Real-time planning method for autonomous vehicles based on temporal logic reduction

Abstract

This invention relates to a real-time planning method for autonomous vehicles based on temporal logic reduction. The method includes: modeling a highway driving task using signal temporal logic language based on a custom predicate function set, obtaining a corresponding signal temporal logic reduction for the highway driving task, and storing it in an onboard path planner; the onboard path planner reads the signal temporal logic reduction instructions under the current driving state, and plans a trajectory set based on the extracted time constraints, state constraints, and vehicle state information of the driving task; and performs an asynchronous temporal robustness risk assessment on the trajectory set to select the optimal trajectory. Compared with existing technologies, this invention makes the autonomous driving system more robust and reliable, enabling it to participate in traffic safely and effectively.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving path planning, and in particular to a real-time planning method for autonomous vehicles based on temporal logic reduction. Background Technology

[0002] The core challenge of robot learning is enabling robots to perform various tasks according to human instructions, requiring robots to possess the ability to understand and execute diverse tasks. Large Language Models (LLMs), trained on massive amounts of internet text through self-supervised learning, have become an important tool for robots to understand and execute natural language instructions.

[0003] The proposed general framework utilizes LLMs as "co-pilots" in driving scenarios, automatically performing certain driving tasks through natural language processing. However, specific AI robot systems still face challenges such as bridging perception and symbolic representation, human-machine interactive learning, and integrating task and motion planning. In autonomous driving systems, precisely mathematically quantified driving tasks and traffic rule formulation are particularly important to bridge the gap between natural language instructions generated by LLMs and machine-executable logical instructions.

[0004] Formal methods are increasingly being used to ensure the safety of autonomous driving systems. Currently, formal modeling of traffic rules and driving tasks for autonomous driving mainly relies on ontology-based and first-order logic-based methods. Ontologies use Semantic Web Rule Language (SWRL) to qualitatively represent rules, which is easy to modularize, but lacks quantitative descriptions of specific traffic scenarios and temporal information, hindering the effective modeling of time-related tasks.

[0005] Temporal logic-based modeling can compensate for the shortcomings of ontology modeling. Metric Temporal Logic (MTL) allows for the specification of attribute time intervals. Related research uses MTL to quantitatively model traffic rules for autonomous vehicles under conditions of maintaining safe distance, speed control, and traffic convergence. Signal Temporal Logic (STL) is used to formalize the modeling of traffic rules and driving tasks, quantitatively expressing the system's satisfaction with the STL specification through robustness, and calculating robustness using a reinforced temporal algorithm. Existing research has proposed the concept of robustness trajectories and established STL specifications for intersection traffic rules, but the representation and modeling of driving tasks based on temporal logic remains insufficient.

[0006] Formal rules have wide applications in trajectory planning and monitoring. This research proposes a model-based predictive monitoring system using STL specifications, which can be integrated into trajectory planners to generate rule-compliant trajectories and play a role in real-time trajectory evaluation and correction. Rules are crucial in computing the reachability set of autonomous vehicles, enabling planners to consider rule-compliant trajectories. Rule manuals facilitate decision-making by prioritizing traffic rules and resolving conflicts when one rule must be violated to satisfy another. Neural networks are used to predict traffic violations, enhancing the understanding of human behavior and supporting better anticipation and response in autonomous driving systems. While existing work considers traffic rules in trajectory evaluation, robustness assessment of specifications with stronger temporal attributes remains insufficiently explored. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art by providing a real-time planning method for autonomous vehicles based on temporal logic reduction. This method makes the autonomous driving system more robust and reliable, enabling it to participate in traffic safely and effectively.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] A real-time planning method for autonomous vehicles based on temporal logic reduction, the method comprising:

[0010] Based on a custom predicate function set, signal timing logic language modeling is performed on highway driving tasks to obtain the corresponding signal timing logic reduction and store it in the vehicle path planner.

