A method for generating safety constraint trajectories for autonomous vehicles

Abstract

This invention discloses a method for generating a safety constraint trajectory for an autonomous vehicle, belonging to the field of autonomous driving technology. The method includes the following steps: S1, acquiring environmental state information, sensor observation information, and the vehicle's current state information; S2, constructing a safety constraint topology based on the risk uncertainty state vector, wherein the safety constraint topology includes a set of safe state nodes and a set of safe reachable state transition edges; S3, dynamically evolving the safety constraint topology based on the risk uncertainty state vector to update the set of safe state nodes and the set of safe reachable state transition edges; S4, generating the trajectory of the autonomous vehicle within the dynamically evolved safety constraint topology. This invention can effectively address uncertainties in the traffic environment, such as changes in the behavior of surrounding traffic participants and changes in the reliability of sensor information, thereby improving the robustness of the system.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically to a method for generating safety constraint trajectories for autonomous vehicles. Background Technology

[0002] With the rapid development of autonomous driving technology, ensuring the safety of autonomous vehicles in different traffic environments has become a critical issue that urgently needs to be addressed. Currently, many autonomous driving systems employ rule-based safety constraint methods, such as setting hard rules like minimum safe distances and maximum accelerations to ensure vehicle safety. However, these rules are typically static and do not take into account uncertainties in complex dynamic environments, especially prediction errors in the behavior of traffic participants and changes in the reliability of sensor information.

[0003] In addition, most existing trajectory generation methods rely on fixed safety boundaries or deep learning-based path search algorithms. When faced with highly uncertain traffic environments, these methods often cannot adjust safety constraints in real time, resulting in generated trajectories that may not be robust enough or are too conservative, thus affecting vehicle driving efficiency.

[0004] Therefore, a new method is needed to adjust and generate trajectories that meet safety constraints in real time based on uncertainties and risk information around the vehicle in a dynamically changing traffic environment, so as to ensure the safety and driving efficiency of autonomous vehicles. Summary of the Invention

[0005] The purpose of this invention is to provide a method for generating safety constraint trajectories for autonomous vehicles. By introducing risk uncertainty state vectors and dynamically evolving safety constraint topology, this method solves the problem that existing technologies cannot effectively cope with uncertainty and complex environments, thereby improving the safety and driving efficiency of autonomous vehicles in dynamic traffic environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for generating a safety constraint trajectory for an autonomous vehicle, comprising the following steps: S1. Obtain environmental state information, sensor observation information, and vehicle current state information of the autonomous vehicle, and construct a risk uncertainty state vector based on the environmental state information, the sensor observation information, and the vehicle current state information; S2. Construct a safety constraint topology based on the risk uncertainty state vector. The safety constraint topology includes a set of safe state nodes and a set of safe reachable state transition edges. S3. Dynamically evolve the safety constraint topology based on the risk uncertainty state vector to update the set of safe state nodes and the set of safe reachable state transition edges; S4. Generate a vehicle trajectory within the state space defined by the safety constraint topology after dynamic evolution. The vehicle trajectory consists of a set of continuous vehicle states, and each vehicle state in the vehicle trajectory belongs to the set of safe state nodes. Furthermore, the state transition relationship between adjacent vehicle states belongs to the set of safe reachable state transition edges.

[0007] Preferably, the risk uncertainty state vector includes the following four components: uncertainty index of surrounding traffic participant behavior, change index of environmental perception credibility, uncertainty index of vehicle's own state estimation, and change index of environmental observability.

[0008] Preferably, the risk uncertainty state vector is obtained through a risk uncertainty mapping function, which is used to map the environmental state information, the sensor observation information, and the vehicle current state information into the risk uncertainty state vector.

[0009] Preferably, the risk uncertainty state vector is obtained through a risk uncertainty mapping function, which is used to map the environmental state information, the sensor observation information, and the vehicle current state information into the risk uncertainty state vector.

