Automatic driving vehicle remote takeover method based on parallel control

Abstract

The invention belongs to the technical field of automatic driving, and particularly relates to an automatic driving vehicle remote takeover method based on parallel control, which is implemented in parallel at an automatic driving vehicle end and a cloud end by utilizing a parallel control method so as to realize quick and stable switching between a cloud end control right and a vehicle end control right. Obtaining the current state information of the automatic driving vehicle and the traffic information of the current environment of the automatic driving vehicle, and calculating a driving risk coefficient; based on a parallel control concept and a virtual simulation model, a cloud reinforcement learning agent based on a decision module and an evaluation module is constructed. The safety state of the automatic driving vehicle can be actively evaluated, active takeover is carried out when the automatic driving system breaks down or fails, the vehicle is actively takeover by the cloud controller or the cloud safety officer, the problem of halfway parking and waiting caused by a traditional takeover mode does not exist, the continuity of a driving task can be ensured, the comfort of passengers is improved, and the driving efficiency is improved. The traffic efficiency is improved; and the safety is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and mainly relates to a remote takeover method for an autonomous driving vehicle based on parallel control. Background Art

[0002] Autonomous driving, also known as driverless cars or wheeled robots, is an integrated and intelligent product that integrates environmental perception, decision-making and planning, control and execution. Autonomous driving has a history of several decades since its inception. With the advent of artificial intelligence and the 5G era, autonomous driving is gradually becoming more practical. Autonomous driving can replace most drivers' work, improve road traffic efficiency, reduce traffic accidents, and improve travel safety. At present, autonomous driving vehicles can make decisions independently in most scenarios, but there are still special situations that require remote safety officers to intervene. Since the system cannot cover all driving scenarios, when the vehicle exceeds its designed operating domain, the safety officer needs to take over remotely to help the vehicle get out of trouble. When existing unmanned vehicles encounter some system failures or special situations that require remote control switching, they generally exit the autonomous driving mode first, initiate a takeover request to the remote driving platform, and then help the vehicle get out of trouble through manual intervention, and then the remote driver exits. The handover of control during the switching process directly affects the safety of the vehicle. Whether the driver can output smoothly after the switch is related to the smoothness of the autonomous driving vehicle. Summary of the invention

[0003] The purpose of the present invention is to provide a method for remote takeover of an autonomous driving vehicle based on parallel control, so as to solve the problem that during the remote driving mode switching process of the autonomous driving vehicle, the vehicle control right switching is not smooth, affecting driving safety.

[0004] To solve the above problems, the present invention proposes a method for remote takeover of an autonomous driving vehicle based on parallel control. The method uses a parallel control method to be executed in parallel on the autonomous driving vehicle side and the cloud side, thereby realizing a fast and smooth switch between the cloud side control and the vehicle side control, including:

[0005] Build a virtual simulation model in the cloud to obtain the current state information of the autonomous driving vehicle and the traffic information of the environment in which the autonomous driving vehicle is currently located, and calculate the driving risk coefficient; Based on the parallel control concept and virtual simulation model, build a cloud-based reinforcement learning agent based on the decision module and the evaluation module. Through the parallel control method, parallel execution is performed between the vehicle-side control and the cloud-based virtual simulation. When the vehicle is driving normally, the vehicle-side decision is mainly made, and the cloud-based control is executed in parallel;

[0006] The virtual simulation model matches the current vehicle status information and the traffic information of the environment, evaluates the current safety status of the autonomous vehicle, calculates the total field strength of the vehicle's driving risk field, and calculates the vehicle's driving risk coefficient based on the forces acting on the vehicle in the driving risk field;

[0007] When the vehicle is in a dangerous state and beyond the decision-making ability of the vehicle, the cloud controller will take over safely and actively intervene. By parallel optimization and minimizing function differences, the cloud decision is close to the vehicle decision, realizing a fast and smooth switch of control between the cloud and the vehicle. After the control is smoothly transferred, the cloud reduces driving risks through active intervention, and then returns control to the vehicle. When the degree of danger is too high, the cloud safety officer will take over to ensure driving safety.

