Self-driving vehicle safety control method and system based on hierarchical scene self-adaption

Abstract

The invention discloses an automatic driving vehicle safety control method and system based on hierarchical scene self-adaption, and the method is realized based on a vehicle nonlinear control affine system model, and comprises a decision control layer, a finite state machine and a safety correction layer. The method comprises the following steps that a decision control layer senses environment information to generate control input and outputs target semantics to a finite-state machine; the finite-state machine switches between driving scene states based on the target semantics and a preset state jump condition, and schedules a corresponding security constraint set; the security correction layer receives the control input and the currently activated security constraint set, and performs minimum modification on the control input by solving a quadratic programming problem on the premise of meeting all activated basic security constraints; and the computer systematically executes a computer program and outputs a final safety control quantity to a vehicle actuator. According to the invention, the limitation in the prior art is solved, and dynamic balance of safety and efficiency under multiple scenes is realized.

Description

A hierarchical scenario-adaptive safety control method and system for autonomous vehicles Technical Field

[0001] This invention relates to the fields of vehicle safety control technology and autonomous driving, and in particular, to an autonomous vehicle safety control method and system based on hierarchical scene adaptation. Background Technology

[0002] Autonomous driving systems, as a product of the deep integration of artificial intelligence, environmental perception, and vehicle safety control technologies, have become a core development direction in the field of intelligent transportation. However, how to provide verifiable safety guarantees in highly dynamic interactive environments has become a major bottleneck. Existing hierarchical planning architectures and end-to-end methods struggle to provide deterministic safety commitments under the high uncertainty of open worlds. Against this backdrop, Control Barrier Functions (CBFs), due to their mathematical forward invariance, can transform safety constraints into linear inequality constraints of quadratic programming. By modifying the nominal controller at the minimum cost, they ensure system safety and have become a commonly used technique for real-time control of autonomous vehicles.

[0003] Traditional obstacle avoidance (CBF) schemes typically use Euclidean distance-based CBFs to define the safety set. Since vehicle dynamics is a second-order system, first-order CBFs cannot explicitly couple the control inputs to the constraint equations. Higher-order obstacle avoidance functions (HOCBFs) address this issue through Lie derivative recursive chains. While HOCBFs establish a theoretical framework for higher-order systems, their effectiveness in avoiding dynamic obstacles is poor. Therefore, HOCBFs incorporate DCBF theory to derive dynamic higher-order obstacle avoidance functions (DHOCBFs).

[0004] However, the Dynamic Higher-Order Control Barrier Function (DHOCBF) has significant limitations in practical applications: First, under extreme conditions, the amplified higher-order terms of the sudden change in the acceleration of the preceding vehicle generate huge negative drift terms, pushing the braking control input to the overtaking vehicle's actuator to its physical limits, resulting in an empty feasible region for the quadratic programming solver and failure of safety control. Second, the inherent local optimum characteristic of CBF makes it prone to getting trapped in local optima and overly conservative control in narrow interactive environments. In scenarios such as intersection passage, fixed lateral drift constraints cannot adapt to time-varying geometric constraints, making it difficult to balance formal safety and traffic efficiency.

[0005] In summary, to address the problems of infeasibility of optimization solutions under extreme conditions, vehicles' susceptibility to local optima in multi-task scenarios, and low traffic efficiency in the existing technologies, a hierarchical scenario-adaptive safety control method and system for autonomous vehicles is proposed. Summary of the Invention

[0006] The main objective of this invention is to provide a hierarchical scenario-adaptive safety control method and system for autonomous vehicles, in order to solve the problem that optimization solutions are not feasible under extreme conditions in the existing technology, as well as the problem that vehicles are prone to getting stuck in local optima and have low traffic efficiency in multi-task scenarios.

[0007] Based on the first main aspect of the present invention, a hierarchical scene-adaptive safety control method for autonomous vehicles is provided. This method is implemented based on a nonlinear control affine system model of the vehicle, which includes a decision control layer, a finite state machine, and a safety correction layer. The method includes the following steps:

[0008] The decision control layer perceives environmental information, generates control inputs, and outputs target semantics to the finite state machine; the environmental information includes the vehicle state, the surrounding obstacle state set, and the road topology.

[0009] The finite state machine switches between multiple predefined driving scenario states based on the target semantics and preset state transition conditions, and schedules the corresponding safety constraint set based on the current driving scenario state.

[0010] The safety correction layer receives the control input and the currently active safety constraint set, and minimizes the modification of the control input by solving a quadratic programming problem, provided that the vehicle state satisfies all active safety constraint sets.

