WO2022251995A1 - Real-time vehicle stabilising system and method - Google Patents

Abstract

A real-time vehicle stabilising system and method. The system comprises: a sensor unit (10), which comprises an electromagnetic wave emitter (101) for emitting a point cloud to irradiate the road surface ahead, an electromagnetic wave receiver (102) for receiving point cloud position data and taking same as a ground status S1, and a vehicle status sensor (103) for obtaining a vehicle status S2; a control decision-making unit (30), which takes S1 and S2 as statuses S and inputs same into a pre-trained reinforcement learning model, so as to obtain an output vehicle control signal; and an execution unit (40), which is used for receiving a control signal from the control decision-making unit, so as to execute a vehicle stabilising intervention action A. A refined road surface status is obtained by collecting an emitted electromagnetic wave point cloud matrix, and on the basis of a reinforcement learning algorithm, corresponding execution feedback is made in real time for various road surface statuses, such that vehicle stability is maximized, thereby improving the comfort of driving a vehicle.

Description

Real-time vehicle stabilization system and method thereof technical field

This application relates to a system and method for real-time vehicle stabilization, in particular to a real-time vehicle stabilization system RVSS (Real time Vehicle Stabilizing System) and its methods.

Background technique

At present, the vehicle stabilization system VSS (Vehicle Stabilizing System) mainly includes the system ABS (anti-lock braking system), ASR (drive slip adjustment device) or ESP (electronic stability program), and is used to improve the vehicle in critical driving conditions (such as oversteer or The controllability and stability of the vehicle in the case of understeer, the main actuator is the brake system of the 4 wheels.

technical problem

The known vehicle stabilization systems are generally very rough for the relevant road foundations, and the selection modes of the foundations are mainly divided into asphalt, cement, mud, sand and the like. However, the actual road conditions are more complex and refined, with complex road conditions such as various potholes, congestion, and inclinations, and these uncertainties cause abnormal movement of the wheels and tilting and shaking of the vehicle body, which greatly reduces the driving safety of the vehicle. comfort.

technical solution

The main purpose of this application is to provide a real-time vehicle stabilization system RVSS (Real time Vehicle Stabilizing System) that can collect refined data on the road surface and control the execution unit in real time based on the reinforcement learning algorithm according to different road conditions on the road surface to achieve vehicle body stability. System) and its methods.

This application discloses a real-time vehicle stabilization system RVSS (Real time Vehicle Stabilizing System), the real-time vehicle stabilization system includes: sensor unit, including: electromagnetic wave transmitter emits point cloud to illuminate the road ahead, electromagnetic wave receiver receives point cloud position data as ground state S1, and vehicle state sensor is used to obtain vehicle state S2 ; Control decision-making unit: Input S1 and S2 as state S into the pre-trained reinforcement learning model to obtain the output vehicle control signal; Execution unit: used to receive the control signal of the control decision-making unit and execute vehicle stability intervention action A.

In order to pre-train the reinforcement learning model, the real-time vehicle stabilization system also includes: the sensor unit also includes: the driving state sensor is used to obtain one or two or more data of vehicle tilt, acceleration and steering, when the vehicle performs action A, The data change parameter of the driving state sensor is set as T, and a parameter R is set as the parameter T decreases as the parameter T increases, as a feedback reward.

In order to pre-train the reinforcement learning model, the real-time vehicle stabilization system also includes: model training unit: based on "environmental state S + action A + next state S' + feedback reward R after action A" as training data, keep trying and improving , so that the A action tends to maximize the feedback reward R, and train a reinforcement learning model.

Further, the electromagnetic wave transmitter and electromagnetic wave receiver include: light wave transmitter and optical camera, laser transmitter and lidar, microwave transmitter and microwave radar, ultrasonic transmitter and ultrasonic radar, wherein the electromagnetic wave transmitter and electromagnetic wave receiver Can be assembled into one unit.