[0011] The on-board route planner reads the signal timing logic reduction instructions under the current driving state and extracts the time constraints, state constraints and vehicle state information of the driving task from them.

[0012] Based on the extracted time constraints, state constraints, and vehicle state information of the driving task, trajectory planning is performed on the vehicle to obtain an initial trajectory set. An asynchronous temporal robustness risk assessment is then conducted on the initial trajectory set to select the optimal trajectory.

[0013] Preferably, the step of performing signal-temporal logic language modeling on the highway driving task based on a custom predicate function set to obtain the signal-temporal logic specification corresponding to the highway driving task is specifically as follows: encoding the custom predicate function set, the logical operators of the signal-temporal logic, and the timing operators to perform signal-temporal logic language modeling on the highway driving task to obtain the signal-temporal logic specification corresponding to the highway driving task.

[0014] Preferably, the custom predicate function set includes:

[0015] Basic predicate functions are used to describe the relationship between the vehicle and the scene in terms of direction, lane, and rules;

[0016] Risk predicate function, used to quantitatively evaluate the asynchronous time robustness of the current signal sequential logic reduction;

[0017] The task predicate function is used to describe the execution logic of a driving task under the signal timing logic specification.

[0018] Preferably, the basic predicate is a binary predicate used to determine whether the corresponding signal timing logic specification is satisfied, including:

[0019] This indicates that comfort is improved by reducing speed variations, where G is a global operator and a is the vehicle acceleration;

[0020] Indicates whether the current vehicle speed v is less than or equal to the lane limit v. limit ;

[0021] This indicates whether the lane change from the current lane to the target lane involves crossing only the dashed line or the solid line.

[0022] Indicates the target lane type, including both entrances and exits;

[0023] This indicates whether there are relevant traffic participants in the left front, left rear, right front, right rear, front, and rear directions, respectively;

[0024] Lon_safety_distance(ego,o front )=G(|s ego -s o |≥s min ): Indicates whether to maintain a safe longitudinal distance from vehicles in front and behind, where ego and o front This refers to the state parameters of the vehicle and its adjacent vehicles, s ego s o These represent the longitudinal position of the vehicle and the longitudinal position of the surrounding vehicle, respectively. min Minimum vertical distance;

[0025] Lat_safety_distance(ego,o)=G(|d ego -d o |≥d min ): Indicates whether to maintain a safe lateral distance from adjacent vehicles, where d ego d o These are the lateral positions of the vehicle itself and the lateral positions of the surrounding vehicles, dmin This represents the minimum lateral distance.

[0026] Preferably, the risk predicate function Risk_monitor(φ) measures the asynchronous timing robustness of the signal timing logic reduction φ by the time difference between the arrival of the two vehicles at the same position.

[0027] Preferably, the task predicate function includes:

[0028] The predicate function for the overtaking task is Suipass(ego,o). front ):=s ego -s o ≥δ sur , among which ego, o front This refers to the state parameters of the vehicle and the vehicle in front, s ego s o δ represents the longitudinal position of the vehicle and the longitudinal position of the vehicle in front, respectively. sur The overtaking distance threshold is defined as the time when the longitudinal position of the vehicle exceeds that of the vehicle in front at a certain moment, indicating that the overtaking task is completed.

[0029] Change lane task predicate function Change_lane left / right :=|d ego,t -d target |≤δ d The subscripts left and right correspond to changing lanes to the left and right, respectively. ego d o These are the lateral positions of the vehicle itself and the lateral positions of the surrounding vehicles, d min Minimum lateral distance;

[0030] The predicate function for the emergency avoidance task is Evasion;

[0031] The safe parking predicate function Pull_over := v ego == 0.

[0032] The predicate function MaxBraking() triggers the maximum braking deceleration.