[0010] Preferably, the security constraint topology is represented in the form of a directed graph. ,in The set of safe state nodes represents a subset of the vehicle state space, and the set of safe reachable state transition edges represents the safe state node set. This represents the vehicle state transition relationship allowed under the aforementioned risk uncertainty state vector constraint.

[0011] Preferably, the dynamic evolution of the security constraint topology is determined by the dynamic evolution function Ψ( The dynamic evolution function is implemented such that: in The safety constraint topology of the previous time step. Given the risk uncertainty state vector at the current time step, the dynamic evolution function generates the vector by performing a pruning operation on the set of safe state nodes and the set of safe reachable state transition edges. .

[0012] Preferably, the pruning of the set of safe state nodes is achieved through a node feasibility determination function, which is a Boolean function. ,when At that time, remove the vehicle state from the set of safety state nodes from the previous time step. .

[0013] Preferably, the pruning of the set of safe reachable state transition edges is achieved through a state transition feasibility determination function, which is a Boolean function. ,when At that time, remove the edges from the set of safe reachable states from the vehicle state in the previous time step. Vehicle status The directed edge.

[0014] Preferably, the dynamic evolution of the security constraint topology follows a continuity constraint rule: when At that time, the dynamically evolved safety constraint topology Safety-constrained topology before dynamic evolution A subset of, i.e. .

[0015] Preferably, generating the vehicle trajectory within the state space defined by the safety constraint topology includes: solving the following optimization problem within the search space defined by the safety constraint topology to obtain the vehicle trajectory. : in, Given a sequence of vehicle states, and for any two adjacent vehicle states... satisfy and Objective function J( The planning objective is used to characterize the trajectory and does not change the structure of the safety constraint topology during the solution process.

[0016] Preferably, the node feasibility determination function and the state transition feasibility determination function are executed in the following order: the safety constraint topology of the previous time step is first preliminarily screened to remove nodes and edges that directly conflict with the risk uncertainty state vector, and then the remaining nodes and edges are verified a second time based on vehicle dynamics feasibility and nodes and edges that do not meet the vehicle dynamics constraints are removed; the second verification removes all edges connected to the node at the same time as removing the corresponding node.

[0017] This invention provides a method for generating safety constraint trajectories for autonomous vehicles. It has the following advantages: 1. Adapting to complex dynamic environments: By introducing a risk uncertainty state vector, this invention can effectively cope with uncertainties in the traffic environment, such as changes in the behavior of surrounding traffic participants and changes in the reliability of sensor information, thereby improving the robustness of the system.

[0018] 2. The balance between safety and efficiency: Through dynamically evolving safety constraint topology, this invention can flexibly adjust the trajectory generation strategy while maintaining safety, thereby improving vehicle driving efficiency.

[0019] 3. Real-time adjustment of safety constraints: Unlike the static safety rules of traditional machine learning, the safety constraint topology of this invention can adjust the safety boundary according to real-time risk information, so that the trajectory generation process can flexibly respond to the needs of different scenarios.

[0020] 4. No additional security constraint determination required: The method of this invention enables the trajectory generation process to be completed directly within the security constraint topology, avoiding additional security determination for each generated trajectory and improving computational efficiency. Attached Figure Description

[0021] Figure 1 This is an overall framework diagram of the present invention; Figure 2 This is a schematic diagram illustrating the process of constructing the risk uncertainty state vector of the present invention; Figure 3 This is a schematic diagram illustrating the security constraint topology construction and dynamic evolution process of the present invention; Figure 4 This is a schematic diagram of the trajectory generation process of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] See Figures 1-4 This invention provides a method for generating safety constraint trajectories for autonomous vehicles, applicable to generating vehicle trajectories that meet safety constraints based on environmental perception information during autonomous vehicle operation. The method is based on knowledge graph technology, constructing a risk uncertainty state vector, and then building and dynamically evolving a safety constraint topology based on this vector. The vehicle trajectory is generated within the state space defined by the safety constraint topology, thereby achieving a unification of the trajectory generation process and safety constraints.