[0008] If the driving risk coefficient β risk <W 1 , the decision-making control on the vehicle side is the main one, and the cloud control is executed in parallel. If the driving risk coefficient W 1 ≤β risk <W 2 , the cloud controller automatically takes over and actively intervenes through the parallel control method. If the driving risk coefficient β risk ≥W 2 , the cloud security officer takes over control.

[0009] The vehicle driving risk field is composed of the "potential field energy" determined by the static elements of the road, the "dynamic field energy" determined by the moving objects on the road, and the "behavior field" of the vehicle driving behavior. In order to quantify the relationship between field force and safety, the driving risk coefficient is introduced. The driving risk coefficient indicates the safety level of the car: the lower the driving risk coefficient of the vehicle, the safer the car is, and the vehicle is autonomously controlled by the vehicle side; when the vehicle driving risk coefficient is medium, the car has a safety risk, which is beyond the decision-making range of the vehicle side and is taken over by the cloud controller; when the driving risk coefficient of the car is too high, the cloud safety officer will take over remotely to ensure driving safety.

[0010] Furthermore, the following method is used to calculate the driving risk coefficient and allocate the control rights of the autonomous driving car:

[0011] a) Driving risk coefficient calculation method: According to the effects of various elements of people, vehicles and roads in the surrounding environment on driving risk, a vehicle driving risk field model is constructed by establishing mathematical models of the kinetic energy field, potential energy field and behavior field formed by various elements in the driving environment, and the total field strength of the driving risk field of the vehicle is calculated;

[0012]

[0013] In the formula, is the total field strength vector of the driving risk field at the position of vehicle j, is the field strength vector at the position of vehicle j in the kinetic energy field, is the total vector strength of the potential energy field at the position of vehicle j, is the field strength vector of the behavior field at the position of vehicle j; and are the field strength vectors of a single kinetic energy field, potential energy field, and behavior field at the position of vehicle j respectively.

[0014] b) Calculate the driving risk coefficient of the vehicle based on the forces acting on the vehicle in the driving risk field;

[0015]

[0016] Where: F j is the force exerted on vehicle j in driving risk; is the total field strength of the driving risk field at point j; M j is the equivalent mass; R j is the road impact factor; v j is the speed of the vehicle; θ j is the angle between the vehicle and the speed direction; D j It is an undetermined risk factor, which is determined by the behavioral characteristics of different driving modes;

[0017]

[0018] Where: β risk is the driving risk coefficient of the vehicle, F j_max is the maximum force that vehicle j is subjected to in driving risk, F j_min The minimum value of the force applied to vehicle j in driving risk;

[0019] c) Determine the range of control rights for autonomous driving according to the driving risk coefficient of the autonomous driving vehicle, where the driving risk coefficient is divided into three ranges: safe, dangerous and very dangerous, and each range corresponds to a control right;

[0020] 1) Safety (vehicle status)

[0021] If β risk <W 1 , it means that the vehicle is in a safe state. At this time, the vehicle control is controlled by the autonomous driving vehicle. 1 It is the threshold at which the vehicle is in a dangerous state, determined by the operational design domain and the intended functional safety.

[0022] 2) Danger

[0023] If W 1 ≤β risk <W 2, it means that the vehicle is in a dangerous state. At this time, the vehicle control is controlled by the cloud controller, where W 2 It is the threshold of a very dangerous state, determined by the critical operational design domain and the intended functional safety.

[0024] 3) Very dangerous

[0025] If β risk ≥W 2 , it means that the vehicle is in a very dangerous state, and the control of the vehicle is controlled by the cloud safety officer.

[0026] 3. According to the method for remote takeover of an autonomous driving vehicle based on parallel control as described in claim 1, it is characterized in that: the virtual simulation model includes: an autonomous driving vehicle model, a traffic participant model, a road traffic facility model and an environmental condition model; the autonomous driving vehicle model is a model constructed based on the characteristics of the autonomous driving vehicle; the traffic participant model is a model constructed based on the characteristics of other vehicles and pedestrians other than the autonomous driving vehicle; the road traffic facility model is a model constructed based on roads and traffic facilities, and the environmental condition model is a model constructed based on environmental information.