[0033] Preferably, the step of performing trajectory planning on the vehicle based on the extracted time constraints, state constraints, and vehicle state information of the driving task to obtain an initial trajectory set specifically involves:

[0034] Sampling of spatial position x and task completion time t, for a signal time-series logic reduction, a two-dimensional search space of horizontal time and vertical position is obtained. A series of discrete polynomial trajectories are generated in this two-dimensional search space, and an initial trajectory set is obtained after preliminary screening by kinematic boundaries.

[0035] Preferably, the spatial position x and the task completion time t are sampled. For a signal time-series logic reduction, a two-dimensional search space with horizontal time and vertical position is obtained. A series of discrete polynomial trajectories are generated in this two-dimensional search space. After preliminary screening by kinematic boundaries, an initial trajectory set is obtained, including:

[0036] Different state variables [s1,s1′,s1″,T] are selected for different signal timing logic specifications. ij =[s i ,s t ′ arget ,s t " arget ,T j ], where i and j represent the index of the vertical position and the index of the task completion time, respectively, and s t ′ arget s t " arget This represents the given target velocity and acceleration values, where s1, s1′, s1″, and T represent the endpoint position, velocity, acceleration, and task completion time.

[0037] The cost function in the process of generating polynomial trajectories is the smoothing cost J. t The weighted sum of time cost g(T) and endpoint constraint h(X1).

[0038] Preferably, the asynchronous temporal robustness risk assessment of the initial trajectory set to select the optimal trajectory includes:

[0039] According to vertical cost C lon and horizontal cost C lat With asynchronous timing robustness θ φ The weighted sum is used to construct the joint cost function C. tot The optimal trajectory is selected from the initial trajectory set.

[0040] Preferably, after selecting the optimal trajectory, the method further includes performing trajectory rolling planning, specifically:

[0041] At the start of each planning cycle, the vehicle controller gives instructions to calculate the long-term trajectory and short-term risk assessment. If there is no collision in the short term and a safe parking trajectory exists in the long term, the verification is passed and the feasibility of the current trajectory is recognized.

[0042] If a collision is likely to occur in the near future, the vehicle will use the safe stopping backup trajectory from the previous moment to decelerate. Once the previous moment passes verification, the vehicle will accelerate back to the original desired speed.

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

[0044] (1) The real-time planning method for autonomous vehicles based on the temporal logic specification of highway scenarios proposed in this invention uses signal temporal logic language to formulate driving tasks and traffic rules in a precise and mathematical way. It can solve the key problem of the gap between natural language and machine-executable logical instructions. It defines asynchronous temporal robustness to quantify the risk of trajectory not conforming to safety specifications and integrates this metric into trajectory planning to improve the safety of planning.

[0045] (2) The timing logic specification for all driving tasks in the highway scenario constructed by this invention covers the action instructions required by the autonomous driving system to handle various highway scenarios. It has good task coverage and provides a sound framework for dealing with the diverse needs of highway driving.

[0046] (3) Through rolling trajectory planning, if there is no collision in the short term and a safe stopping trajectory exists in the long term, the current trajectory is verified and deemed feasible; if a collision will occur in the short term, the safe stopping backup trajectory from the previous moment is adopted for deceleration, and the speed is accelerated back to the original expected speed after verification in the next moment. This long-term and short-term risk calculation architecture effectively avoids overly conservative behavior while ensuring safety and no collisions. Attached Figure Description

[0047] Figure 1 This is a framework diagram of the real-time planning method for autonomous vehicles based on temporal logic reduction in highway scenarios according to the present invention.

[0048] Figure 2 This is a risk definition diagram for asynchronous timing robustness in an embodiment of the present invention;

[0049] Figure 3 This refers to the calculation of drivable areas based on reachability set analysis in this embodiment of the invention.

[0050] Figure 4 This is the rolling planning framework for the method of the present invention;

[0051] Figure 5 The results are the trajectory simulation results in the embodiments of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0053] Example

[0054] like Figure 1As shown, this embodiment proposes a real-time planning method for autonomous vehicles based on temporal logic reduction in highway scenarios. The method includes the following steps:

[0055] Based on a custom predicate function set, signal timing logic language modeling is performed on highway driving tasks to obtain the corresponding signal timing logic reduction and store it in the vehicle path planner.