Further, the point cloud emitted by the electromagnetic wave transmitter refers to a point matrix, an intersection matrix formed by lines, or an extracted point matrix in a light curtain.

Further, the ground state S1 parameter is an electromagnetic wave point cloud position matrix and/or an electromagnetic wave point cloud size matrix.

Further, the vehicle state sensor is one or two or more of an adjustable suspension system height sensor, an adjustable suspension system damping sensor, a vehicle speed sensor, an acceleration sensor, and a steering angle sensor, and is used to calculate the arrival of the wheels. The position and angle of the wheel on the point cloud, the height of the adjustable suspension system, and the damping parameters together form the vehicle state S2.

Further, the executive unit includes: an adjustable suspension system for adjusting the height of the wheel from the ground and/or damping parameters.

Further, the execution unit further includes: a vehicle speed controller, a braking device, and a steering device.

The present application also discloses a real-time vehicle stabilization method based on a reinforcement learning algorithm, which includes the following steps: a) the electromagnetic wave transmitter emits a point cloud to illuminate the road ahead, and the electromagnetic wave receiver receives the point cloud position data as the ground state S1; b) the vehicle state S2 is added to S1 to obtain the state S; c) Input the state S into the pre-trained reinforcement learning model to obtain the vehicle stability intervention action A; d) The driving state sensor parameter change T generated by the vehicle stability intervention action A, set a The parameter R decreases as the parameter T increases, as a feedback reward; e) The next state S' is obtained after the vehicle stability intervention action A; f) A reinforcement learning model is trained based on "state S + vehicle stability intervention action A + next state S'+feedback reward R" is used as training data, constantly trying and improving, so that the vehicle stability intervention action A tends to maximize the feedback reward R.

Beneficial effect

This application discloses a real-time vehicle stabilization system and its method, which solves the disadvantage that the existing vehicle stabilization system cannot perform regulation on the refined road surface state. This application obtains the refined state of the road surface by collecting the emitted electromagnetic wave point cloud matrix , based on the reinforcement learning algorithm, one-to-one corresponding execution feedback is made in real time for various road surface states (potholes, congestion, tilt, etc.) to maximize vehicle stability and greatly improve vehicle driving comfort.

Description of drawings

In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the accompanying drawings that need to be used in the description of the specific embodiments or prior art. Obviously, the accompanying drawings in the following description The drawings are some implementations of the present application, and those skilled in the art can obtain other drawings based on these drawings without creative work.

FIG. 1 and FIG. 2 are schematic diagrams of a real-time vehicle stabilization system provided by an embodiment of the present application.

FIG. 3 is a diagram of the training process of the reinforcement learning algorithm of the present application.

FIG. 4( a ) is a schematic diagram of the point cloud emitted by the electromagnetic wave emitter provided in the embodiment of the present application to irradiate the road surface.

Fig. 4(b) is a top view of Fig. 4(a).

Fig. 4(c) is a perspective view of Fig. 4(a).

FIG. 5 is a schematic diagram of point cloud signals received by the electromagnetic wave receiver provided in the embodiment of the present application.

FIG. 6 is a schematic diagram of a road surface state S1 digital matrix provided by an embodiment of the present application.

FIG. 7 is a schematic diagram of different positions and angles of the wheels reaching the road surface state S1 provided by the embodiment of the present application.

FIG. 8 is a schematic diagram of the wheel obtaining the maximum feedback reward R according to the road surface state S1 provided by the embodiment of the present application.

BEST MODE FOR CARRYING OUT THE INVENTION

The technical solutions of the present application will be clearly and completely described below in conjunction with the accompanying drawings. Apparently, the described embodiments are some of the embodiments of the present application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

In the description of this application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" etc. The indicated orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, as well as the position marks of each component in the drawings, which are only for the convenience of describing the application and simplifying the description, rather than indicating or implying the system or component referred to Must be in a particular orientation, constructed, and operate in a particular orientation, and thus should not be construed as limiting of the application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and should not be construed as indicating or implying relative importance.