[0011] The computer system executes the computer program and outputs the final safety control parameters to the vehicle actuators.

[0012] As a further preferred option, in the aforementioned method, the multiple predefined driving scenario states specifically include: basic mode, lane-changing obstacle avoidance mode, and intersection passage mode;

[0013] The preset state transition conditions are as follows:

[0014] When the vehicle receives a lane change command, its speed is higher than the vehicle in front, and the vehicle is not in the intersection area, the driving scenario state switches from the basic mode to the lane change and obstacle avoidance mode.

[0015] When the lateral deviation of the vehicle is less than the preset threshold, the driving scenario state returns from lane change and obstacle avoidance mode to basic mode.

[0016] When a vehicle enters an intersection area, the driving scenario switches from the basic mode to the intersection passage mode, and after the vehicle passes the intersection conflict point, it returns to the basic mode from the intersection passage mode.

[0017] As a further preferred embodiment, in the aforementioned method, the security constraint set includes a basic security constraint set and a task-oriented constraint set;

[0018] The basic safety constraint set includes time-prediction-based road boundary constraints and speed-adaptive dynamic elliptical obstacle avoidance constraints.

[0019] The task-oriented guidance constraint set includes lane-change guidance constraints and intersection traffic constraints.

[0020] As a further preferred embodiment, in the aforementioned method, the time-prediction-based road boundary constraints include left road boundary constraints and right road boundary constraints, and the specific construction and execution steps are as follows:

[0021] By confining the vehicle to a feasible area, and based on the Frenet coordinate system model in the nonlinear control affine system model of the vehicle, combined with the current lateral deviation of the vehicle, the tangent angle between the vehicle's heading and the center line of the road, and the prediction step size, the lateral position of the vehicle at the next moment is predicted to obtain the predicted value.

[0022] Based on the predicted values, left road boundary constraints and right road boundary constraints are constructed.

[0023] As a further preferred embodiment, in the aforementioned method, the left road boundary constraint is as follows:

[0024]

[0025] The right road boundary constraint is shown in the following formula:

[0026]

[0027] in, This indicates the left road boundary constraint. Indicates the right road boundary constraint. Indicates the predicted value. Represents the state vector. This represents the minimum lateral deviation within the feasible region. This represents the maximum lateral deviation within the feasible region.

[0028] As a further preferred embodiment, in the aforementioned method, the steps for constructing the velocity-adaptive dynamic elliptical obstacle avoidance constraint are as follows:

[0029] A double-circle envelope model is used to describe obstacles. First, the physical center of an obstacle sub-circle is defined. At the same time, anchor points that dynamically change with the vehicle speed are introduced to obtain a virtual sub-circle, and the physical center of the virtual sub-circle is defined. The radius of the obstacle sub-circle is the same as that of the virtual sub-circle.

[0030] Based on the physical center of the obstacle sub-circle and the physical center of the virtual sub-circle, the center position of the dynamic ellipse is calculated, and the major axis and minor axis of the dynamic ellipse are further calculated.

[0031] Finally, based on the center position of the dynamic ellipse, a dynamic ellipse control function is constructed;

[0032] The dynamic elliptic control function is shown in the following equation:

[0033] in, This represents a dynamic elliptical obstacle avoidance constraint based on velocity adaptation. This represents the dynamic elliptic control function. Represents the state vector. Indicates the vehicle's position. Indicates the safety margin parameter. Indicates the center of the dynamic ellipse;

[0034] Specifically, the safety margin parameter is greater than or equal to the sum of the polar radius of the dynamic ellipse in the vehicle direction and the radius of the vehicle's envelope circle.

[0035] As a further preferred embodiment, in the aforementioned method, the steps for constructing the intersection traffic constraints are as follows:

[0036] The system obtains the longitudinal distance, longitudinal relative speed, and current lateral deviation of the vehicle and the obstacle at a preset potential intersection conflict point, and calculates a conflict time; the preset potential intersection conflict point is the longitudinal overlap position between the vehicle and the obstacle.

[0037] Based on the conflict time, the intersection traffic constraints are constructed.

[0038] The intersection traffic constraints are visualized through a dynamic parameter set.

[0039] Based on a second key aspect of the present invention, a hierarchical scene-adaptive autonomous driving vehicle safety control system is provided, which applies the aforementioned hierarchical scene-adaptive autonomous driving vehicle safety control method, including:

[0040] The perception and decision-making module is used to perceive environmental information, generate control inputs, and simultaneously output target semantics.