[0056] The on-board route planner reads the signal timing logic reduction instructions under the current driving state and extracts the time constraints, state constraints and vehicle state information of the driving task from them.

[0057] Based on the extracted time constraints, state constraints, and vehicle state information of the driving task, trajectory planning is performed on the vehicle to obtain an initial trajectory set. An asynchronous temporal robustness risk assessment is then conducted on the initial trajectory set to select the optimal trajectory.

[0058] The method of this embodiment will now be described in detail.

[0059] Step 1: Establish a general specification

[0060] General specifications refer to the requirements that autonomous vehicles must typically meet outside of driving tasks, including:

[0061] ① Maintain a safe longitudinal distance (mandatory requirement): Your vehicle must maintain a safe distance from the vehicle in front in the same lane to prevent collisions, even in the event of emergency braking. This rule applies to every vehicle within the sensor's range and must be assessed accordingly.

[0062]

[0063] Where: G represents the global operator, meaning the following condition must be satisfied at all times; This indicates logical implication; It refers to the attributes of the vehicle in front, specifically its existence and state of motion; ego, o front These refer to the state parameters of the vehicle and the vehicle in front, respectively; Lon_safe_distance() is the longitudinal safety distance predicate, representing a mapping function from state parameters to Boolean variables.

[0064] ② Maintain a safe lateral distance (mandatory requirement): Your vehicle must maintain a safe distance from vehicles in adjacent lanes to minimize the risk of collision, especially when changing lanes or when vehicles in adjacent lanes make sudden movements. This rule applies to all vehicles within the sensor's range and must be constantly assessed and followed to ensure that an appropriate lateral distance is maintained.

[0065]

[0066] Where: ∨ represents the logical "OR"; These are attributes of adjacent vehicles, referring to their existence and movement state; ego, o adiacent These refer to the state parameters of the vehicle and adjacent vehicles, respectively; `Lat_safe_distance()` is the lateral safety distance predicate. ③ Maximum speed limit (mandatory requirement): The vehicle must strictly adhere to the specified maximum speed limit to ensure safe and legal operation. This rule applies to all driving conditions and road types. The vehicle's speed should not exceed the prescribed limit and must be continuously monitored and adjusted to ensure compliance.

[0067]

[0068] in: It refers to the vehicle's speed attribute.

[0069] ④ Ensure comfort (soft requirement): The vehicle should minimize discomfort caused by sudden acceleration, deceleration or violent operation in order to provide a comfortable riding experience.

[0070]

[0071] in: It refers to the comfort attribute of the vehicle.

[0072] Step 2: Extract possible driving tasks from highway scenarios. Driving tasks typically specify a series of actions; therefore, temporal logic is a natural choice for the formal expression of driving tasks. Our formal specification is based on high-level decision-making instructions given by the upper-level decision-making system. Based on the general specification, this invention defines a comprehensive set of basic driving tasks for highway scenarios. These driving tasks are divided into two categories: tasks with reference targets and tasks without reference targets. As shown in Table 1, these tasks cover key operations such as overtaking, yielding, following, lane changing, cruising, safe stopping, entrance ramps, and exit ramps. By including these tasks, this invention ensures that the autonomous driving system has the action instructions required to handle various highway scenarios. This demonstrates the completeness of the architecture in terms of task coverage, providing a robust framework for addressing the diverse needs of highway driving.

[0073] Step 3: From natural language description of driving tasks to signal timing logic language.

[0074] As shown in Table 1 below, signal timing logic STL language modeling is performed for driving tasks in highway scenarios.

[0075] Table 1

[0076]

[0077] The first category consists of tasks with a reference target and their formal specifications are as follows:

[0078] ① Overtaking task: The action of a vehicle moving from its current lane to an adjacent lane to overtake a slower-moving vehicle ahead. This task is subject to specific rules and considerations designed to ensure the safety and efficiency of lane-changing operations.