In the description of this application, it should be noted that unless otherwise specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense, for example, it can be a fixed connection or a detachable connection. Connection, or integral connection; it can be mechanical connection or electrical connection; it can be direct connection or indirect connection through an intermediary, and it can be the internal communication of two components. Those of ordinary skill in the art can understand the specific meanings of the above terms in this application in specific situations.

It should be understood that the systems and methods of the present application can be used with any type of vehicle, including conventional vehicles, hybrid electric vehicles (HEV), extended-range electric vehicles (EREV), battery electric vehicles (BEV), motorcycles, battery cars, passenger cars, sports SUVs, crossovers, trucks, vans, buses, recreational vehicles (RV) and more. These are just some of the possible applications, as the systems and methods described herein are not limited to the exemplary embodiments shown in FIGS. 1-8 and can be implemented in many different ways.

In Fig. 1 and Fig. 2, an embodiment of a real-time vehicle stabilization system is shown according to a wheeled motor vehicle (50), equipped with: a sensor unit (10), including: an electromagnetic wave emitter ( 101), the electromagnetic wave receiver (102) used to receive point cloud position data as road surface state S1, the vehicle state sensor (103) used to obtain vehicle state S2, used to obtain one or both of vehicle tilt, acceleration and steering The driving state sensor (104) of one or more kinds of data; the control decision-making unit (30): input S1 and S2 as the state S into the pre-trained reinforcement learning model, and obtain the output vehicle control signal; the execution unit (40) Used to receive the control signal from the control decision-making unit, and execute the vehicle stability intervention action A, including: an adjustable suspension system (401) used to adjust the height and damping coefficient of the wheel (60), used to reduce or close the oil circuit to control the vehicle speed The vehicle speed controller (402), the brake device (403) used to connect the brake circuit to brake the vehicle, and the steering device (404) used to control the steering wheel to steer the vehicle; the model training unit (20): based on the obtained Training data to train a reinforcement learning model.

Electromagnetic Wave Transmitter (101) and Electromagnetic Wave Receiver (102): Electromagnetic waves include radio waves, microwaves, infrared rays, visible light, and ultraviolet rays. A device capable of emitting the electromagnetic waves is called an electromagnetic wave transmitter. The device that can receive the signal reflected from the target and compare it with the transmitted signal, and after proper processing, can obtain the relevant information of the target is called an electromagnetic wave receiver. Known electromagnetic wave transmitters and electromagnetic wave receivers, including but not limited to: light wave transmitters and optical cameras, laser transmitters and lidars, microwave transmitters and microwave radars, ultrasonic transmitters and ultrasonic radars, wherein electromagnetic wave transmitters and electromagnetic wave The receiver can be divided into two devices or assembled into one device and installed at the same position. The position in the figure is a schematic position, and the actual selection type and installation position can be changed flexibly. The point cloud signal emitted by the electromagnetic wave transmitter refers to the intersection point matrix composed of point matrix and line or the extraction point matrix in the light curtain. After proper processing, the point cloud position signal and point cloud size signal are obtained. The electromagnetic wave point cloud position matrix and/or the electromagnetic wave point cloud size matrix are used as the road surface state S1 parameter.

The vehicle state sensor (103) includes: one or two or more of an adjustable suspension system height sensor, an adjustable suspension system damping sensor, a vehicle speed sensor, an acceleration sensor, and a steering angle sensor. According to the collected road surface state S1, the height and damping of the wheels can be adjusted to stabilize the vehicle body, but there is still a time difference between the road surface state S1 reaching the wheels, so it is necessary to obtain the parameters of the vehicle state sensor to calculate the arrival of the wheels on the road surface The corresponding position, direction, and adjustable suspension system height and damping parameters in the state S1, and the parameters of these vehicle state sensors constitute the vehicle state S2.

