left, and at the same time increases the torque of the right rear wheel to balance the vehicle body posture. Through the coordinated control of steering and torque, the system enables the vehicle to bypass the obstacle along a smooth trajectory, and passengers do not feel a sharp turn.

[0200] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. An obstacle avoidance control method for an electric vehicle, characterized in that, Including: Obtaining obstacle data and generating a vector model of the obstacle; Projecting the obstacle onto the relative motion direction of the current vehicle based on the vector model, and determining whether a collision will occur with respect to the operating parameters of the current vehicle; When the collision probability exceeds a preset risk value, generating an obstacle avoidance control instruction deviating from the obstacle projection; wherein, the obstacle avoidance control instruction includes an automatic response instruction for the steering angle of the current vehicle under the current path curvature and a self-feedback torque enhancement instruction based on the road surface friction.

2. The method for obstacle avoidance control of an electric vehicle according to claim 1, wherein, The vector model is constituted based on the spatio-temporal evolution of the dynamic characteristic parameters of the three-dimensional contour model of the obstacle generated after multi-sensor fusion in the current vehicle; wherein, the three-dimensional contour model is a spatio-temporal expansion model of the obstacle within the relevant contact distance including the current vehicle and the obstacle generated after dynamic compensation of the real-time motion state of the vehicle. When performing dynamic compensation of the motion state, within the relevant contact distance, the current vehicle performs vector marking based on the real-time motion state of the vehicle and the motion state of the obstacle, and uses the vector marking as the expansion range marking of the spatio-temporal expansion model.

3. The method for obstacle avoidance control of an electric vehicle according to claim 1, characterized in that, The obstacle projection includes configuring a dynamic safety corridor based on the current vehicle. Among them, the dynamic safety corridor is constructed based on the origin of the current vehicle as the lane coordinate system, and the dynamic safety corridor is configured with a collision limitation mechanism, and the collision limitation mechanism is based on the coupling coefficient between the projection area under the real-time motion state parameters of the vehicle and the vehicle envelope. The width of the dynamic safety corridor is dynamically adjusted according to the road surface adhesion coefficient and the vehicle yaw moment margin, and the coupling coefficient is used to weightedly calculate the collision probability.

4. The obstacle avoidance control method for an electric vehicle according to claim 1, characterized in that, The projecting the obstacle onto the relative motion direction of the current vehicle based on the vector model further includes: Configuring an offset determination mechanism based on virtual piloting, and the offset determination mechanism is used to generate a planning target point of the vehicle path in the reverse direction of the obstacle projection direction; The offset determination mechanism is constrained by the steering gradient of the current path curvature change rate, and limits the correction amplitude of the steering angle not to be greater than a preset amplitude value; The constraint condition is associated with the planning target point of the vehicle path and performs dynamic coupling with the vehicle heading angle deviation.

5. The method for obstacle avoidance control of an electric vehicle according to claim 1, wherein, The determining whether a collision will occur with respect to the operating parameters of the current vehicle includes: Obtaining the real-time motion parameters of the vehicle and constructing a dynamic collision model based on the obstacle projection; wherein, the collision model includes the probability collision area of the current vehicle relative to the safety profile in the motion direction; According to the dynamic collision model, determining the collision probability field based on the current vehicle trajectory; Through the probability collision field, determining the spatio-temporal overlap risk value of the current vehicle in the probability collision area, and outputting the real-time corrected collision probability.

6. The method for obstacle avoidance control of an electric vehicle according to claim 5, characterized in that, The preset risk value is divided based on multiple levels of risk thresholds, and each level of threshold in the multiple levels of risk thresholds corresponds to a different obstacle avoidance mode; wherein, the trigger boundary of the obstacle avoidance mode corresponding to each level of threshold in the multiple levels of risk thresholds is determined by the steering wheel operation tendency parameter of the current vehicle.

7. A method for obstacle avoidance control of an electric vehicle according to claim 1, characterized in that, The generating the obstacle avoidance control instruction deviating from the obstacle projection includes: And based on the real-time collision probability and the multiple levels of risk thresholds, determining the target obstacle avoidance mode and the obstacle avoidance control instruction corresponding to the target obstacle avoidance mode; The vehicle braking command is based on the correlation between the probability collision coefficient of the probability collision area and the vehicle yaw angular velocity. When they are positively correlated, the vehicle braking command is an acceleration braking command based on self-feedback torque enhancement; When they are negatively correlated, the vehicle braking command is a braking response delay command based on the steering angle automatic response command.

8. A method for obstacle avoidance control of an electric vehicle according to claim 1, characterized in that, The generation of the obstacle avoidance control command deviating from the obstacle projection further includes: Construct a first Kalman prediction model based on the obstacle motion acceleration, and generate a trajectory cloud map of the obstacle within the prediction time period; wherein, the resolution of the trajectory cloud map is adaptively adjusted according to the sensor sampling frequency and the vehicle speed, and the superposition analysis result is used to correct the confidence weight of the vector model; According to the trajectory cloud map and the current vehicle's motion braking area, determine the avoidance space and the response command of the obstacle avoidance control command corresponding to the avoidance space.

9. The obstacle avoidance control method for an electric vehicle according to claim 1, characterized in that, The generation of the obstacle avoidance control command deviating from the obstacle projection further includes: Establish a steering-torque coupling control domain, and define a control parameter priority matrix under different risk levels; Use the model predictive control algorithm to perform constraint decoupling on the steering angle change rate and the torque response rate; Among them, the dimension of the priority matrix includes the vehicle lateral acceleration threshold value and the drive motor thermal load state parameter.

10. An obstacle avoidance control system for an electric vehicle, characterized in that, It includes: Vector model building module: Obtain obstacle data and generate a vector model of the obstacle; Collision determination module: Project the obstacle onto the relative motion direction of the current vehicle based on the vector model, and determine whether a scrape will occur relative to the running parameters of the current vehicle; Obstacle avoidance control module: Used to generate an obstacle avoidance control command deviating from the obstacle projection when the scrape probability exceeds the preset risk value; wherein, the obstacle avoidance control command includes the steering angle automatic response command of the current vehicle under the current path curvature and the self-feedback torque enhancement command based on the road surface friction.

Images