[0079] The STL specification is as follows:

[0080]

[0081] in:

[0082] This indicates that there is a target vehicle in front of your vehicle.

[0083] Stay_in_lane U [0,t] Safe_to_change_lane: The vehicle should remain in the current lane until time t should satisfy the Safe_to_change_lane condition. U is the Until operator, indicating that the Stay_in_lane state should be maintained until time t.

[0084] This is a logical implied symbol, meaning that if the condition on the left (there is a car in front, stay in the lane before safely changing lanes) is true, then the condition on the right should also be true.

[0085] This indicates that within the time interval [0, t1], Change_lane left The event had to happen, which meant the vehicle should change to the left lane.

[0086] This indicates that within the time interval [0, Tt], the event Surpass(ego, o front This must happen, Change_lane left The event is completed within the time interval [0, t1].

[0087] In general, the overtaking rule can be interpreted as follows: if there is a vehicle in front, the vehicle must remain in its own lane until it is safe to change lanes. Then, within the time interval [0, Tt], the vehicle should overtake the vehicle in front and complete the left lane change within the range [0, t1].

[0088] ② Emergency Avoidance Tasks: In the field of autonomous vehicles, emergency avoidance is crucial, especially in situations where a vehicle in front suddenly decelerates, a vehicle behind rapidly approaches, or a vehicle unexpectedly enters the lane. These examples highlight the need for autonomous driving systems to demonstrate agile reaction capabilities and skillfully execute emergency avoidance maneuvers when faced with unexpected events.

[0089] The STL specification is as follows:

[0090] 1) Vehicle cut-in: This is defined as the lateral distance between the vehicle and the vehicle in front does not meet the safe distance, and the vehicle's speed relative to the vehicle in front is faster than the predetermined speed difference threshold.

[0091]

[0092] Where: s o s represents the longitudinal position of the chariot. ego Represents the longitudinal position of the vehicle; v o The speed attribute of the vehicle, v ego Represents the vehicle's speed attribute; δ s δ v These represent the threshold values ​​for distance and speed, respectively. When a vehicle suddenly cuts into the lane, the vehicle brakes at maximum deceleration until the distance to the vehicle in front is restored to a safe level.

[0093]

[0094] Where MaxBraking() represents the predicate function that triggers the maximum braking deceleration.

[0095] 2) Emergency braking of the vehicle in front or approaching vehicle behind: The scenario where the vehicle in front suddenly decelerates and the vehicle behind rapidly approaches can be defined by the STL formula:

[0096]

[0097] Where: a o The acceleration attribute of the vehicle, a ego Represents the vehicle's acceleration attribute; δ a These represent the threshold values ​​for acceleration.

[0098] If the risk monitoring module is triggered based on the conditions met by the vehicle itself and surrounding traffic participants, the system must ensure that evasive action is taken at some point within the corresponding time interval.

[0099]

[0100] Where: Risk_monitor() represents the predicate function for risk detection.

[0101] ③ Following vehicle task: The vehicle automatically maintains a safe distance from the vehicle in front on the highway, while maintaining a stable cruising speed.

[0102] The STL specification is as follows:

[0103]

[0104] The second category, tasks without a reference target, and their formal specifications are as follows:

[0105] ① Lane change task: The vehicle changes from the current lane to the adjacent lane.

[0106] The STL specification is as follows:

[0107]

[0108] ② Cruise mission: The action of maintaining the vehicle's position within a specific lane while driving on the road.

[0109] The STL specification is as follows:

[0110]

[0111] Where: v suggestion This indicates a suggested speed.

[0112] ③ Safe Pullover: The vehicle exits the main traffic flow and stops at the roadside or a designated parking area. This is typically used to handle various situations, such as emergencies, traffic violations, or to allow other vehicles to pass. In performing this task, the vehicle follows a series of steps to safely and efficiently transition from active driving to a stationary state.