[0024] The method of this invention can be executed by the planning system in an autonomous vehicle. Each step of the method can be implemented by a computer program executed by a processor, or by dedicated hardware or a combination of hardware and software. This invention does not limit the specific vehicle type, sensor type, or underlying control model.

[0025] In one specific embodiment, the method of the present invention operates within the autonomous driving system of an autonomous vehicle. The autonomous driving system includes at least a perception system, a localization system, a planning system, and an execution system.

[0026] The perception system is used to acquire information about the vehicle's surrounding environment, including road structure information, static obstacle information, and dynamic traffic participant information. The positioning system is used to acquire information about the vehicle's current state, which includes at least the vehicle's position, speed, and heading.

[0027] The planning system generates a vehicle trajectory based on environmental state information, the vehicle's current state information, and sensor observation information. The planning system includes a trajectory generation module, and the method of this invention is executed by the trajectory generation module.

[0028] The execution system is used to generate control commands based on the trajectory output by the trajectory generation module, and then apply the control commands to the vehicle's actuators.

[0029] The method of the present invention does not limit the communication method between the systems, and the systems can interact with each other through the vehicle bus or communication interface.

[0030] The symbols and terms used in the embodiments are defined as follows: Let Represents the discrete time step; the vehicle state vector is denoted as... ,in, This represents the vehicle's x-coordinate in the global coordinate system. This represents the vehicle's ordinate in the global coordinate system. Indicates the speed of the vehicle. The heading angle of the vehicle is represented by ; the control input vector is denoted as . ,in, Indicates the vehicle's acceleration. The vehicle's steering angular velocity is represented by the risk uncertainty state vector. Safety constraint topology is denoted as Among the states For a set of nodes in a safe state, Let be the set of safe and reachable state transition edges; the topological evolution function is denoted as . The node feasibility determination function is denoted as The state transition feasibility determination function is denoted as: Trajectory is represented as a sequence of vehicle states The trajectory objective function is denoted as .

[0031] The method of this invention is implemented in the planning module of an autonomous vehicle. The planning module exchanges data with the perception module, localization module, and vehicle execution module via an onboard communication bus or dedicated interface. The perception module outputs a set of environmental state information. The positioning module output includes the vehicle's current status. The set of sensor observation information is denoted as The following calculations are executed by a program running on the processor within the planning module, or by a dedicated circuit with equivalent functionality.

[0032] Step S1: As Figure 2 As shown, the construction of the risk uncertainty state vector; S1.1 Input. Get the current... Set of environmental state information at time Sensor observation information set and the current status of the vehicle .

[0033] S1.2 defines risk components. (In time...) Construct a risk uncertainty state vector; The meanings of each component are as follows: 1) The Uncertainty Index of Surrounding Traffic Participant Behavior is a measure of the dispersion of predicted behavior of surrounding dynamic targets within a short time window. Its input is the predicted trajectory distribution of each identified target at several future times, calculated using a scalar value obtained by measuring the dispersion of probability distributions.

[0034] 2) An environmental perception reliability change index is used to characterize changes in the set of sensor observations. The data extracts the changes in observation consistency and confidence over time. Its inputs are the sensor confidence output and a multi-frame observation consistency measure; the output is a scalar value.

[0035] 3) The vehicle's own state estimation uncertainty index is used to characterize the vehicle's own state estimation error covariance or its measure output by the localization and fusion modules. Its input is the error estimation information from the localization module, and its output is a scalar value.

[0036] 4) The Environmental Observability Change Index characterizes the degree to which the observable environmental coverage of a vehicle at its current coordinates changes over time. Its inputs are sensor field of view, occlusion information, and map coverage information; the output is a scalar value.