[0027] Furthermore, the vehicle status information includes vehicle speed information, vehicle position information, steering wheel angle information, brake and accelerator pedal opening information. Its beneficial effects are: combining various vehicle status information to ensure the authenticity of the virtual simulation model, the cloud controller can take over quickly, and improve the smoothness of control switching.

[0028] Furthermore, the cloud-based reinforcement learning agent based on the decision-making module and the evaluation module is constructed, including: using the motion state feedback of the autonomous driving vehicle during driving as the input of the reinforcement learning model, the cloud-based decision-making module of the reinforcement learning agent outputs the decision result x according to the current state, and the cloud-based evaluation module outputs the decision result x according to the state-action value function U 1 (x 1 ,u 1 ) evaluates the decision result and calculates the new value function U 2 (x 2 ,u 2 ) as the vehicle-side decision input. The cloud-side decision module selects the optimal decision instruction Y according to the maximum judgment function value, and minimizes the value function difference between the cloud-side evaluation module and the vehicle-side judgment module, so that the cloud-side decision is close to the vehicle-side decision output, realizing a fast and smooth switch of control rights between the cloud and the vehicle.

[0029]

[0030] in, The value function of the cloud-side evaluation module and the vehicle-side evaluation module is the cumulative discount reward.

[0031]

[0032] When H(t)<δ, δ can be optimized and calibrated according to the needs of smooth control. The evaluation module outputs a trigger command to the vehicle-side decision module to complete the smooth handover of control from the vehicle-side to the cloud.

[0033] Furthermore, after the control right is successfully handed over, the driving risk coefficient is reduced by selecting the optimal vehicle speed and wheel turning angle with the goal of minimizing the driving risk coefficient, and then the control right is returned to the vehicle end; the specific method is as follows:

[0034] Establish the objective function of the car's decision-making problem based on the driving risk coefficient:

[0035]

[0036] Where: β k (x k ,dec k ) indicates that the vehicle is at x in the kth stage k Execute decision dec in state k The driving risk coefficient after risk (t) is the optimal indicator function. (x j ,y j ) is the coordinate of the vehicle position in stage j, v j is the driving speed, δ i is the wheel angle, is the road adhesion coefficient, F X and F Y They represent the longitudinal force and lateral force on the vehicle tires respectively; f is the rolling resistance coefficient; C D is the air resistance coefficient; A is the frontal area of ​​the vehicle; M = m / L, W = C D A / 21.15,

[0037] When the vehicle is traveling at normal speed, the time t p After that, the position that the vehicle will reach can be calculated according to the formula.

[0038]

[0039] Let R(t) represent the turning radius of the vehicle, then:

[0040]

[0041] Where: K is the stability factor; L is the wheelbase of the vehicle; δ is the steering angle.

[0042] Safety-oriented cloud-based active intervention is the optimal solution process for the driving safety risk objective function, and ultimately the optimal path and speed decision sequence are obtained by reducing the driving risk coefficient.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention constructs a virtual simulation model in the cloud, and then constructs a cloud-based reinforcement learning agent based on a decision module and an evaluation module. The parallel control method is used to execute in parallel between the vehicle-side control and the cloud-side virtual simulation, and the driving risk coefficient of the driving safety of the autonomous driving vehicle is calculated. When dangerous driving occurs, the vehicle-side control right is allocated: in general scenarios, the autonomous driving system makes its own decisions on the vehicle side when safety conditions are met; when danger occurs, the vehicle side switches the control right to the cloud-side controller for safe takeover and active intervention; when a very dangerous state occurs, the control right is taken over by the cloud-side safety officer. Through the parallel control method, the cloud-side virtual simulation and the vehicle-side driving are executed in parallel, and the cloud can obtain the real-time output of the vehicle side. In dangerous scenarios, the cloud-side controller can take over smoothly and quickly to ensure driving safety and driving smoothness; the present invention can actively evaluate the safety status of the autonomous driving vehicle, and take over actively when the autonomous driving system fails or fails. The cloud-side controller or the cloud-side safety officer takes over the vehicle actively, and there is no problem of stopping and waiting in the middle caused by the traditional takeover method, which can ensure the continuity of the driving task, increase passenger comfort, improve traffic efficiency and improve safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Flowchart of the autonomous driving remote takeover method of the present invention.