The driving status sensor (104) includes: angular velocity sensor (ie gyroscope), acceleration sensor (ie accelerometer), magnetic induction sensor (ie electronic compass), these three types of sensors can be freely combined to obtain multi-axis sensors, and the three types of sensors are designed into one The sensor is referred to as the nine-axis sensor. Used to obtain one or two or more data of vehicle tilt, acceleration and steering. However, the driving state sensors include but are not limited to the above sensors, for example, the steering angle sensor installed under the steering wheel can also obtain vehicle steering data. In addition, some detection parameters of the driving state sensor and the vehicle state sensor are the same, and one sensor can be used to share the obtained parameters. The parameters obtained by the driving state sensor can detect the running stability of the vehicle, and the smaller the detected changes in the parameters of inclination, acceleration and steering, the more stable the vehicle is running. Here, the driving state sensor data change parameter is set as T, and a parameter R is set to decrease as the parameter T increases as a feedback reward. That is, the larger R is, the more stable the vehicle is.

The control decision-making unit (30) is equipped with a microcomputer, an interface of a wiring harness, and the like. The above-mentioned microcomputer has a CPU (Central Processing Unit: Central Processing Unit), ROM (Read Only Memory: Read Only Memory), RAM (Random Access Memory: Random Access Memory), I/O and CAN (Controller Area Network: A well-known structure of a controller area network) communication device or the like. The control unit is mainly used to input S1 and S2 as the state S into the pre-trained reinforcement learning model, obtain the output vehicle control signal, and let the execution unit execute the vehicle stability intervention action A. Maximize the feedback reward R, and finally achieve the purpose of vehicle stability.

Adjustable suspension system (401): The adjustable suspension system adjusts the height and damping of the wheel suspension according to the instructions of the control decision-making unit, so that the vehicle is in the best stable driving state. The current adjustable suspension system of automobiles can be divided into the following three categories according to the control type: air adjustable suspension system, hydraulic adjustable suspension system, and electromagnetic adjustable suspension system.

Vehicle speed regulator (402): used to reduce or close the oil circuit to control the vehicle speed.

Brake device (403): used to connect the brake circuit to brake the vehicle.

Steering device (404): used to control the steering wheel to steer the vehicle.

Model training unit (20): Based on "environmental state S + vehicle stability intervention action A + vehicle stability intervention action A and the next state S' + feedback reward R" as training data, keep trying and improving, so that A action tends to feedback The reward R is the largest, and a reinforcement learning model is trained. The model training unit can adopt the following three modes: 1. Online mode: the model training unit is installed on the vehicle, obtains data in real time for model training, and inputs the state S into the trained model to obtain vehicle stability intervention action A; 2. Offline Mode: That is, the model training unit is not installed on the vehicle. By collecting the data on the vehicle, the background reinforcement learning model training is performed, and then the trained reinforcement learning model is imported into the vehicle that needs to be installed. Finally, when the vehicle is in use, the state S is input The trained model gets the vehicle stabilization intervention action A; 3. Offline + online mode: the model is first trained in the offline mode and imported to the vehicle that needs to be installed, but the model is updated and optimized when the vehicle is in use, and the vehicle is output Stabilization intervention action A. These three methods have their own advantages and disadvantages: the advantage of method 1 is that the structure is simple, but the model accuracy is not high; the advantage of method 2 is that it can collect more vehicle data for fusion, making the model more accurate; the advantage of method 3 is that On the basis of method 2, in addition to the higher accuracy of the model, the unique characteristics of the vehicle can be added to the model in the subsequent use of the vehicle, so that the vehicle stability intervention action A is more suitable for the condition of the vehicle.

FIG. 3 is a diagram of the training process of the reinforcement learning algorithm of the present application. Reinforcement learning as a sequential decision (Sequential Decision Making) problem, which needs to continuously select some behaviors, and get the maximum benefit from the completion of these behaviors as the best result. Without any label telling the algorithm what to do, it first tries to make some behaviors - then gets a result, and gives feedback on the previous behavior by judging whether the result is right or wrong. The previous behavior is adjusted by this feedback, and the algorithm can learn what behavior to choose under what circumstances to get the best results through continuous adjustment of the algorithm.

