A man-machine co-driving method based on a dynamic security arbitrator

Abstract

The application relates to a man-machine co-driving method based on a dynamic safety arbitrator. First, a model prediction control method is used to construct an autonomous control task of an intelligent vehicle, then a dynamic safety arbitrator is constructed to dynamically adjust the priorities of the autonomous control task and a human control task of the intelligent vehicle, and finally a man-machine shared control scheme is constructed to fuse the autonomous control and the human control of the intelligent vehicle when the intelligent vehicle cannot pass through a complex environment by relying on an autonomous control system, so that the robot can pass through smoothly. The method can effectively fuse the autonomous control and the human control of the intelligent vehicle, and realizes man-machine co-driving.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, and in particular to a human-computer co-driving method based on a dynamic safety arbitrator. Background Technology

[0002] Although the development of intelligent vehicles has entered a period of rapid progress, achieving significant advancements in environmental perception, driving decision-making, and precise control, drivers still cannot completely disengage from driving tasks at this stage. In Level 2 and Level 3 autonomous driving according to the SAE (Society of Automotive Engineers) classification, drivers must manually take over vehicle control when required by the system, such as when the autonomous driving system determines its functionality is limited or experiences a partial malfunction. Real-world Level 3 autonomous vehicles have experienced takeover scenarios in road tests, including dense pedestrian / bicycle environments, traffic light detection malfunctions, construction zones, unclear lane markings, and other vehicles cutting in. These scenarios share a common characteristic for drivers: complex traffic environments and a high cognitive load. It is worth noting that in Level 3 autonomous driving environments, because drivers do not need to constantly observe the traffic environment, they tend to engage in non-driving-related tasks. This leads to more safety issues and instability factors related to takeover, primarily including reduced situational awareness, prolonged reaction time, and unstable driving performance.

[0003] Traditional control handover primarily employs operation-triggered methods. When human intervention occurs, control is completely transferred from the system to human control using a binary mode (from 1 to 0). This necessitates immediate and effective takeover measures by humans upon regaining control; otherwise, drastic or unstable operations may lead to system instability. If control is gradually transferred during the handover process, the safety and stability of the takeover can be improved. Shared control is a control method that achieves this goal, generally categorized into direct shared control and indirect shared control.

[0004] Direct shared control means that the control from both parties acts directly on the actuators. This can lead to conflicts in driving authority between the driver and the automated system within a small deviation range, potentially causing mutual interference between human and autonomous control, resulting in control instability. Indirect shared control does not interfere with the driver's operation. The intelligent system does not directly participate in control but instead superimposes or modifies the driver's input and the controller input according to a certain ratio, transmitting the result to the vehicle's control actuators. Typically, this is a weighted sum of the driver's and controller inputs, with the weights equal to one. While indirect shared control, due to its non-intrusive nature, can effectively reduce conflicts between the driver and the control system and improve stability, the weighted fusion method can make the control output uninterpretable, potentially leading to safety hazards. Summary of the Invention

[0005] The purpose of this invention is to provide a human-machine co-driving method based on a dynamic safety arbitrator, which can effectively integrate intelligent vehicle autonomous control and human control to achieve human-machine co-driving.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a human-machine co-driving method based on a dynamic safety arbitrator. First, a model predictive control method is used to construct the autonomous control task of the intelligent vehicle. Then, a dynamic safety arbitrator is constructed to dynamically adjust the priority of the autonomous control task of the intelligent vehicle and the human control task. Finally, a human-machine shared control scheme is constructed. When the intelligent vehicle cannot rely on the autonomous control system to pass through complex environments, the autonomous control of the intelligent vehicle and the human control are integrated to help the robot pass through smoothly.

[0007] Furthermore, the autonomous control task of the intelligent vehicle is to drive according to a known trajectory. Model predictive control is used to enable the intelligent vehicle to track the known trajectory, and the output serves as the autonomous control quantity for the intelligent vehicle. Tracking the known trajectory is related to the vehicle's position and the trajectory position. The objective function is constructed as the difference between the vehicle's position and the reference path. To minimize energy loss and prevent abrupt changes in the control quantity, it is necessary to keep the control quantity as small as possible. Therefore, the square of the control quantity is added to the objective function, and then physical constraints on the control quantity are imposed.