[0046] Figure 2 Schematic diagram of the control right switching interval of the present invention.

[0047] Figure 3 Parallel programming control structure diagram of the present invention. DETAILED DESCRIPTION

[0048] The present invention is described with reference to the accompanying drawings.

[0049] like Figure 1-Figure 3As shown, a method for remote takeover of an autonomous driving vehicle based on parallel control is implemented by parallel control method in the autonomous driving vehicle and cloud, so as to realize fast and smooth switching of cloud control right and vehicle control right; the autonomous driving vehicle is used to obtain vehicle status information and traffic information of the autonomous driving vehicle, and control the autonomous driving vehicle according to the control instructions of the autonomous driving system, cloud virtual system or remote driving platform of the autonomous driving vehicle; the autonomous driving vehicle is provided with sensors, radars, high-precision maps, cameras and other devices for obtaining vehicle status information of the autonomous driving vehicle. The vehicle status information here includes vehicle speed information, vehicle position information, steering wheel angle information, brake and accelerator pedal opening information; the autonomous driving vehicle has a communication module for communicating with both the cloud virtual system and the remote driving platform, and can send the vehicle status information and traffic information of the autonomous driving vehicle to the cloud virtual system and the remote driving platform in real time through the network environment, and can receive control instructions from the cloud virtual system and the remote driving platform.

[0050] The cloud virtual system is based on the virtual car model of the autonomous driving vehicle and its traffic environment constructed in the cloud, and the vehicle state and traffic environment in the model are synchronized with the actual physical world in real time. The cloud virtual system constructs a virtual scene for its main elements according to the design operating conditions and design operating domain of the autonomous driving vehicle on urban roads, and provides a virtual scene. The cloud virtual car model can obtain the current vehicle state information and traffic information, and correspond one-to-one with the autonomous driving vehicle state information and traffic environment information at the same time, so as to realize the parallel execution in the virtual simulation model. Historical data can improve the accuracy of the cloud virtual simulation platform and improve the accuracy of the safety assessment and prediction of the entire life cycle of the autonomous driving vehicle. The current vehicle state information, the current traffic information and the surrounding environment are used to calculate the driving risk coefficient to evaluate the current safety state of the autonomous driving vehicle, and then the control right of the autonomous driving vehicle is allocated, and the control right is handed over to the corresponding control subject according to the range of the driving risk coefficient. When the control authority of the autonomous driving vehicle is transferred to the cloud virtual simulation platform, a control instruction can be sent to the autonomous driving vehicle so that the autonomous driving vehicle works according to the control instruction of the cloud controller.

[0051] The remote driving platform is provided with a display device. After receiving the autonomous driving vehicle and traffic information, the remote driving platform can display the vehicle status information and traffic information through the display device. The display form can be text, image, sound, and the display method can be a display or a virtual reality device. When the driving risk factor is too high and the cloud controller cannot meet the driving requirements of the autonomous driving vehicle, the cloud safety officer needs to take over. The remote driving platform reminds the cloud safety officer to prepare for the takeover. The reminder method can be sound, vibration, light, etc. The remote driving platform will convert the driving operation of the cloud safety officer into control instructions and send them to the autonomous driving vehicle (including steering wheel angle, expected speed, expected acceleration, lighting control, pulling up / releasing the parking brake, etc.), so that the autonomous driving vehicle ignores the control instructions from the cloud virtual platform and the remote driving platform and executes the control instructions from the remote driving platform.