[0113] The STL specification is as follows:

[0114]

[0115] ④Task 4 Entrance (Exit) Ramp: Entering and exiting a ramp can be seen as a lane-changing process, where vehicles transition from the main road to the ramp, and vice versa.

[0116]

[0117] Step 4: Extract predicates and functions from the signal-timing logic specification. Signal-timing logic is a formal logic language defined on predicates, requiring the definition of predicates and functions formalized for driving tasks. The basic elements are systematically organized and defined through a task-level hierarchical structure.

[0118] This invention defines predicate functions at three levels. Based on the custom predicate function set, signal timing logic reductions corresponding to various driving tasks can be constructed.

[0119] The first category is basic predicates.

[0120] The satisfaction of a binary predicate is determined by a feature function that outputs 0-1 variables to indicate whether the STL formula is satisfied under quantitative semantics. Basic predicates describe the relationship between the vehicle and the scene from three aspects: direction, lane, and rules. Due to space limitations, we cannot list all predicates here; however, a scene can be fully defined in this way using a finite number of predicates.

[0121] ① This indicates that comfort is improved by reducing speed variations;

[0122] ② Indicates whether the speed limit is met;

[0123] ③ Indicates whether the lane change from the current lane to the target lane only crosses the dashed line (solid line);

[0124] ④ Indicates the type of the target lane;

[0125] ⑤ These six predicates indicate whether there are relevant traffic participants in the left front, left rear, right front, right rear, front, and rear directions, respectively;

[0126] ⑥Lon_safety_distance(ego,o front )=G(|s ego -s o |≥s min ): Indicates whether a safe distance is maintained from vehicles in front and behind;

[0127] ⑦Lat_safety_distance(ego,o)=G(|d ego -d o |≥d min ): Indicates whether to maintain a safe distance from adjacent vehicles.

[0128] Where: v limit This refers to the speed limit for the corresponding lane; s min ,d min These are the minimum longitudinal distance and the minimum lateral distance, respectively, and their calculations reference the responsibility-sensitive safety model.

[0129] The second category is risk predicates.

[0130] The `Risk_monitor(φ)` function is used to quantitatively evaluate the asynchronous time robustness of a specification φ. By comparing it with a predetermined threshold, the spatiotemporal satisfaction of the specification φ can be determined. Figure 3In the given ramp example, the protocol ensures that the red and blue cars will not collide. Asynchronous timing robustness is measured by the time difference between the arrival of the two cars at the same location.

[0131] The third type is the task predicate function.

[0132] The task function clarifies the specific implementation and execution logic of driving actions under the STL specification. It inputs the time task and state constraints specified in STL into the controller and planner. Although real-time trajectory solving based on STL constraints has not yet been implemented, it is feasible to perform task satisfaction evaluation and generate trajectories that meet the constraints based on the STL specification.

[0133] ① The predicate function for the overtaking task is Surpass(ego, o front ):=s ego,t -s o,t ≥δ sur , among which ego, o front This refers to the state parameters of the vehicle and the vehicle in front, s ego s o δ represents the longitudinal position of the vehicle and the longitudinal position of the vehicle in front, respectively. sur This is the overtaking distance threshold.

[0134] The termination conditions for the autonomous driving task are determined and input into the onboard trajectory planner. The overtaking task is considered complete when the vehicle's longitudinal position surpasses that of the vehicle in front at a certain moment.

[0135] ② The predicate function for lane change tasks: Change_lane left / right :=|d ego,t -d target |≤δ d The subscripts left and right correspond to changing lanes to the left and right, respectively. ego d o These are the lateral positions of the vehicle itself and the lateral positions of the surrounding vehicles, d min This represents the minimum lateral distance.

[0136] When a left lane change is required, the target lateral position constraint will be input into the planner, and the system will then execute the left lane change task.