[0037] S1.3 Risk Mapping Function. A risk uncertainty mapping function is used. Map the input to Mapping function The implementation includes performing feature extraction, time window statistics, predictive distribution analysis, and numerical normalization on the input data, outputting scalar metrics for each component. The mapping function is explicitly given in the implementation using deterministic operations, and each intermediate quantity in the mapping process is defined and maintained consistent throughout the implementation code.

[0038] In one specific implementation, the risk uncertainty mapping function Build it as follows.

[0039] First, regarding the uncertainty indicators of the behavior of surrounding traffic participants. From the set of environmental state information The system obtains the trajectory prediction distribution of each dynamic traffic participant within a predetermined prediction time window, calculates the covariance matrix of the trajectory prediction distribution, and constructs a behavioral uncertainty measure based on the trace value of the covariance matrix, thereby obtaining... .

[0040] Secondly, regarding the indicators of changes in environmental perception credibility From sensor observation information set The system acquires confidence information output from the multi-source sensing module and statistically analyzes the temporal changes in confidence information within a predetermined time window. Based on the degree of confidence fluctuation, it constructs a measure of perception reliability change, thereby obtaining... .

[0041] Secondly, regarding the uncertainty index of the vehicle's own state estimation. From the vehicle's current status information The system obtains the state covariance information output by the positioning and state estimation module, and calculates the state estimation uncertainty measure based on the covariance information, thereby obtaining... .

[0042] Finally, regarding indicators of observable environmental change... Based on environmental status information and sensor observation information The observable area information is used to calculate the proportion of observable coverage of the vehicle in the current time step, and the change in observable coverage over time is combined to construct a measure of environmental observability change, thereby obtaining... .

[0043] Risk uncertainty mapping function In the implementation process, the calculation results of each uncertainty component are subjected to uniform numerical normalization to ensure that each component is within a comparable numerical range, thereby forming a risk uncertainty state vector. Step S2 Construction of security-constrained topology, such as Figure 3 As shown; S2.1 Input. Receive the risk uncertainty state vector output from step S1. With the current vehicle state space definition .

[0044] S2.2 State Discretization and Node Candidate Set. For the vehicle state space... Discretization yields a set of candidate states. Discretization is performed according to the state representation used by the vehicle, for example, defining discrete grid points in the lateral, longitudinal, velocity, and heading dimensions. Each candidate node... This corresponds to the discrete states that a vehicle can take at a certain time step.

[0045] S2.3 Initial screening of candidate nodes. Perform a preliminary geometric feasibility check to eliminate candidate nodes that collide with static obstacles or are located outside the road. This yields the initial set of nodes. .

[0046] S2.4 Edge candidate generation. (For) Each pair of nodes in Generate edge candidate set (satisfying the adjacency relationship based on time index) Edge candidate generation is based on the vehicle motion model. If control input exists Make If the basic dynamic constraints are satisfied within the dynamic sampling accuracy range, then the pair is considered as an edge candidate.

[0047] S2.5 Initial Topology Construction. Let the initial security-constrained topology be... The initial topology is further filtered in subsequent steps using a decision function to become the final topology. .

[0048] Step S3: Dynamic evolution of the security constraint topology; S3.1 Input. Obtain the topology from the previous time step. With the current risk uncertainty vector r k .

[0049] S3.2 Definition of Topological Evolution. Topological evolution is represented as... Among them Ψ( The pruning function is defined as a function that performs pruning and updating of nodes and edges in a certain order. The pruning operation includes two parts: pruning based on risk constraints and pruning based on vehicle dynamics constraints. The pruning order is as follows: first, pruning that directly conflicts with risk uncertainty is performed, and then pruning based on vehicle dynamics feasibility is performed. This order is implemented as a fixed process in the implementation.

[0050] In one specific implementation, the function The implementation process includes the following steps.

[0051] First, perform node pruning based on risk uncertainty constraints.

[0052] For the set of safe state nodes Each node in The node feasibility determination function is adopted. A decision is made. The node feasibility decision function is used to determine the vehicle status. In the current state vector of risk uncertainty Whether retention is permitted under the represented risk conditions.