[0052] The present invention comprises the following steps when used:

[0053] Step 1: Use various sensors on the autonomous driving vehicle to obtain the current status information of the autonomous driving vehicle and the traffic information of the environment in which the autonomous driving vehicle is currently located, and send it to the cloud platform in real time through the 5G network.

[0054] Vehicle status information includes vehicle location information, destination information, vehicle speed information, steering wheel angle information, accelerator and brake pedal opening information. Traffic information includes road topology, weather conditions, number of obstacles, obstacle types, obstacle locations and speeds, traffic signal information, etc.

[0055] Step 2: The cloud virtual simulation model conducts a safety risk assessment on the vehicle safety status based on the vehicle status information and current traffic information. The method used is as follows:

[0056] First, the relationship between various risk factors is uniformly expressed in time and space, and then the risk is quantified. The field theory is introduced into the risk assessment of the traffic system, and the three types of risks (field strength) generated by various elements (field sources) of scene traffic to vehicles are reconstructed: the "potential field" determined by static elements in the road environment, the "kinetic field" determined by moving objects on the road, and the "behavior field" determined by the characteristics of vehicle driving behavior, which are finally unified into a driving risk field. The vehicle driving risk field model is constructed by establishing mathematical models of the kinetic field, potential field and behavior field formed by various elements in the driving environment, and the total field strength of the driving risk field of the vehicle is calculated; according to the effects of various human-vehicle-road elements in the surrounding environment on driving risks, the vehicle driving risk field model is constructed by establishing mathematical models of the kinetic field, potential field and behavior field formed by various elements in the driving environment, and the total field strength of the driving risk field of the vehicle is calculated;

[0057]

[0058] In the formula, is the total field strength vector of the driving risk field at the position of vehicle j, is the field strength vector at the position of vehicle j in the kinetic energy field, is the total vector strength of the potential energy field at the position of vehicle j, is the field strength vector of the behavior field at the position of vehicle j; and are the field strength vectors of a single kinetic energy field, potential energy field, and behavior field at the position of vehicle j

[0059] The driving risk coefficient of the vehicle is calculated based on the forces acting on the vehicle in the driving risk field.

[0060]

[0061] Where: F j is the force exerted on vehicle j in driving risk; is the total field strength of the driving risk field at point j; M j is the equivalent mass; R j is the road impact factor; v j is the speed of the vehicle; θ j is the angle between the vehicle and the speed direction; D j It is an undetermined risk factor, which is determined by the behavioral characteristics of different driving modes.

[0062]

[0063] Where: β risk is the driving risk coefficient of the vehicle, F j_max is the maximum force that vehicle j is subjected to in driving risk, F j_min The minimum value of the force applied to vehicle j in driving risk.

[0064] The control range of the autonomous driving control right is determined according to the driving risk coefficient of the autonomous driving vehicle.

[0065] The driving risk factor is divided into three ranges: safe, dangerous and very dangerous, and each range corresponds to a control right;

[0066] 1) Safety (vehicle status)

[0067] If β risk <W 1 , it means that the vehicle is in a safe state. At this time, the vehicle control is controlled by the autonomous driving vehicle. 1 It is the threshold value of the vehicle in a dangerous state, determined by the critical operational design domain and the expected functional safety.

[0068] 2) Danger

[0069] If W 1 ≤β risk <W 2 , it means that the vehicle is in a dangerous state. At this time, the vehicle control is controlled by the cloud controller, where W 2 It is the threshold of a very dangerous state, determined by the design operation domain and the intended functional safety.

[0070] 3) Very dangerous

[0071] If β risk ≥W 2 , it means that the vehicle is in a very dangerous state, and the control of the vehicle is controlled by the cloud safety officer.

[0072] The driving risk coefficient is calculated based on the vehicle status and the takeover requirement is output: if the vehicle status is safe, execute step three; if the vehicle status is dangerous, execute step four; if the vehicle status is very dangerous, execute step five.