[0137] ③ Safe parking predicate function Pull_over := v ego,t ==0

[0138] This system enables drivers to quickly and safely pull over to the side of the road in emergency situations. It detects the driver's intention to pull over and assesses the surrounding environment to identify a safe parking location for the vehicle.

[0139] ④ Emergency avoidance task predicate function Evasion:

[0140] Using ensemble analysis to predict the movement trends of other traffic participants, such as Figure 3 As shown, the predicted occupancy area of ​​surrounding traffic participants is constructed, and the drivable area is extracted from it.

[0141] Step 5: Path planning.

[0142] This embodiment employs a path planner based on asynchronous temporal robustness risk assessment. It includes a set of custom predicate functions, logical operators for signal temporal logic, and a signal temporal logic reduction set encoded by these operators. By reading the signal temporal logic language instructions under the current driving state, and extracting the time constraints, state constraints, and vehicle state information of the driving task, it performs trajectory planning for the vehicle to obtain an initial trajectory set. An asynchronous temporal robustness risk assessment is then performed on this initial trajectory set to select the optimal trajectory, thus completing the path planning task.

[0143] Trajectory generation is used for autonomous driving path planning. Given an initial location and time constraints for the driving task, it generates an optimal path that satisfies safety (collision-free), efficiency, and STL (Standard Time Limits) specifications. A common method for trajectory generation is to represent the trajectory using a polynomial function.

[0144] Sampling the spatial location x and task completion time t reduces the search complexity compared to sampling the entire state space. Generally, for an STL specification, the planner ultimately obtains a two-dimensional search space of time and longitudinal location, in which a series of discrete polynomial trajectories are generated, and then preliminary screening is performed using kinematic boundaries. Of course, different state variables can be selected for different STL specifications.

[0145] [s1,s1′,s1″,T] ij =[s i ,s t ′ arget ,s t " arget ,T j (16)

[0146] Where: i and j represent the indices for the vertical position and the task completion time, respectively; s t ′ arget s t " arget The given target velocity and acceleration values ​​are represented by s1, s1′, s1″, and T, which represent the endpoint position, velocity, acceleration, and task completion time, respectively.

[0147] Considering the smoothing cost J tWe construct the cost function by considering the time cost g(T) and the endpoint constraint h(X1).

[0148] C = k j J t +k t g(T)+k f h(X1) (17)

[0149] Where: k j k t k f These are the weighting coefficients.

[0150] Considering the smoothing cost, time cost, and endpoint constraint, a cost function is constructed.

[0151] C = k j J t +k t g(T)+k f h(X1) (17)

[0152] Where: k j k t k f These are the weighting coefficients.

[0153] The joint cost function is a weighted sum of vertical cost, horizontal cost, and asynchronous time robustness. Calculating asynchronous time robustness requires transforming the coordinates from the Frenet coordinate system to the Cartesian coordinate system. By combining this with the objective function, the trajectory set is evaluated, and the optimal trajectory is selected.

[0154] C tot =k lat C lat +k lon C lon +k risk θ φ (18)

[0155] Figure 5 This paper outlines the trajectory generation process for overtaking maneuvers. At the initial time step, the overtaking command, state vector, and prediction time domain are received. First, a function is invoked to obtain the target region and constraints. Then, a set of trajectories is sampled and generated. The final step calculates the cost by checking comfort score, time score, and asynchronous time robustness conforming to the STL specification.

[0156] Step 6: Design the rolling planning framework. This embodiment designs a framework as follows: Figure 4The planning cycle shown is designed to perform rolling planning following STL instructions provided by the decision-making system. The goal of the planning process is to generate a trajectory that meets the specified STL requirements while ensuring safety and efficiency. At the beginning of each planning cycle, three decision instructions (such as...) will be received from the decision module. Figure 4 When c=1, three decision instructions (Task1, Task2, and Task3) are received. The system calculates the long-term (e.g., 5-second) trajectory and the short-term (e.g., 1.5-second) risk assessment. If there is no collision in the short-term and a safe stopping trajectory exists in the long-term, the AI ​​trajectory is verified and deemed feasible. If a collision is likely in the short-term, the previously established safe stopping backup trajectory is used for deceleration. Once verification is passed in the next moment, the system accelerates back to the original expected speed. This long-term and short-term risk calculation architecture effectively avoids overly conservative behavior while ensuring safety and preventing collisions.