[0008]

[0009] Where a and b are the objective function weights, x(k+i) is the intelligent vehicle state at time k+i, p(k+i) is the reference trajectory at time k+i, u(k+i) is the control input at time k+i, and N... p It is the prediction step size, u min It is the lower bound of the control variable, u max It is the upper limit of the control quantity.

[0010] Furthermore, after completing the construction of the autonomous control task for the intelligent vehicle, a dynamic safety arbiter is constructed using the control obstacle function as a quadratic programming constraint. The input priorities of humans and intelligent vehicles are dynamically adjusted according to the real-time task situation. The objective function and constraints of the dynamic safety arbiter are constructed for the input priority strategies of intelligent vehicles and humans in complex environments to ensure the safety of intelligent vehicles.

[0011] Furthermore, the dynamic safety arbitrator uses a control obstacle function and quadratic programming to prioritize the trajectory tracking task and the human-manually controlled obstacle avoidance task, and solves the decision quantity to adjust the priority of human and intelligent agent inputs.

[0012] The control objective of the dynamic safety arbitrator is to allow human intervention when encountering obstacles, prioritizing human input. The dynamic safety arbitrator's constraints are set as obstacle constraints to determine task control priority. In obstacle avoidance tasks, the vehicle must avoid obstacles detected by its sensors, and the distance between its current position and its current position must be greater than a safe distance. Therefore, the obstacle avoidance task safety condition is expressed as the following obstacle control function:

[0013] h(x)=Dd (2)

[0014] Where D is the distance between the intelligent vehicle and the obstacle, and d is the safe distance that the intelligent vehicle should maintain from the obstacle.

[0015] Furthermore, the objective function of the dynamic safety arbitrator is flexible and variable, enabling human-centered arbitration, where the objective function tracks human input as much as possible, or task-centered arbitration, where the objective function has no bias and only optimizes the task error. In simple environments, no human intervention is required, allowing the intelligent vehicle to control autonomously. In complex environments where the intelligent vehicle cannot pass autonomously, the input priorities of humans and the intelligent vehicle are adjusted. By integrating autonomous control of the intelligent vehicle and human control, the ability to pass through complex environments is guaranteed.

[0016] Since the control barrier function is used as a quadratic programming constraint to obtain the task priority decision quantity, the task priority decision quantity is defined as follows:

[0017] q∈{0,1} (3)

[0018] Where q=1 indicates that the autonomous control task of the intelligent vehicle takes priority, and q=0 indicates that the manual control task of the human takes priority.

[0019] The objective function is designed to enable the vehicle to drive autonomously within the safety set using the model predictive controller, tracking the input of the model predictive controller. A threshold q is set, and when only the model predictive controller fails to meet the safety constraints and human input is required to ensure the intelligent vehicle can safely complete the task, the control priority is transferred to the human. At the same time, the constraints of the obstacle function can maintain the safety of the entire process.

[0020] Furthermore, by utilizing the control barrier function, constraints are constructed for complex environments that require human intervention. When the intelligent vehicle cannot autonomously pass through the complex environment and will trigger the constraints, the dynamic safety arbitrator provides a decision quantity, adjusts the input priorities of humans and the intelligent vehicle, and mixes the inputs of humans and the intelligent vehicle to ensure that the constraints are met, i.e., the vehicle can pass through the complex environment.