[0041] A finite state machine module, which has a built-in safety constraint library, is used to switch between multiple preset driving scenario states according to the target semantics and preset transition conditions, and to schedule the set of safety constraints in the safety constraint library based on the current driving scenario state.

[0042] The safety constraint library contains a set of safety constraints pre-stored in relation to the preset driving scenario state;

[0043] A safety correction module, which incorporates a quadratic programming safety controller, is used to schedule the quadratic programming safety controller to minimize the modification of the control input based on the control input and the currently active set of safety constraints.

[0044] The output module is used to output the final safety control values.

[0045] According to a third key aspect of the present invention, an electronic device is provided, comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0046] The memory stores a computer program, which, when executed by the processor, causes the processor to perform the aforementioned hierarchical scene-adaptive autonomous vehicle safety control method.

[0047] Based on a fourth principal aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon.

[0048] When the program is executed, it implements the aforementioned hierarchical scene-adaptive autonomous vehicle safety control method.

[0049] Compared with existing technologies, this invention provides a hierarchical scene-adaptive safety control method and system for autonomous vehicles. Firstly, addressing the problem that existing dynamic high-order control obstacle functions are prone to quadratic programming infeasibility under extreme conditions, this invention proposes a dynamic elliptical control obstacle function. By explicitly coupling velocity information into the definition of the safety set, the relative order of the constraints is mathematically reduced, eliminating the influence of sudden acceleration changes in obstacles on the constraint derivatives. Simultaneously, the active contraction characteristic of the safety envelope is utilized to increase the safety margin, ensuring the feasibility of solving the control algorithm under extreme dynamic conditions.

[0050] Secondly, by actively contracting the safety envelope, the present invention utilizes spatial margin to achieve earlier and smoother braking operations, optimizes the smoothness of the control process, expands the low-intensity braking area, and improves driving safety and comfort.

[0051] Finally, this invention introduces a hierarchical control architecture based on a finite state machine. Through dynamic scheduling, it avoids excessive compression of the solution space caused by the simultaneous activation of all constraints, solves the constraint conflict problem in multi-task scenarios, and ensures the system's adaptability to complex traffic environments. At the same time, it introduces task-oriented guidance constraints based on spatiotemporal prediction, dynamically adjusts the constraint combination according to the decision intention of the upper layer, guides vehicles to actively avoid impassable areas, breaks the traditional local optimum trap, and improves traffic efficiency. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.

[0053] Figure 1 shows the execution flowchart of a hierarchical scene adaptive autonomous vehicle safety control method and system according to an embodiment of the present invention;

[0054] Figure 2 shows a decision control layer architecture diagram of a hierarchical scene adaptive autonomous vehicle safety control method and system in one embodiment of the present invention.

[0055] Figure 3 shows a finite state machine architecture diagram of a hierarchical scene adaptive autonomous vehicle safety control method and system in one embodiment of the present invention.

[0056] Figure 4 shows a safety correction layer architecture diagram of a layered scenario-adaptive autonomous vehicle safety control method and system according to an embodiment of the present invention. Detailed Implementation

[0057] The preferred embodiments of the present invention will be described in detail below to provide a clearer understanding of the purpose, features, and advantages of the invention. It should be understood that the following embodiments are not intended to limit the scope of the invention, but are merely illustrative of the essential spirit of the technical solution of the invention.

[0058] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known techniques associated with the invention may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0059] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0060] The following is a description of the specific meanings of technical terms, English abbreviations, and formula parameters that may be used in this invention:

[0061] CBF: Control Barrier Function, a mathematical function used to ensure system safety. By designing a scalar function relating to the system state and constraining its rate of change over time, it is possible to formally guarantee that the system state always operates within a pre-defined safe set.

[0062] QP: Quadratic Programming, a special type of mathematical optimization problem, aims to find a set of decision variables that minimize or maximize a quadratic objective function while satisfying a set of linear inequalities and constraints.

[0063] Controlled affine systems: a class of nonlinear control systems whose dynamic equations can be expressed as the time derivative of the state vector equals a drift term that is only related to the state plus a control term, which is a linear combination of control inputs.

[0064] FSM: Finite State Machine, an abstract computational model for modeling the behavior of a system with a finite number of discrete states and transitions between these states based on input events or conditions.

[0065] DECBF: Dynamic Elliptic Controlled Barrier Function.