[0073] Step 3: By calculating the current driving risk coefficient, it is in a safe state, indicating that the autonomous driving vehicle's perception device, decision-making device, and execution device are working properly, and the autonomous driving vehicle can make correct decisions based on the current environment and traffic conditions without the intervention of other devices. In this case, the autonomous driving vehicle makes decisions on the vehicle side and sends the driving data to the cloud virtual simulation model.

[0074] Step 4: By calculating the current driving risk coefficient, it is a dangerous state, which means that the autonomous driving vehicle may be close to the operating design domain, etc., and cannot make reasonable decisions based on its own situation. At this time, the cloud controller needs to take over the vehicle and send control instructions to the autonomous driving vehicle. The autonomous driving vehicle will hand over the control to the cloud controller and execute the instructions sent by the cloud controller. The autonomous driving vehicle will normally send the vehicle status and traffic information to the cloud control system.

[0075] In order to ensure the smooth switching of vehicle-cloud control and avoid the vehicle state affecting driving safety when the control is switched, cloud-based reinforcement learning agents based on decision modules and evaluation modules are constructed based on the parallel control concept and virtual system model. The motion state feedback of the autonomous vehicle during driving is used as the input of the reinforcement learning model. The cloud-based decision module of the reinforcement learning agent outputs the decision result x according to the current state, and the cloud-based evaluation module outputs the decision result x according to the state-action value function U. 1 (x 1 ,u 1 ) evaluates the decision result and calculates the new value function U 2 (x 2 ,u 2) as the vehicle-side decision input. The cloud-side decision module selects the optimal decision instruction Y according to the maximum judgment function value, and minimizes the value function difference between the cloud-side evaluation module and the vehicle-side judgment module, so that the cloud-side decision is close to the vehicle-side decision output, realizing a fast and smooth switch of control rights between the cloud and the vehicle.

[0076]

[0077] in, The value function of the cloud-side evaluation module and the vehicle-side evaluation module is the cumulative discount reward.

[0078]

[0079] When H(t)<δ, δ can be optimized and calibrated according to the needs of smooth control. The evaluation module outputs a trigger command to the vehicle-side decision module to complete the smooth handover of control from the vehicle-side to the cloud.

[0080] After the control is successfully handed over, the cloud should actively intervene to reduce driving risks and return control as soon as possible. The specific methods are as follows:

[0081] Establish the objective function of the car's decision-making problem based on the driving risk coefficient:

[0082]

[0083] Where: βk(xk,deck) represents the driving risk coefficient of the vehicle after executing the decision deck in the kth stage when it is in the xk state, β risk (t) is the optimal indicator function. (x j ,y j ) is the coordinate of the vehicle position in stage j, v j is the driving speed, δ i is the wheel angle, is the road adhesion coefficient, F X and F Y They represent the longitudinal force and lateral force on the vehicle tires respectively; f is the rolling resistance coefficient; C D is the air resistance coefficient; A is the frontal area of ​​the vehicle; M = m / L, W = C D A / 21.15,

[0084] When the vehicle is traveling at normal speed, the time t p After that, the position that the vehicle will reach can be calculated according to the formula.

[0085]

[0086] Let R(t) represent the turning radius of the vehicle, then:

[0087]

[0088] Where: K is the stability factor; L is the wheelbase of the vehicle; δ is the steering angle.

[0089] Safety-oriented cloud-based active intervention is the optimal solution process for the driving safety risk objective function, and ultimately the optimal path and speed decision sequence are obtained by reducing the driving risk coefficient.

[0090] Step 5: By calculating the current driving risk coefficient, it is a very dangerous state, which means that the vehicle may encounter a temporary controlled road, the automatic driving system fails, and the cloud controller cannot make reasonable decisions and cannot meet safety requirements. At this time, an alarm is issued to the remote driving platform to remind the cloud safety officer to take over the automatic driving vehicle. After the automatic driving vehicle transfers control, the cloud safety officer sends a remote control signal to the automatic driving vehicle by operating the remote driving platform, and the automatic driving vehicle receives the signal command and executes the operation.