[0021] For a given control barrier function h(x), the closed set C is defined as follows:

[0022]

[0023] Consider the following type of affine system:

[0024]

[0025] in, is the derivative of the system state variable, x is the system state variable, and u is the system control variable;

[0026] Define a set:

[0027] K cbf (x)={u∈U:L f h(x)+L g h(x)u+αh(x)≥0} (6)

[0028] Where U is the set that satisfies the following inequality u, and L f It is the Lie derivative of f(x) with respect to the state variables, L g It is the Lie derivative of g(x) with respect to the state variables, and α is the coefficient of h(x), which guarantees that the output is from K. cbf Choose from (x) to ensure the invariant set property of C;

[0029] Actual control quantity u r for:

[0030] u r =(1-q)u a +qu h (7)

[0031] Among them, u a It is the autonomous control quantity of the vehicle, u h It is a quantity controlled by humans;

[0032] Therefore, the control obstacle function is used as a constraint in the quadratic programming to generate the optimal decision controller, and the optimal decision quantity q is obtained. * :

[0033]

[0034] Furthermore, a human-machine shared control scheme is constructed using a zero-space control method. First, two basic tasks are determined: the autonomous control task of the intelligent vehicle and the human control task. The priority of the two tasks is obtained through a dynamic safety arbitrator. Then, the vector of the low-priority task is projected onto the zero space of the high-priority task to obtain the comprehensive output. The low-priority task is partially or fully completed while the high-priority task is completed.

[0035] Furthermore, regarding the autonomous control tasks of intelligent vehicles:

[0036] When within a safe set, the agent's task is to track a trajectory, defined as:

[0037] σ A =[XX ref YY ref (9)

[0038] Where, σ A This is a trajectory tracking task. X is the vehicle's x-coordinate, Y is the vehicle's y-coordinate, and X... ref It is the reference vehicle's x-coordinate, Y ref It is the longitudinal coordinate of the reference vehicle;

[0039] Taking the partial derivatives with respect to the system state variables yields the Jacobian matrix for this task:

[0040]

[0041] in, It is the derivative of the trajectory tracking task, J A is the Jacobian matrix of the task, and p is the system state variable;

[0042] Therefore, the task input is:

[0043]

[0044] in, It is the pseudo-inverse of the Jacobian matrix for this task;

[0045] For tasks controlled by humans:

[0046] Considering human intervention input:

[0047]

[0048] Where, δ h It is the input steering angle of the intelligent vehicle;

[0049]

[0050] Where, σ h It is a human-controlled task. It is the vehicle's yaw angle;

[0051] Taking the partial derivatives with respect to the system state variables, we obtain the Jacobian matrix for this task:

[0052]

[0053] in, It is the derivative of human control tasks, J h It is the Jacobian matrix for this task;

[0054] Therefore, the task input is:

[0055]

[0056] in, It is the pseudo-inverse of the Jacobian matrix for this task;

[0057] When within the safe set, no human intervention is required. In this case, the autonomous control task has higher priority than the human control task, so the task output u1 is:

[0058]

[0059] When encountering complex situations or uncertain environments, human intervention is required. In this case, the priority of human control tasks takes precedence over autonomous control tasks, and the task output u2 is:

[0060]

[0061] Compared with existing technologies, this invention has the following advantages: Addressing the human-machine co-driving problem in intelligent vehicle autonomous driving systems, this invention improves upon traditional control obstacle functions and quadratic programming methods by proposing a dynamic safety arbitrator. This arbitrator determines whether human intervention is required in the current situation, thus providing decision-making parameters to adjust the input priorities of both humans and the intelligent vehicle. Combined with a null-space-based control method, the inputs of humans and the intelligent vehicle are mixed. When the environment is simple and the intelligent vehicle can autonomously complete the task, the input priority of the intelligent vehicle is higher than that of the human. When the environment is complex and the intelligent vehicle cannot autonomously complete the task, requiring human assistance, the input priority of the human is higher than that of the intelligent vehicle. This human-machine shared control method not only compensates for the shortcomings of current autonomous control but also has advantages in solution time. Furthermore, compared to weighted control fusion methods, it has clear physical meaning. Attached Figure Description

[0062] Figure 1 This is a schematic diagram illustrating the implementation principle of the human-machine shared control scheme according to an embodiment of the present invention;

[0063] Figure 2 This is a trajectory diagram of an intelligent vehicle under human-machine shared control in an embodiment of the present invention;

[0064] Figure 3 This is a distance relationship diagram between the intelligent vehicle and obstacles under human-machine shared control in an embodiment of the present invention. Detailed Implementation

[0065] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0066] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0067] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0068] like Figure 1 As shown, this embodiment provides a human-machine co-driving method based on a dynamic safety arbitrator. First, a model predictive control method is used to construct the autonomous control task of the intelligent vehicle. Then, a dynamic safety arbitrator is constructed to dynamically adjust the priority of the autonomous control task of the intelligent vehicle and the human control task. Finally, a human-machine shared control scheme is constructed. When the intelligent vehicle cannot rely on the autonomous control system to pass through complex environments, the autonomous control of the intelligent vehicle and the human control are integrated to help the robot pass through smoothly.