[0066] Cartesian coordinate system: a standard orthogonal coordinate system, also known as a rectangular coordinate system.

[0067] Frenet coordinate system: A curvilinear coordinate system based on a given reference curve. In this coordinate system, the position of any point is described by its longitudinal arc length and lateral offset.

[0068] Lie derivative: In differential geometry and control theory, a differential operation that describes the rate of change of a scalar field along the direction of a vector field.

[0069] As shown in Figure 1, in one embodiment, the present invention provides a hierarchical scene-adaptive autonomous vehicle safety control method. This method is implemented based on a vehicle nonlinear control affine system model, which includes a decision control layer, a finite state machine, and a safety correction layer.

[0070] The method includes the following steps:

[0071] S1, the decision control layer perceives environmental information, generates control input, and outputs target semantics to the finite state machine; the environmental information includes the vehicle state, the surrounding obstacle state set, and the road topology.

[0072] S2, the finite state machine switches between multiple predefined driving scenario states based on the target semantics and preset state transition conditions, and schedules the corresponding safety constraint set based on the current driving scenario state;

[0073] S3, the safety correction layer receives the control input and the currently active safety constraint set, and by solving a quadratic programming problem, minimizes the modification of the control input under the premise that the vehicle state satisfies all active safety constraint sets;

[0074] S4, the computer system executes the computer program and outputs the final safety control quantity to the vehicle actuator.

[0075] The computer system in this invention is an in-vehicle embedded computer system specifically adapted to the in-vehicle application scenarios of autonomous vehicles, and has a built-in vehicle nonlinear control affine system model.

[0076] It includes a processor, memory, and input / output interfaces, and is connected to vehicle sensors and actuators.

[0077] As shown in Figures 2 to 4, the autonomous vehicle safety control method based on hierarchical scenario adaptation proposed in this invention mainly consists of three parts: a decision control layer, a finite state machine (FSM), and a safety correction layer.

[0078] Specifically, the decision control layer is responsible for perceiving environmental information, including the vehicle's status, the status of surrounding obstacles, and the road topology.

[0079] Based on the above information, the decision layer generates control inputs that meet the task requirements and outputs target semantics containing scene semantic features to the FSM.

[0080] The safety correction layer, acting as the bottom-level safety filter, receives control inputs and the set of activated CBF constraints. By solving a quadratic programming (QP) problem, this layer minimizes the modification of the control inputs while constraining the vehicle state to meet all activated safety constraints, thereby outputting the final safety control quantity.

[0081] In one feasible implementation, vehicle dynamics and control affine system modeling is performed first.

[0082] To construct a safety controller based on a control barrier function, it is first necessary to establish a nonlinear control affine system model for the vehicle.

[0083] In this embodiment, the vehicle was modeled in both Cartesian and Frenet coordinate systems.

[0084] The affine system is described by the following closed-loop control affine nonlinear model:

[0085]

[0086] in, Represents the state vector. The dimension of the state vector. Indicates control input, Indicates the number of control inputs. and These represent the drift vector field and the control vector field, respectively.

[0087] In one embodiment, the motion model in the Cartesian coordinate system is described.

[0088] To balance computational efficiency, the vehicle is approximated as a single circular envelope, and obstacles are approximated as double circular envelopes.

[0089] Vehicle dynamics employs a nonlinear motion model, and its state vector is defined as follows: , where represent the x-coordinate, y-coordinate, velocity, and heading angle of the centroid in the envelope method, respectively.

[0090] Control input is defined as , representing acceleration and the tangent of the front wheel steering angle, respectively. This indicates the transpose operation.

[0091] The system dynamic equations are as follows:

[0092]

[0093] in, The x-coordinate of the centroid of the envelope method is represented by the first derivative with respect to time. The ordinate of the centroid of the envelope method is represented by the first derivative with respect to time. This represents the first derivative of velocity with respect to time. Indicates speed, The first derivative of the heading angle at the center of the rear axle of the vehicle with respect to time. Indicates the heading angle. Indicates acceleration. This refers to the vehicle's wheelbase. Indicates the steering angle of the vehicle's front wheels;

[0094] In one embodiment, the Frenet coordinate system model is described in detail:

[0095] To facilitate the construction of road boundaries and guidance constraints, the Frenet coordinate system is introduced.

[0096] The state vector is defined as ,in Indicates the longitudinal arc length. Indicates lateral deviation. For speed, The angle between the vehicle's heading and the tangent to the road's centerline. This indicates the transpose operation.