[0091] In summary, the present invention can actively evaluate the safety status of the autonomous driving vehicle, and actively take over when the autonomous driving system breaks down or fails. The cloud controller or cloud safety officer actively takes over the vehicle, and there is no problem of stopping and waiting in the middle of the journey caused by the traditional takeover method. The continuity of the driving task can be guaranteed, passenger comfort can be increased, traffic efficiency can be improved, and safety can be improved.

[0092] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical scheme of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for remote takeover of an autonomous driving vehicle based on parallel control, characterized by: By using the parallel control method, parallel execution is performed on the autonomous driving vehicle side and the cloud side, thereby realizing a fast and smooth switch between cloud-side control and vehicle-side control, including: building a virtual simulation model on the cloud side, obtaining the current state information of the autonomous driving vehicle and the traffic information of the current environment of the autonomous driving vehicle, and calculating the driving risk coefficient; based on the parallel control concept and the virtual simulation model, building a cloud-based reinforcement learning intelligent body based on the decision-making module and the evaluation module, and through the parallel control method, parallel execution is performed between the vehicle-side control and the cloud-based virtual simulation. When the vehicle is driving normally, the vehicle-side decision is mainly made, and the cloud-based control is executed in parallel; the virtual simulation model is synchronized with the current vehicle state information and the environment in which it is located. The cloud controller matches the traffic information of the environment, evaluates the current safety status of the autonomous driving vehicle, calculates the total field strength of the vehicle's driving risk field, and calculates the driving risk coefficient of the vehicle based on the force acting on the vehicle in the driving risk field. When the vehicle is in a dangerous state and exceeds the decision-making ability of the vehicle, the cloud controller takes over safely and actively intervenes. Through parallel optimization and minimization of function differences, the cloud decision is close to the vehicle decision, realizing a fast and smooth switching of control rights between the cloud and the vehicle. After the control is smoothly transferred, the cloud reduces the driving risk through active intervention, and then returns the control to the vehicle. When the degree of danger is too high, it is handed over to the cloud safety officer to ensure driving safety.

2. The method for remote takeover of an autonomous driving vehicle based on parallel control according to claim 1 is characterized in that: The driving risk coefficient is calculated using the following method, and the control right of the autonomous driving vehicle is allocated; a) Calculation method of driving risk coefficient: According to the effects of various elements of people, vehicles and roads in the surrounding environment on driving risk, a vehicle driving risk field model is constructed by establishing mathematical models of the kinetic energy field, potential energy field and behavior field formed by various elements in the driving environment, and the total field strength of the driving risk field of the vehicle is calculated; In the formula, is the total field strength vector of the driving risk field at the position of vehicle j, is the field strength vector at the position of vehicle j in the kinetic energy field, is the sum of the potential energy field strength vector strength at the vehicle j position, is the field strength vector of the behavior field at the position of vehicle j; and are the field strength vectors of a single kinetic energy field, potential energy field, and behavior field at the position of vehicle j respectively. b) Calculate the driving risk coefficient of the vehicle based on the forces acting on the vehicle in the driving risk field; Where: F j is the force exerted on vehicle j in driving risk; is the total field strength of the driving risk field at point j; M j is the equivalent mass; R j is the road impact factor; v j is the speed of the vehicle; θ j is the angle between the vehicle and the speed direction; D j It is an undetermined risk factor, which is determined by the behavioral characteristics of different driving modes; Where: β risk is the driving risk coefficient of the vehicle, F j_max is the maximum force that vehicle j is subjected to in driving risk, F j_min The minimum value of the force applied to vehicle j in driving risk; c) Determine the range of control rights for autonomous driving according to the driving risk coefficient of the autonomous driving vehicle, where the driving risk coefficient is divided into three ranges: safe, dangerous and very dangerous, and each range corresponds to a control right; 1) Safety (vehicle status) If β risk <W1, it indicates that the vehicle is in a safe state. At this time, the vehicle control right is controlled by the autonomous driving vehicle end, where W1 is the threshold value of the vehicle in a dangerous state, which is determined by the operating design domain and the expected functional safety. 2) Danger If W1 ≤ β risk <W2, it indicates that the vehicle is in a dangerous state. At this time, the vehicle control right is controlled by the cloud controller, where W2 is the threshold of the extremely dangerous state, which is determined by the critical operation design domain and the expected functional safety. 3) Very dangerous If β risk ≥W2, it means that the vehicle is in a very dangerous state, and the control of the vehicle is controlled by the cloud safety officer.