[0069] This embodiment constructs a simulation using a vehicle kinematics model and three obstacles. The trajectory of the intelligent vehicle under human-machine shared control is as follows: Figure 2 As shown. The distance between the intelligent vehicle and obstacles under human-machine shared control is as follows. Figure 3 As shown. Human control input uses an Xbox 360 controller.

[0070] I. Construction of Autonomous Control Tasks for Intelligent Vehicles

[0071] The autonomous control task of an intelligent vehicle is to travel according to a known trajectory. Model predictive control methods are used to minimize the distance between the intelligent vehicle's position and the trajectory points, and physical constraints are imposed on the intelligent vehicle to enable it to autonomously track the trajectory.

[0072] In this embodiment, a model predictive control method is used to enable the intelligent vehicle to track a known trajectory, and the output serves as the autonomous control variable for the vehicle. Tracking the known trajectory is related to the vehicle's position and the trajectory position. The objective function is constructed as the difference between the vehicle's position and the reference path. To minimize energy loss and prevent abrupt changes in the control variable, the control variable should be kept as small as possible. Therefore, the square of the control variable is added to the objective function, and then physical constraints on the control variable are imposed.

[0073]

[0074] Where a and b are the objective function weights, x(k+i) is the intelligent vehicle state at time k+i, p(k+i) is the reference trajectory at time k+i, u(k+i) is the control input at time k+i, and N... p It is the prediction step size, u min It is the lower bound of the control variable, umax It is the upper limit of the control quantity.

[0075] II. Construction of Dynamic Security Arbitrator

[0076] After completing the construction of the autonomous control task for the intelligent vehicle, a dynamic safety arbiter is constructed using the control obstacle function as a quadratic programming constraint. The input priorities of humans and intelligent vehicles are dynamically adjusted according to the real-time task situation. The objective function and constraints of the dynamic safety arbiter are constructed for the input priority strategies of intelligent vehicles and humans in complex environments to ensure the safety of intelligent vehicles.

[0077] The dynamic safety arbitrator can dynamically adjust the priority of human and agent inputs while ensuring system safety. Furthermore, the objective function of the dynamic safety arbitrator can be flexibly set according to requirements. Control barrier functions are typically used as constraints in conjunction with quadratic programming to ensure safety when system performance and safety conflict. Traditional control barrier functions and quadratic programming directly solve for control variables to guarantee system safety. In this embodiment, the dynamic safety arbitrator uses control barrier functions and quadratic programming to prioritize trajectory tracking and human-manually controlled obstacle avoidance tasks, solving for decision variables to adjust the priority of human and agent inputs.

[0078] The control objective of the dynamic safety arbitrator is to allow human intervention in control when encountering obstacles, prioritizing human input. We set the constraints of the dynamic safety arbitrator as obstacle constraints to determine the task control priority. In obstacle avoidance tasks, the vehicle must avoid obstacles detected by its sensors, and the distance between its current position and its current position must be greater than a safe distance. Therefore, the safety conditions for obstacle avoidance tasks are expressed as the following obstacle control function:

[0079] h(x)=Dd (2)

[0080] Where D is the distance between the intelligent vehicle and the obstacle, and d is the safe distance that the intelligent vehicle should maintain from the obstacle.

[0081] III. Construction of the Objective Function for the Dynamic Security Arbitrator

[0082] The objective function of the dynamic safety arbitrator is flexible and variable, enabling either human-centered arbitration (where the objective function tracks human input as closely as possible) or task-centered arbitration (where the objective function has no bias and only optimizes task errors). In this embodiment, human intervention is minimized in simple environments, allowing the intelligent vehicle to control itself autonomously. In complex environments where autonomous vehicle control is insufficient, the input priorities of humans and the intelligent vehicle are adjusted. By fusing autonomous vehicle control and human control, passage through complex environments is guaranteed.