[0097] The Frenet coordinate system model is specifically as follows:

[0098]

[0099] in, This represents the first derivative of the longitudinal arc length with respect to time. Indicates the longitudinal arc length. This represents the first derivative of the vehicle's current lateral deviation with respect to time. This indicates the current lateral deviation of the vehicle. Indicates speed, This represents the first derivative of the angle between the vehicle's heading and the tangent to the road centerline with respect to time. This indicates the angle between the vehicle's heading and the tangent to the road's centerline. This represents the curvature of the road centerline at the current projection point. Indicates acceleration. Indicates the steering angle of the vehicle's front wheels;

[0100] This model is used to construct the left and right boundary constraints of roads. and .

[0101] Although the state vectors in the two coordinate systems are different, they share the same control input. Therefore, they can be integrated into the same quadratic programming (QP) problem for solution.

[0102] In one feasible implementation, the finite state machine and step S2 are described as follows:

[0103] To achieve dynamic coordination between safety constraints and task objectives, this embodiment designs a finite state machine. .

[0104] Where S represents the state space. Indicates the set of events that are triggered. This represents the state transition rule.

[0105] The state space S is defined as follows: These correspond to three driving scenarios:

[0106] Basic mode Applicable to regular road driving, only activates the basic safety constraint set. Includes road boundary constraints ( Includes left road boundary constraints and right road boundary constraints ) and dynamic elliptical obstacle avoidance constraints .

[0107] Lane change obstacle avoidance mode Activated when a lane change command is received. In this mode, in addition to basic safety constraints, lane change guidance constraints are also activated. It is used to guide the vehicle to complete the lateral position transition.

[0108] Intersection traffic patterns Activated when a vehicle enters the intersection area. In this mode, additional intersection guidance constraints are activated. It is used to handle dynamic game theory at intersections.

[0109] The state transition logic for the driving scenario is as follows:

[0110] When the upper-level planning issues a lane-changing instruction The vehicle's speed is higher than the vehicle in front and it is not at an intersection. At that time, from Jump to ;

[0111] When the lateral deviation is less than the set threshold When, the state transitions to return. .

[0112] Similarly, when entering the intersection area State transition to After passing the point of conflict, return .

[0113] in, Indicates a lane change instruction. Indicates the status of the intersection area. This indicates that the intersection has been cleared and the signal has been triggered.

[0114] In one embodiment, the basic security constraint set The detailed construction process is explained below:

[0115] Basic security constraint set It includes two types of constraints: road boundary constraints based on discrete-time prediction ( ) and speed-adaptive dynamic elliptical obstacle avoidance constraints ( ).

[0116] Road boundary constraints based on discrete-time prediction (Including left road boundary constraints) and right road boundary constraints .

[0117] First, explain the road boundary constraints ( The construction process of )

[0118] To overcome the errors of continuous-time CBF when dealing with discrete systems, this embodiment establishes road boundary constraints based on discrete state prediction.

[0119] Assume that vehicles must be restricted to the feasible area. Inside.

[0120] Using the Frenet coordinate system model, for the next time step ( ) vehicle lateral position Further predictions:

[0121]

[0122] in, This represents the current lateral deviation. For speed, The angle between the vehicle's heading and the tangent to the road's centerline. This indicates the prediction step size.

[0123] Based on this prediction, the following two control barrier functions are constructed:

[0124] Left road boundary constraints As shown in the following formula:

[0125]

[0126] Left road boundary constraints As shown in the following formula:

[0127]

[0128] in, This indicates the left road boundary constraint. Indicates the right road boundary constraint. This indicates the predicted value, i.e., the lateral position of the vehicle. Represents the state vector. This represents the minimum lateral deviation within the feasible region. This represents the maximum lateral deviation within the feasible region;

[0129] This constraint ensures that the vehicle remains within the physical boundaries of the road throughout the prediction time domain.

[0130] Furthermore, the dynamic elliptic control barrier function Perform the build (DECBF):

[0131] To address the unsolvable optimization problem caused by higher-order terms in DHOCBF when the preceding vehicle brakes suddenly, this embodiment designs a speed-modulated safety envelope that geometrically contracts when the vehicle decelerates, thereby increasing the relative safety margin.

[0132] Specifically, (1) definition of virtual bifocal configuration:

[0133] Obstacles are described using a double-circle envelope model.