[0053] When the judgment result is not allowed, the node will be... x Remove from the set of safe state nodes.

[0054] The feasibility assessment of a node should include at least the following deterministic checks: Based on environmental and vehicle status information, determine whether the vehicle status corresponding to the node meets the minimum safe distance constraint with dynamic traffic participants within the prediction time window. Based on the environmental observability change index, determine whether the state of the corresponding node is within the observable coverage range at the current time step; Based on vehicle speed and heading information, determine whether the state corresponding to a node belongs to the set of permissible motion states under the current risk level.

[0055] The above checks are implemented using logical judgment, and each judgment condition has a clearly defined threshold in the implementation.

[0056] Secondly, edge pruning is implemented based on risk uncertainty constraints.

[0057] For the set of safe reachable state transition edges Each state transition edge in Adopting a state transition feasibility determination function Make a judgment.

[0058] When the determination result is not allowed, the corresponding state transition edge is removed from the set of safe and reachable state transition edges.

[0059] The determination of the feasibility of a state transition should include at least the following deterministic checks: Determine whether the state transition trajectory segment described by the vehicle motion model conflicts with the predicted trajectory of any dynamic traffic participant within the prediction time window; Determine whether the state transition meets the verification requirements under the current perceived credibility condition; Determine whether the state transition falls within the permissible range of control input changes under the current state of risk uncertainty.

[0060] After completing the node and edge pruning based on risk uncertainty constraints, the intermediate topology is obtained.

[0061] Subsequently, the intermediate topology is pruned based on vehicle dynamics constraints.

[0062] Vehicle dynamics constraints include longitudinal acceleration constraints, lateral acceleration constraints, and heading rate of change constraints. Nodes or state transition edges that do not satisfy the vehicle dynamics constraints are removed, and all state transition edges connected to that node are removed simultaneously when a node is removed.

[0063] In this embodiment, the function Ψ( The cutting process follows a fixed execution order: First, perform the trimming based on risk uncertainty constraints, then perform the trimming based on vehicle dynamics constraints; The execution order is implemented as a fixed process and does not change with the running state.

[0064] Through the above pruning and updating operations, the safety constraint topology of the current time step is obtained. Safety constraint topology is used to limit the state space for subsequent trajectory generation.

[0065] S3.3 Risk-driven node pruning. Each node in Adopting a node feasibility determination function Perform a decision. The decision function is a Boolean function: when Remove from the node set at that time. Node determination is based on the following checks: the minimum safe distance constraint between the node and dynamic obstacles, the observability threshold of the node at future moments, and the acceptability of the node's velocity / heading combination at the current risk level. The node determination process is deterministic and is implemented using well-defined logical expressions.

[0066] S3.4 Risk-Driven Edge Clipping. (For...) Each side of the middle Adopting a state transition feasibility determination function Perform a decision. The decision function is a Boolean function: when When the edge is removed from the edge set, the edge is removed. Edge determination is based on the following checks: the probability of the corresponding trajectory segment colliding with any dynamic target within the prediction time window, the verifiability of the trajectory segment under the current perception confidence, and whether the instantaneous requirements of the trajectory segment on vehicle dynamics during execution exceed the allowable range.

[0067] S3.5 Continuity constraint implementation. Define the aggregate risk metric function. Where the coefficient The weights are pre-defined non-negative weights, the selection of which is fixed in the implementation description and explicitly stated in the embodiments of the specification. In the implementation, the continuity constraint is written as: If but At the program implementation level, this constraint is guaranteed by only allowing node / edge removal operations and prohibiting node / edge addition operations. That is, under the condition that the risk increases or does not decrease, the topology only performs shrinkage operations.