[0083] Since we use the control barrier function as a quadratic programming constraint to obtain the task priority decision quantity, we define the task priority decision quantity as follows:

[0084] q∈{0,1} (3)

[0085] Where q=1 indicates that the autonomous control task of the intelligent vehicle takes priority, and q=0 indicates that the manual control task of the human takes priority.

[0086] The objective function is designed to enable the vehicle to drive autonomously within the safety set using the model predictive controller, tracking the input of the model predictive controller. A threshold q is set, and when only the model predictive controller fails to meet the safety constraints and human input is required to ensure the intelligent vehicle can safely complete the task, the control priority is transferred to the human. At the same time, the constraints of the obstacle function can maintain the safety of the entire process.

[0087] IV. Construction of Constraints for Dynamic Security Arbitrators

[0088] By utilizing a control obstacle function, constraints are constructed for complex environments requiring human intervention. When an autonomous vehicle cannot navigate such an environment (i.e., it would violate the constraints), a dynamic safety arbitrator provides a decision value, adjusting the input priorities of both humans and the vehicle, and mixing their inputs to ensure the constraints are met, allowing the vehicle to pass through the complex environment. For example, when the vehicle encounters multiple obstacles on a reference trajectory, the distances between the vehicle and these obstacles are designed as a control obstacle function. Upon encountering an obstacle, the dynamic safety arbitrator prioritizes human input, mixing human and vehicle inputs to navigate around the obstacle.

[0089] For a given control barrier function h(x), the closed set C is defined as follows:

[0090]

[0091] Consider the following type of affine system:

[0092]

[0093] in, is the derivative of the system state variable, x is the system state variable, and u is the system control variable;

[0094] Define a set:

[0095] K cbf (x)={u∈U:L f h(x)+L g h(x)u+αh(x)≥0} (6)

[0096] Where U is the set that satisfies the following inequality u, and L f It is the Lie derivative of f(x) with respect to the state variables, L gIt is the Lie derivative of g(x) with respect to the state variables, and α is the coefficient of h(x), which guarantees that the output is from K. cbf Choose from (x) to ensure the invariant set property of C;

[0097] Actual control quantity u r for:

[0098] u r =(1-q)u a +qu h (7)

[0099] Among them, u a It is the autonomous control quantity of the vehicle, u h It is a quantity controlled by humans;

[0100] Therefore, the control obstacle function is used as a constraint in the quadratic programming to generate the optimal decision controller, and the optimal decision quantity q is obtained. * :

[0101]

[0102] V. Construction of Human-Machine Sharing Control Scheme

[0103] A human-machine shared control scheme is designed based on a zero-space control method. First, two basic tasks are determined: the autonomous control task of the intelligent vehicle and the human control task. The priority of the two tasks is obtained through a dynamic safety arbitrator. Then, the vector of the lower priority task is projected onto the zero space of the higher priority task to obtain the comprehensive output. The lower-priority task is partially or fully completed while the higher-priority task is completed.

[0104] In this embodiment, we use an Xbox 360 controller to input the steering angle of the intelligent vehicle as the human control input. We then use a null-space-based method to mix human and intelligent agent inputs, ensuring high priority for human input and avoiding conflicts in control tasks to achieve shared control. Therefore, we consider two basic tasks: autonomous control by the intelligent agent and human control, and obtain the priorities of these two tasks through the aforementioned dynamic safety arbitrator.

[0105] 1. Intelligent vehicle autonomous control task

[0106] When within a safe set, the agent's task is to track a trajectory, defined as:

[0107] σ A =[XX ref YY ref (9)

[0108] Where, σ A This is a trajectory tracking task. X is the vehicle's x-coordinate, Y is the vehicle's y-coordinate, and X... ref It is the reference vehicle's x-coordinate, Yref It is the longitudinal coordinate of the reference vehicle.