[0134] set up The physical center of the obstacle subcircle (radius is) Introduce an anchor point that dynamically changes with the vehicle's speed. (A virtual sub-circle is obtained with radius of...) ), which is defined as follows:

[0135]

[0136] in, Indicates the speed scaling factor; Indicates speed; Indicates the heading angle; A constant represents the lower bound of scaling; This indicates the transpose operation.

[0137] Then calculate the center of the dynamic ellipse. :

[0138]

[0139] (2) Calculation of dynamic ellipse geometric parameters.

[0140] To ensure that the dynamic ellipse strictly contains both the obstacle subcircle and the virtual subcircle, the major axis of the dynamic ellipse... and short axis It is uniquely determined by the following formula:

[0141]

[0142]

[0143] in, Represents the radius of the obstacle subcircle or virtual subcircle.

[0144] (3) DECBF function definition.

[0145] Based on the above dynamic geometric parameters, a dynamic ellipse control obstacle function is defined:

[0146]

[0147] in, Represented by a velocity-adaptive dynamic elliptical loop obstacle avoidance constraint. This represents the dynamic elliptic control function. Represents the state vector. Indicates the vehicle's position. Indicates the safety margin parameter (satisfying) ,in Let be the polar radius of the dynamic ellipse in the direction of the vehicle. (where the radius is the radius of the vehicle's envelope circle). This represents the center of the dynamic ellipse.

[0148] Specifically, this construction method explicitly couples the velocity term into the safety set definition.

[0149] right When differentiating, the derivative of the velocity term (acceleration) appears directly in the first derivative, thus reducing the relative order with respect to the position constraint from 2 to 1.

[0150] This eliminates the acceleration term of obstacles when introducing DCBF theory, thus avoiding the acceleration caused by sudden braking of the vehicle in front. The sudden drop in speed leads to a sharp reduction in the QP solution domain. Furthermore, as the vehicle decelerates, the safety set expands, increasing the QP solution domain and enabling smoother braking.

[0151] In one embodiment, for a task-oriented guidance constraint set The detailed construction process is explained below:

[0152] This set contains lane change guidance constraints. ) and intersection traffic constraints ( The problem of discrepancies between the local optimum properties of the distance barrier function and the upper-level planning intent is addressed by using a finite state machine (FSM) for dynamic scheduling.

[0153] Specifically, (1) lane change guidance constraints ( ).

[0154] For lane-changing scenarios, in order to prevent vehicles from being stuck in the original lane, guidance constraints based on spatiotemporal convergence prediction are constructed.

[0155] First, the meeting time. calculate:

[0156] Define the longitudinal time required for the vehicle's center to reach the rear end of the envelope of the preceding vehicle's ellipse (an ellipse expanded with the vehicle's radius):

[0157]

[0158] in, This represents the relative distance between the outermost end of the elliptical envelope after the longitudinal obstacle expands and the center of the vehicle. Represents longitudinal relative velocity, This indicates the current lateral deviation of the vehicle. This represents the curvature of the road centerline at the current projection point.

[0159] Since the obstacle vehicle is represented by the double circle method, there will be two ellipses. Therefore, when defining the longitudinal time required for the center of the vehicle to reach the rear end of the elliptical envelope of the preceding vehicle, the ellipse closest to the vehicle should be selected.

[0160] Finally, constraints are constructed:

[0161] Based on convergence time Establish constraints that require vehicles to complete a lateral position transition before reaching the rendezvous point:

[0162]

[0163]

[0164] in, This indicates lane-change guidance constraints. Represents the state vector. This indicates the current lateral deviation of the vehicle. This represents the predicted lateral displacement during the convergence time. Indicates the center of the target lane. Indicates the meeting time. Indicates speed, This indicates the angle between the vehicle's heading and the tangent to the road's centerline. Indicates the direction of lane change, when Change lanes to the right when the wind is strong, and to the left when the wind is weak.

[0165] (2) Intersection traffic constraints ).

[0166] In intersection scenarios, the lateral intrusion of obstacles creates dynamically shrinking channels. Intersection traffic is modeled as a spatiotemporal accessibility problem.

[0167] First, conduct conflict time calculate:

[0168] Calculate the time it takes for the vehicle to reach the potential collision point (longitudinal overlap) with the obstacle. :

[0169]

[0170] in, Indicates the direction coefficient. , This represents the longitudinal distance between the center of the ellipse after the obstacle has expanded and the center of gravity of the vehicle. Indicates scene activation identifier, At this point, the longitudinal distance between the boundary of the ellipse at the center of the obstacle's ellipse and the nearest point of the vehicle is... Represents longitudinal relative velocity, This indicates the current lateral deviation of the vehicle. This represents the curvature of the road centerline at the current projection point.