3. The method for remote takeover of an autonomous driving vehicle based on parallel control according to claim 1 is characterized in that: The virtual simulation model includes: an autonomous driving vehicle model, a traffic participant model, a road traffic facility model and an environmental condition model; the autonomous driving vehicle model is a model constructed based on the characteristics of the autonomous driving vehicle; the traffic participant model is a model constructed based on the characteristics of other vehicles and pedestrians other than the autonomous driving vehicle; the road traffic facility model is a model constructed based on roads and traffic facilities, and the environmental condition model is a model constructed based on environmental information.

4. The method for remote takeover of an autonomous driving vehicle based on parallel control according to claim 1 is characterized in that: The vehicle status information includes vehicle speed information, vehicle position information, steering wheel angle information, and brake and accelerator pedal opening information.

5. The method for remote takeover of an autonomous driving vehicle based on parallel control according to claim 1 is characterized by: The cloud-based reinforcement learning agent based on the decision-making module and the evaluation module is constructed, including: The motion state feedback of the autonomous vehicle during driving is used as the input of the reinforcement learning model. The cloud-based decision-making module of the reinforcement learning agent outputs the decision result x according to the current state, and the cloud-based evaluation module outputs the decision result x according to the state-action value function U. 1 (x 1 ,u 1 ) evaluates the decision result and calculates the new value function U 2 (x 2 ,u 2 ) as the vehicle-side decision input. The cloud-side decision module selects the optimal decision instruction Y according to the maximum judgment function value, and minimizes the value function difference between the cloud-side evaluation module and the vehicle-side judgment module, so that the cloud-side decision is close to the vehicle-side decision output, realizing a fast and smooth switch of control rights between the cloud and the vehicle. in, The value function of the cloud-side evaluation module and the vehicle-side evaluation module is the cumulative discount reward. When H(t)<δ, δ can be optimized and calibrated according to the needs of smooth control. The evaluation module outputs a trigger command to the vehicle-side decision module to complete the smooth handover of control from the vehicle-side to the cloud.

6. The method for remote takeover of an autonomous driving vehicle based on parallel control according to claim 5 is characterized by: After the control is successfully transferred, the driving risk factor is reduced by selecting the optimal vehicle speed and wheel turning angle with the goal of minimizing the driving risk factor, and then the control is returned to the vehicle end; the specific method is as follows: Establish the objective function of the car's decision-making problem based on the driving risk coefficient: Where: β k (x k ,dec k ) indicates that the vehicle is at x in the kth stage k Execute decision dec in state k The driving risk coefficient after risk (t) is the optimal indicator function. (x j ,y j ) is the coordinate of the vehicle position in stage j, v j is the driving speed, δ i is the wheel angle, is the road adhesion coefficient, F X and F Y They represent the longitudinal force and lateral force on the vehicle tires respectively; f is the rolling resistance coefficient; C D is the air resistance coefficient; A is the frontal area of ​​the vehicle; M = m / L, W = C D A / 21.15, When the vehicle is traveling at normal speed, the time t p After that, the position that the vehicle will reach can be calculated according to the formula. Let R(t) represent the turning radius of the vehicle, then: Where: K is the stability factor; L is the wheelbase of the vehicle; δ is the steering angle. Safety-oriented cloud-based active intervention is the optimal solution process for the driving safety risk objective function, and ultimately the optimal path and speed decision sequence are obtained by reducing the driving risk coefficient.

Images