Plain language explanation: We have trained an artificial brain Agent, which can make judgments on the status of the environment, read the status of the environment, and make actions. After the artificial brain makes an action, the environment will give the Agent a reward feedback Reward based on the action received from the Agent, and the artificial brain will make improvements based on the reward feedback from the environment, so as to make better Improve actions. That's it An iterative process, the Agent keeps trying and improving itself. So how to make Agent become far-sighted enough to optimize the current fixed action from a long-term perspective, rather than quick success. So Agent Every step needs to be aligned towards the side that obtains the greatest benefit.

Figures 4-8 are schematic diagrams of the real-time vehicle stabilization provided by the embodiment of the present application. Referring now to Figures 4-8, a specific method for realizing real-time vehicle stabilization by using a reinforcement learning algorithm will be described.

Step a) The electromagnetic wave transmitter emits a point cloud to irradiate the road ahead, and the electromagnetic wave receiver receives the point cloud position data as the road surface state S1. As shown in Figure 4 (a) (b) (c): the electromagnetic wave transmitter (101) on the vehicle (50) emits a quadrilateral point cloud matrix of point cloud ABCD to the road surface, and there are potholes (70) and congestion on the road surface (80).

FIG. 5 is a schematic diagram of a signal point cloud received by an electromagnetic wave receiver provided in an embodiment of the present application. Among them, at the positions of potholes (70) and pockets (80), due to changes in the state of the road surface, the point positions of the point cloud at the positions will be offset by a certain amount, and the road state S1 digital matrix shown in Figure 6 is obtained. This time, the road state S1 digital matrix in Figure 6 has been simplified for illustration. The actual road state S1 digital matrix is denser, with a larger amount of data, including more information, such as: in addition to potholes and congestion, there are also road slopes, etc. More information, in addition to the electromagnetic wave point cloud position matrix and the electromagnetic wave point cloud size matrix, the point cloud size refers to the diameter of the point, and this parameter can reflect the material properties of the road surface. Of course, although the road surface state S1 digital matrix is just a digital matrix, the actual information it contains must be richer than the known information, which can only be interpreted through the reinforcement learning model, which is also the strength of the reinforcement learning model.

Step b) Add vehicle state S2 to S1 to obtain state S. According to the collected road surface state S1, the height and damping of the wheels can be adjusted to stabilize the vehicle body. However, there is still a time difference t when the road surface state S1 reaches the wheels, so it is necessary to obtain the parameters of the vehicle state sensor to calculate the arrival of the wheels. The corresponding position, direction, and adjustable suspension system height and damping parameters in the road surface state S1, and the parameters of these vehicle state sensors constitute the vehicle state S2. Among them, the state S obtained by adding the vehicle state S2 to the road surface state S1 can be simply fused with the digital matrix S1 plus the digital matrix S2, and a new digital matrix S can be generated by adding the time parameter t, or the speed and acceleration in S2 , steering angle, time t and other parameters are calculated when the wheels arrive at the S1 point cloud matrix, the parameters such as the position and direction of the wheels simplify the road surface state S1, and then add the adjustable suspension system height and damping parameters to obtain the state S digital matrix.

Figure 7 is a schematic diagram of the different positions and angles of the wheels reaching the road state S1 provided by the embodiment of the present application. As shown in Figure 7(a) the wheel position, the positions passed by the wheels are the area (901) and the area (902) in the figure, and the actual road state The S1 digital matrix can be simplified as the area(901) plus area(902) matrix. As shown in Figure 7(b) the wheel position, the wheel position has offset and steering, then the area that affects the wheel becomes area (903) and area (904), and the actual road surface state S1 digital matrix can be simplified as area (903 ) plus the area (904) matrix.