[0171] Finally, constraint construction:

[0172]

[0173]

[0174]

[0175] in, Indicates traffic restrictions at the intersection. Represents the state vector. Indicates the direction coefficient. and These represent the predicted lateral positions of the vehicle and the obstacle at the time of the conflict. Indicates the lateral safety factor. Indicates the horizontal safety threshold. This indicates the current lateral deviation of the vehicle. Indicates the current lateral deviation of the obstacle. Indicates the time of conflict. Indicates speed, This indicates the angle between the vehicle's heading and the tangent to the road's centerline. Indicates the speed of the obstacle. Indicates the angle between the obstacle's heading and the tangent to the road centerline. Indicates the direction of lane change.

[0176] Parameter set Dynamically assign values ​​based on upper-level decisions:

[0177] If the decision is to rush: This restricts the vehicle from passing in front of obstacles;

[0178] If the decision is to yield: This restricts the vehicle from passing behind the obstacle.

[0179] in, Indicates the direction coefficient. Indicates the lateral safety factor. This indicates a scene activation identifier.

[0180] It is worth noting that since the obstacle is enveloped by a double circle, when rushing ahead, the ellipse is formed by the circle of the front of the car, and when yielding, the ellipse is formed by the circle of the following car.

[0181] In one feasible implementation, the security correction layer and step S3 are described as follows:

[0182] Finally, the safety correction layer solves the following quadratic programming (QP) problem:

[0183]

[0184] in, Indicates control input, As slack variables, This indicates finding the minimum value with respect to the control input and the slack variable. Indicates the nominal control input. Represents the weight matrix. Relaxing weights;

[0185] And meet the following conditions:

[0186]

[0187]

[0188] in, Represents the barrier function Regarding the control Lie derivative of the state vector Represents the barrier function Regarding the control Lie derivative of the control input. Represents the barrier function. The index identifier representing the basic set of security constraints. Represents the basic set of security constraints. Represents the guiding barrier function Regarding the control Lie derivative of the state vector Represents the guiding barrier function Regarding the control Lie derivative of the control input. Information indicating obstacles, Represents the state vector. Represents a time variable. Indicates the index identifier of the task-oriented constraint set. Indicates the first Extended K-class functions corresponding to each basic set of security constraints No. Extended K-class functions of a task-oriented constraint set This represents a task-oriented constraint set.

[0189] This design prioritizes basic safety constraints when multiple constraints conflict. While satisfying the needs, we should also take into account the guiding constraints as much as possible. To achieve the optimal balance between safety and efficiency.

[0190] The technical terms, principles, or means related to the technical solutions of the present invention mentioned in the above embodiments, which are not described in detail above, are all well-known technologies or common practices that are known to those skilled in the art.

[0191] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A safety control method for autonomous vehicles based on hierarchical scenario adaptation, characterized in that, This method is based on a vehicle nonlinear control affine system model, which includes a decision control layer, a finite state machine, and a safety correction layer. The method comprises the following steps: the decision control layer senses environmental information, generates control input, and outputs target semantics to the finite state machine; the environmental information includes the vehicle state, a set of surrounding obstacle states, and the road topology; the finite state machine switches between multiple predefined driving scenario states based on the target semantics and preset state transition conditions, and schedules the corresponding safety constraint set based on the current driving scenario state; the safety correction layer receives the control input and the currently active safety constraint set, and, by solving a quadratic programming problem, minimizes the modification of the control input while ensuring that the vehicle state satisfies all active safety constraint sets; the computer system executes the computer program and outputs the final safety control quantity to the vehicle actuators.

2. The autonomous vehicle safety control method based on hierarchical scenario adaptation according to claim 1, characterized in that, The predefined driving scenario states specifically include: basic mode, lane change and obstacle avoidance mode, and intersection passage mode; the preset state transition conditions are as follows: when the vehicle receives a lane change command, its speed is higher than the vehicle in front, and the vehicle is not in the intersection area, the driving scenario state transitions from basic mode to lane change and obstacle avoidance mode; when the vehicle's lateral deviation is less than a preset threshold, the driving scenario state returns from lane change and obstacle avoidance mode to basic mode; when the vehicle enters the intersection area, the driving scenario state transitions from basic mode to intersection passage mode, and after the vehicle passes the intersection conflict point, it returns from intersection passage mode to basic mode.