[0068] S3.6 Secondary Dynamics Verification. After completing the risk-based pruning described above, perform vehicle dynamics feasibility verification on the remaining nodes and edges. Dynamics feasibility verification includes determining acceleration boundaries, steering angular velocity limits, and lateral acceleration limits. Nodes or edges that do not meet the dynamics feasibility requirements are removed from the set, and all edges connected to the removed nodes are also removed simultaneously.

[0069] S3.7 Final Topology Output. After the above trimming, the final safety-constrained topology is obtained. The final topology is used to define the trajectory search space in step S4.

[0070] Step S4 Trajectory generation within a constrained safety topology, such as Figure 4 As shown; S4.1 Input. Receive the final security constraint topology output from step S3. and the current status of the vehicle in .

[0071] S4.2 Trajectory Feasibility Constraints. The candidate trajectory set is constrained to be within... The sequence of all states in: The set is explicitly represented in the implementation as a data structure for graph traversal or path search.

[0072] S4.3 Establishment of the Trajectory Optimization Problem. Trajectory generation is formulated as a constrained optimization problem: The file objective function This is a deterministic function used to characterize the planning objective. (Under implementation) It may include a linear combination of the following terms: trajectory smoothness term, trajectory deviation from reference route term, control input magnitude term, and path feasibility term. The weights used are fixed in the embodiments and specified in the description. Important requirement: The solution process of the objective function must not change the safety constraint topology. The structure of the solver is such that it cannot be modified outside the constraint set. or .

[0073] S4.4 Solution Method. Based on the representation of the topology, a graph search or a numerical optimization method based on topological constraints is used to solve the problem. Graph search is implemented by... Using shortest path algorithms or costly depth-first / breadth-first search, the cost is determined by J( The solution is represented by a segmented structure. Numerical optimization is implemented by establishing discretized control variables in the path space defined by the topology and using a constrained optimization solver to obtain the optimal path. The solution algorithm is implemented with explicit steps and data structures, and the intermediate variables and convergence criteria are clearly defined in the implementation documentation.

[0074] S4.5 Output and Interface. The optimal trajectory obtained from the solution. The output is from the planning module to the vehicle control module. The output format includes a state sequence, corresponding timestamps, and the proposed control input sequence. The control module generates vehicle execution commands based on the output.

[0075] Regarding the temporal correlation between nodes and edges, the set of safe-state nodes and the set of safe-reachable state transition edges can simultaneously encode temporal indices (i.e., nodes are represented as...). This allows for direct consideration of temporal constraints in the planning process. Time indexes are used to maintain the comparability of nodes and ensure temporal consistency of states during edge generation.

[0076] Regarding the setting of weights and thresholds. These are used in the paper to calculate the aggregate risk metric. weight Furthermore, the threshold in the judgment function is a fixed parameter in the implementation. The specific value of the parameter is clearly defined and recorded as a system configuration in the embodiment, and is not considered as a limiting element of the present invention.

[0077] Regarding the flexibility of the implementation: To adapt to different vehicle platforms and sensor configurations, the granularity of state discretization, the fineness of the decision function, and the discretization step size for optimization can be adjusted during implementation. These adjustments must maintain consistency between the symbols and the meanings of the functions to ensure the logical coherence and reproducibility of steps S1 to S4.

[0078] Regarding the deterministic nature of the computation order, the implementation strictly follows this sequence: Step S1 → Step S2 → Step S3 → Step S4. In Step S3, the pruning order is risk pruning first, followed by dynamic pruning, and only shrinkage operations are allowed without reducing risk. This sequence represents a deterministic program flow, implemented using explicit function call sequences and data structure operations.

[0079] Regarding error tolerance and numerical stability, each mapping function, decision function, and solver should employ well-defined numerical tolerance and robustness handling strategies during implementation to ensure the determinism of topology evolution and trajectory solving under boundary conditions. These implementation details are listed in the engineering implementation documentation and are not intended to limit the scope of this invention.