Step c) Input the state S into the pre-trained reinforcement learning model to obtain the vehicle stabilization intervention action A. Through the reinforcement learning model, the digital matrix of the state S is input, and the digital matrix of the output vehicle stability intervention action A is obtained, wherein the parameters in the digital matrix of A include: wheel height adjustment parameters and wheel damping adjustment parameters. That is, whether the wheels should rise or fall when encountering various road conditions, and whether the damping should be adjusted to be softer or harder to adapt to the road surface, making the vehicle more stable and improving driving comfort. Furthermore, in addition to the adjustment of the vehicle's adjustable suspension system, other control means in the execution unit can also be used, such as: vehicle speed controller, braking device and steering device. When using the vehicle speed controller, braking device and steering device for regulation, the existing driving conditions should be fully considered, especially the driving safety and driving comfort of the vehicle.

Step d) The parameter change T of the driving state sensor produced by the vehicle stability intervention action A, set a parameter R that decreases with the increase of the parameter T, as a feedback reward; Sensors are used to judge vehicle stability, that is, the smaller and/or smoother the parameter changes in vehicle tilt, acceleration and steering, the better the vehicle stability. Of course, different weight parameters can be added in front of the three parameters to define the different importance of vehicle tilt, acceleration and steering. The specific weight parameters can be defined according to the driving experience of the actual experimental situation, or made into different options for the driver Passengers are free to choose.

As shown in Figure 8, the schematic diagram of the wheel getting the maximum feedback reward R according to the road surface state S1. As shown in Figure 8(a), the road state S1 is crowded. The wheel at the position (sa) can only go up, so the parameter change T of the driving state sensor can be minimized; the wheel at the position (sb) can only go down, and the parameter T is the smallest. . As shown in Figure 8(b), there are potholes in the road surface state S1, the wheel at the position (sc) can only go down, and the parameter T is the smallest; the wheel at the position (sd) can only shrink up, and the parameter T is the smallest; the vehicle here The stability intervention action A is to achieve the minimum T and the maximum R through the change of the adjustable suspension system parameters.

Step e) The next state S' is obtained after the vehicle stabilization intervention action A.

Step f) Train a reinforcement learning model, based on "state S + vehicle stability intervention action A + next state S' + feedback reward R" as training data, keep trying and improving, so that vehicle stability intervention action A tends to feedback reward R maximum.

Further, for feedback reward R to perform well in the long run, we need to consider not only immediate rewards, but also future rewards we will receive. Therefore, set Rt=rt+γRt+1, rt is the immediate reward after executing t steps, Rt+1 is the future reward after executing the next t+1 steps, and γ is the discount factor with a value between 0 and 1 , the farther away we are from future rewards, the less we consider them.

Further, the reinforcement learning model can be trained using the Q-learning method, and the Q-learning update formula is as follows: Q(s,a)←Q(s,a)+α[r+γMAXa'Q(s', a')−Q(s,a)], according to the next state s', select the largest Q(s',a') value multiplied by the decay coefficient γ plus the real return value as the Q reality, and according to the past Q table The Q(s,a) inside is used as a Q estimate to update the Q-table, where α is the learning rate.

Furthermore, in ordinary Q-learning, when the state S and action A are discrete and the dimension is not high, Q-Table can be used to store the Q value corresponding to each state S and action A, and when the state S and action A are When high-dimensional continuous, use Q-Table to store state S and action A, because the amount of data is too large to store is very difficult. Therefore, the reinforcement learning model can be trained using the DQN (CNN+Q-Learning) method. First, the convolutional neural network CNN is introduced, and the Q-table update is transformed into a function fitting problem, and the Q-table is generated by fitting a function function. The Q value makes similar states get similar output actions.

Industrial Applicability

This application discloses a real-time vehicle stabilization system and its method, which solves the disadvantage that the existing vehicle stabilization system cannot perform regulation on the refined road surface state. This application obtains the refined state of the road surface by collecting the emitted electromagnetic wave point cloud matrix , based on the reinforcement learning algorithm, one-to-one corresponding execution feedback is made in real time for various road conditions (potholes, congestion, tilt, etc.) to maximize vehicle stability and greatly improve vehicle driving comfort. High market application value.