3. The autonomous vehicle safety control method based on hierarchical scenario adaptation according to claim 1, characterized in that, The safety constraint set includes a basic safety constraint set and a task-oriented constraint set; the basic safety constraint set includes road boundary constraints based on time prediction and dynamic elliptical obstacle avoidance constraints based on speed adaptation; the task-oriented constraint set includes lane-changing guidance constraints and intersection traffic constraints.

4. The autonomous vehicle safety control method based on hierarchical scenario adaptation according to claim 3, characterized in that, The time-prediction-based road boundary constraints include left road boundary constraints and right road boundary constraints. The specific construction and execution steps are as follows: restrict the vehicle to the feasible area, and based on the Frenet coordinate system model in the vehicle nonlinear control affine system model, combine the current lateral deviation of the vehicle, the angle between the tangent of the vehicle's heading and the road centerline, and the prediction step size to predict the lateral position at the next moment and obtain the predicted value; based on the predicted value, the left road boundary constraints and right road boundary constraints are constructed.

5. The autonomous vehicle safety control method based on hierarchical scenario adaptation according to claim 4, characterized in that, The left road boundary constraint is shown in the following formula: The right road boundary constraint is shown in the following formula: in, This indicates the left road boundary constraint. Indicates the right road boundary constraint. Indicates the predicted value. Represents the state vector. This represents the minimum lateral deviation within the feasible region. This represents the maximum lateral deviation within the feasible region.

6. The autonomous vehicle safety control method based on hierarchical scenario adaptation according to claim 3, characterized in that, The construction and execution steps of the speed-adaptive dynamic elliptical obstacle avoidance constraint are as follows: An obstacle is described using a double-circle envelope model. First, a physical center of an obstacle sub-circle is defined, and anchor points that dynamically change with the vehicle's speed are introduced to obtain a virtual sub-circle. The physical center of the virtual sub-circle is then defined. The obstacle sub-circle and the virtual sub-circle have the same radius. Based on the physical centers of the obstacle sub-circle and the virtual sub-circle, the center position of the dynamic ellipse is calculated, and the major and minor axes of the dynamic ellipse are further calculated. Finally, based on the center position of the dynamic ellipse, a dynamic ellipse control function is constructed. The dynamic ellipse control function is shown in the following equation: in, This represents a dynamic elliptical obstacle avoidance constraint based on velocity adaptation. This represents the dynamic elliptic control function. Represents the state vector. Indicates the vehicle's position. Indicates the safety margin parameter. The center of the dynamic ellipse is indicated; specifically, the safety margin parameter is greater than or equal to the sum of the polar radius of the dynamic ellipse in the vehicle direction and the radius of the vehicle's envelope circle.

7. The autonomous vehicle safety control method based on hierarchical scenario adaptation according to claim 3, characterized in that, The steps for constructing the intersection traffic constraints are as follows: The longitudinal distance between the center of the expanded ellipse of the obstacle and the centroid of the vehicle, the longitudinal distance between the center of the obstacle ellipse and the nearest point of the vehicle from the ellipse boundary, the longitudinal relative velocity, and the current lateral deviation of the vehicle are obtained. The conflict time between the vehicle and the obstacle at the potential conflict point is calculated. The potential conflict point is the longitudinal overlap position between the vehicle and the obstacle. Based on the conflict time, the intersection traffic constraints are constructed. The intersection traffic constraints are visualized using a dynamic parameter set.

8. A hierarchical scene-adaptive autonomous vehicle safety control system, employing the hierarchical scene-adaptive autonomous vehicle safety control method as described in any one of claims 1-7, characterized in that, include: The perception and decision-making module is used to perceive environmental information, generate control inputs, and simultaneously output target semantics. A finite state machine module, which has a built-in safety constraint library, is used to switch between multiple preset driving scenario states according to the target semantics and preset transition conditions, and to schedule the set of safety constraints in the safety constraint library based on the current driving scenario state. The safety constraint library contains a set of safety constraints pre-stored in relation to the preset driving scenario state; A safety correction module, which incorporates a quadratic programming safety controller, is used to schedule the quadratic programming safety controller to minimize the modification of the control input based on the control input and the currently active set of safety constraints. The output module is used to output the final safety control values.

9. An electronic device, characterized in that, include: The system includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory stores a computer program, and when the program is executed by the processor, the processor performs the hierarchical scene adaptive autonomous vehicle safety control method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed, it implements the autonomous vehicle safety control method based on hierarchical scene adaptation as described in any one of claims 1 to 7.

Images