[0109] Taking the partial derivatives with respect to the system state variables yields the Jacobian matrix for this task:

[0110]

[0111] in, It is the derivative of the trajectory tracking task, J A is the Jacobian matrix for this task, and p is the system state variable.

[0112] Therefore, the task input is:

[0113]

[0114] in, It is the pseudo-inverse of the Jacobian matrix for this task.

[0115] 2. Human-controlled tasks

[0116] Considering human intervention input:

[0117]

[0118] Where, δ h It uses an Xbox 360 controller to input the steering angle of the smart car.

[0119]

[0120] Where, σ h It is a human-controlled task. It is the vehicle's yaw angle.

[0121] Taking the partial derivatives with respect to the system state variables, we obtain the Jacobian matrix for this task:

[0122]

[0123] in, It is the derivative of human control tasks, J h It is the Jacobian matrix for this task.

[0124] Therefore, the task input is:

[0125]

[0126] in, It is the pseudo-inverse of the Jacobian matrix for this task.

[0127] 3. Combined tasks

[0128] When within the safe set, no human intervention is required. In this case, the autonomous control task has higher priority than the human control task, so the task output u1 is:

[0129]

[0130] When encountering complex situations or uncertain environments, human intervention is required. In this case, the priority of human control tasks takes precedence over autonomous control tasks, and the task output u2 is:

[0131]

[0132] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for human-machine co-driving based on a dynamic security arbitrator, characterized in that, Firstly, the model predictive control method is used to build the autonomous control task of the intelligent vehicle, then a dynamic safety arbitrator is built to dynamically adjust the priority of the autonomous control task and the human control task of the intelligent vehicle, and finally a human-machine shared control scheme is built to help the robot pass through when the intelligent vehicle cannot rely on the autonomous control system to pass through the complex environment; After the autonomous control task of the intelligent vehicle is completed, the control barrier function is used as a quadratic programming constraint to build a dynamic safety arbitrator, which dynamically adjusts the input priority of the human and the intelligent vehicle according to the real-time task situation, and builds the objective function and constraint of the dynamic safety arbitrator for the input priority strategy of the intelligent vehicle and the human in complex environment to ensure the safety of the intelligent vehicle; The dynamic safety arbitrator is based on the control barrier function and quadratic programming to plan the priority of the trajectory tracking task and the human manual control obstacle avoidance task, and solves the decision variable to adjust the priority of the human and intelligent agent input; The control goal of the dynamic safety arbitrator is to let the human participate in control when encountering obstacles, and the input of the human is given priority; the constraint of the dynamic safety arbitrator is set as the obstacle constraint to determine the task control priority; in the obstacle avoidance task, the vehicle must avoid the obstacles detected by the sensor, and the distance between the position and the current position must be greater than the safety distance, so the obstacle avoidance task safety condition is expressed as the following control barrier function: h(x)=D-d (1) Where D is the distance between the intelligent vehicle and the obstacle, and d is the safety distance that the intelligent vehicle should maintain from the obstacle; Using the control barrier function, constraints are built for complex environments that require human control intervention. When the autonomous control of the intelligent vehicle cannot pass through the complex environment, i.e. the constraint is touched, the dynamic safety arbitrator gives the decision variable to adjust the input priority of the human and the intelligent vehicle, and mixes the input of the human and the intelligent vehicle, so that the constraint is satisfied, i.e. the complex environment can be passed through; For a given control barrier function h(x), define a closed set C as follows: Consider the following affine system: wherein is the derivative of the system state quantity, x is the system state quantity, and u is the system control quantity. Define a set as follows: K cbf (x) = {u e U: L f h(x) + L g h(x)u + a h(x) > 0} (4) where U is the set satisfying the following inequality u, L f is the Lie derivative of f(x) with respect to the state variable, L g is the Lie derivative of g(x) with respect to the state variable, and a is the coefficient of h(x), that is, to ensure that the output is selected from K cbf (x), thereby ensuring the invariance of C Actual control variable u r is: u r = (1 - q)u a + qu h (5) wherein u a is the vehicle autonomous control amount, u h is the human control amount; Thus the control barrier function is taken as a constraint to generate the optimal decision controller in a quadratic programming, and the optimal decision quantity q * is obtained A human-machine shared control scheme is built based on the zero space control method; first, determine two basic tasks: the autonomous control task of the intelligent vehicle and the human control task, and then project the low-priority task vector to the zero space of the high-priority task through the dynamic safety arbitrator to get the comprehensive output, which completes the high-level task while partially or completely completing the low-level task.