[0080] The steps described in this embodiment can be executed by the vehicle's central processing unit or by distributed processing units in a coordinated manner. Structured information is exchanged between functional modules through predefined data interfaces. The specific operational steps, function expressions, and data structures described in the above embodiments are sufficient to enable those skilled in the art to implement this invention without inventive effort.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating a safety constraint trajectory for an autonomous vehicle, characterized in that, Includes the following steps: S1. Obtain environmental state information, sensor observation information, and vehicle current state information of the autonomous vehicle, and construct a risk uncertainty state vector based on the environmental state information, the sensor observation information, and the vehicle current state information; S2. Construct a safety constraint topology based on the risk uncertainty state vector. The safety constraint topology includes a set of safe state nodes and a set of safe reachable state transition edges. S3. Dynamically evolve the safety constraint topology based on the risk uncertainty state vector to update the set of safe state nodes and the set of safe reachable state transition edges; S4. Generate a vehicle trajectory within the state space defined by the safety constraint topology after dynamic evolution. The vehicle trajectory consists of a set of continuous vehicle states, and each vehicle state in the vehicle trajectory belongs to the set of safe state nodes. Furthermore, the state transition relationship between adjacent vehicle states belongs to the set of safe reachable state transition edges.

2. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 1, characterized in that, The risk uncertainty state vector includes the following four components: uncertainty index of surrounding traffic participant behavior, change index of environmental perception credibility, uncertainty index of vehicle's own state estimation, and change index of environmental observability.

3. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 1, characterized in that, The risk uncertainty state vector is obtained through a risk uncertainty mapping function, which maps the environmental state information, the sensor observation information, and the vehicle current state information into the risk uncertainty state vector.

4. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 1, characterized in that, The security constraint topology is represented in the form of a directed graph. ,in The set of safe state nodes represents a subset of the vehicle state space, and the set of safe reachable state transition edges represents the safe state node set. This represents the vehicle state transition relationship allowed under the aforementioned risk uncertainty state vector constraint.

5. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 1, characterized in that, The dynamic evolution of the security constraint topology is determined by the dynamic evolution function Ψ( The dynamic evolution function is implemented such that: in The safety constraint topology of the previous time step. Given the risk uncertainty state vector at the current time step, the dynamic evolution function generates the vector by performing a pruning operation on the set of safe state nodes and the set of safe reachable state transition edges. .

6. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 5, characterized in that, The pruning of the set of safe state nodes is achieved through a node feasibility determination function, which is a Boolean function. ,when At that time, remove the vehicle state from the set of safety state nodes from the previous time step. .

7. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 6, characterized in that, The pruning of the set of safe reachable state transition edges is achieved through a state transition feasibility determination function, which is a Boolean function. ,when At that time, remove the edges from the set of safe reachable states from the vehicle state in the previous time step. Vehicle status The directed edge.

8. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 7, characterized in that, The dynamic evolution of the security constraint topology follows the continuity constraint rule: when At that time, the dynamically evolved safety constraint topology Safety-constrained topology before dynamic evolution A subset of, i.e. .

9. The method for generating a safety constraint trajectory for an autonomous vehicle according to claim 1, characterized in that, Generating a vehicle trajectory within the state space defined by the safety constraint topology includes: solving the following optimization problem within the search space defined by the safety constraint topology to obtain the vehicle trajectory. : in, Given a sequence of vehicle states, and for any two adjacent vehicle states... satisfy and Objective function J( The planning objective is used to characterize the trajectory and does not change the structure of the safety constraint topology during the solution process.

10. A method for generating a safety constraint trajectory for an autonomous vehicle according to claim 7, characterized in that, The node feasibility determination function and the state transition feasibility determination function are executed in the following order: the safety constraint topology of the previous time step is first preliminarily screened to remove nodes and edges that directly conflict with the risk uncertainty state vector, and then the remaining nodes and edges are verified a second time based on vehicle dynamics feasibility and nodes and edges that do not meet the vehicle dynamics constraints are removed; the second verification removes all edges connected to the node at the same time as removing the corresponding node.

Images