Although the present disclosure has been described based on the examples, it should not be understood that the present disclosure is not limited to the examples and structures. The present disclosure also includes various modified examples and modifications within the equivalent range. In addition, various combinations and forms, and other combinations or forms including only one element, more than one element, or less than one element are included in the scope and scope of thought of the present disclosure.

Claims (10)

A system for real-time vehicle stabilization comprising: The sensor unit (10) includes: the electromagnetic wave transmitter (101) emits a point cloud to illuminate the road ahead, the electromagnetic wave receiver (102) receives the point cloud position data as the ground state S1, and the vehicle state sensor (103) is used to obtain the vehicle state S2; Control decision-making unit (30): Input S1 and S2 as state S into the pre-trained reinforcement learning model to obtain output vehicle control signals; Execution unit (40): used to receive the control signal from the control decision-making unit, and execute vehicle stability intervention action A. The system according to claim 1, characterized in that: the sensor unit further includes: a driving state sensor (104) for obtaining one or two or more data of vehicle tilt, acceleration and steering, when the vehicle performs action A , the driving state sensor data change parameter is set to T, and a parameter R is set to decrease as the parameter T increases as a feedback reward. The system according to claim 2, characterized in that it also includes a model training unit (20): based on "environmental state S + vehicle stability intervention action A + vehicle stability intervention action A and the next state S' + feedback reward R" as training The data, constantly trying and improving, make the A action tend to maximize the feedback reward R, and train a reinforcement learning model. The system according to claim 1, wherein the electromagnetic wave transmitter and the electromagnetic wave receiver comprise: a light wave transmitter and an optical camera, a laser transmitter and a laser radar, a microwave transmitter and a microwave radar, an ultrasonic transmitter and an ultrasonic radar, Wherein the electromagnetic wave transmitter and the electromagnetic wave receiver can be assembled into a device. The system according to claim 1, characterized in that: the point cloud emitted by the electromagnetic wave emitter is a point matrix, an intersection matrix composed of lines, or an extracted point matrix in a light curtain. The system according to claim 5, wherein the ground state S1 parameter is an electromagnetic wave point cloud position matrix and/or an electromagnetic wave point cloud size matrix. The system according to claim 1, wherein the vehicle state sensor is one or two or more of an adjustable suspension system height sensor, an adjustable suspension system damping sensor, a vehicle speed sensor, an acceleration sensor, and a steering angle sensor , used to calculate the position and angle of the wheel on the point cloud when the wheel reaches the corresponding point cloud, the height of the adjustable suspension system, and the damping parameters, which together form the vehicle state S2. The system according to any one of claims 1 to 7, characterized in that: the execution unit (40) includes: an adjustable suspension system (401) for adjusting the height of the wheel from the ground and/or damping parameters. The system according to claim 8, characterized in that: the execution unit further comprises: a vehicle speed regulator (402), a braking device (403), and a steering device (404). A method for real-time vehicle stabilization, said real-time vehicle stabilization method comprising the steps of: a) The electromagnetic wave transmitter emits a point cloud to illuminate the road ahead, and the electromagnetic wave receiver receives the point cloud position data as the ground state S1; b) Add the vehicle state S2 to S1 to get the state S; c) Input the state S into the pre-trained reinforcement learning model to obtain the vehicle stability intervention action A; d) The parameter change T of the driving state sensor generated by the vehicle stability intervention action A, and a parameter R is set to decrease with the increase of the parameter T as a feedback reward; e) Get the next state S' after vehicle stabilization intervention action A; f) Train a reinforcement learning model, based on "state S + vehicle stability intervention action A + next state S' + feedback reward R" as the training data, keep trying and improving, so that the vehicle stability intervention action A tends to maximize the feedback reward R .