2. The method of claim 1, wherein the dynamic security arbiter is a hardware component. The autonomous control task of the intelligent vehicle is to travel according to the known trajectory, and the model predictive control method is used to make the intelligent vehicle track the known trajectory, and the output is used as the autonomous control variable of the intelligent vehicle; tracking the known trajectory is related to the position and trajectory position of the intelligent vehicle, and the target function is built as the difference between the vehicle position and the reference path; in order to minimize energy consumption and prevent control variable from changing abruptly, the control variable should be as small as possible, so the square of the control variable is added to the target function, and the physical limit of the control variable constraint is added again: wherein a, b are the weights of the objective function, x(k+i) is the state of the intelligent vehicle at time k+i, p(k+i) is the reference trajectory at time k+i, u(k+i) is the control amount at time k+i, N p is the predicted step, u min is the lower bound of the control amount, u max is the upper bound of the control amount.

3. The method of claim 1, wherein the dynamic security arbiter is based on, The target function of the dynamic safety arbitrator is flexible and variable, which can realize human-centered arbitration, i.e., the target function is to track human input as much as possible, or task-centered, i.e., the target function has no preference and only optimizes the task error; in a simple environment, no human intervention is performed, and the intelligent vehicle is autonomously controlled; in a complex environment where autonomous control of the intelligent vehicle cannot pass, the priority of the input of the human and the intelligent vehicle is adjusted, and the intelligent vehicle is autonomously controlled and the human control is fused to ensure passing through the complex environment; Since the control barrier function is used as a quadratic programming constraint to obtain the task priority decision variable, the decision variable for defining the task priority is: q∈{0,1}(8) Where q=1 indicates that the intelligent vehicle autonomous control task is preferred, and q=0 indicates that the human manual control task is preferred. The purpose of constructing the target function is to make the vehicle pass through the model predictive controller as much as possible in the safety set, automatically drive, and track the input of the model predictive controller; a threshold value of q is set, when only the model predictive controller does not meet the safety constraint, and the mixed human input is needed to ensure that the intelligent vehicle safely completes the task, the control priority is transferred to the human hand, and the constraint of the control barrier function can keep the whole process safe.

4. The method of claim 1, wherein the dynamic security arbiter is based on, For the intelligent vehicle autonomous control task: When in the safety set, the task of the intelligent agent is to track the trajectory, which is defined as: σ A = [X-X ref Y-Y ref ] (9) where σ A is the tracking trajectory task, X is the vehicle lateral coordinate, Y is the vehicle longitudinal coordinate, X ref is the reference vehicle lateral coordinate, Y ref is the reference vehicle longitudinal coordinate; The partial derivative of the system state quantity is taken to obtain the Jacobian matrix of the task: wherein, is the derivative of the tracking trajectory task, J A is the Jacobian matrix of the task, p is the system state quantity; Thus, the task input is: wherein is the pseudo-inverse of the task Jacobian matrix; For the human control task: Consider the human intervention input as: wherein δ h is the input steering angle of the intelligent vehicle; wherein σ h is a human control task, is a vehicle yaw angle; The partial derivative of the system state quantity is taken to obtain the Jacobian matrix of the task: wherein, is the derivative of the human control task, J h is the Jacobian matrix of the task; Thus, the task input is: wherein is the pseudo-inverse of the task Jacobian matrix; When in the safety set, no human input intervention is needed, and the priority of the autonomous control task is higher than that of the human control task, so the output u1 of the task is: When encountering complex situations or uncertain environments, human input intervention is needed, and the priority of the human control task is higher than that of the autonomous control task, so the output u2 of the task is:

Images