[0157] Step 7: Design a simulation platform for the planning architecture proposed in this invention. This invention develops a simulation framework to realize path planning for autonomous vehicles operating on dynamic highways. Figure 5 This paper demonstrates how to convert driving tasks specified by Signal Sequential Logic (STL) into accurate and executable trajectories. The effectiveness of the proposed method is validated by effectively applying the STL specification and utilizing predefined predicates.

[0158] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A real-time planning method for autonomous vehicles based on temporal logic reduction, characterized in that, The method includes: Based on a custom predicate function set, signal timing logic language modeling is performed on highway driving tasks to obtain the corresponding signal timing logic reduction and store it in the vehicle path planner. The on-board route planner reads the signal timing logic reduction instructions under the current driving state and extracts the time constraints, state constraints and vehicle state information of the driving task from them. Based on the extracted time constraints, state constraints, and vehicle state information of the driving task, the vehicle trajectory is planned to obtain an initial trajectory set. An asynchronous time-series robustness risk assessment is then performed on the initial trajectory set to select the optimal trajectory. The custom predicate function set includes: Basic predicate functions are used to describe the relationship between the vehicle and the scene in terms of direction, lane, and rules; Risk predicate function, used to quantitatively evaluate the asynchronous time robustness of the current signal sequential logic reduction; Task predicate function, used to describe the execution logic of driving tasks under signal timing logic specification; The basic predicates are binary predicates, used to determine whether the corresponding signal timing logic specification is satisfied, including: This indicates that comfort is improved by reducing speed variations, where... This is a global operator. To accelerate the vehicle; : Indicates the current vehicle speed Is it less than or equal to the lane limit? ; : Indicates whether only the dashed line is crossed when changing lanes from the current lane to the target lane. or solid line ; : Indicates the target lane type, including entrances and exits; , , , , , : These indicate whether there are relevant traffic participants in the left front, left rear, right front, right rear, front, and rear directions, respectively; : Indicates whether to maintain a safe longitudinal distance from vehicles in front and behind, where This refers to the state parameters of the vehicle itself and the vehicles in front and behind it. These represent the longitudinal position of the vehicle and the longitudinal position of the surrounding vehicle, respectively. Minimum vertical distance; : Indicates whether to maintain a safe lateral distance from adjacent vehicles, where, These refer to the lateral position of the vehicle itself and the lateral position of the surrounding vehicles. This represents the minimum lateral distance.

2. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 1, characterized in that, The method of performing signal-temporal logic language modeling on highway driving tasks based on a custom predicate function set to obtain the signal-temporal logic specification corresponding to the highway driving task is as follows: encoding the custom predicate function set, the logical operators of signal-temporal logic, and the timing operators, performing signal-temporal logic language modeling on highway driving tasks, and obtaining the signal-temporal logic specification corresponding to the highway driving task.

3. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 1, characterized in that, The risk predicate function For signal timing logic reduction Asynchronous timing robustness is measured by the time difference between the arrival of two vehicles at the same location.

4. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 1, characterized in that, The task predicate function includes: Overtaking task predicate function ,in This refers to the state parameters of the vehicle and the vehicle in front. These represent the longitudinal positions of the vehicle and the preceding vehicle, respectively. The overtaking distance threshold is defined as the time when the longitudinal position of the vehicle exceeds that of the vehicle in front at a certain moment, indicating that the overtaking task is completed. lane change task predicate function subscript , These correspond to changing lanes to the left and changing lanes to the right, respectively. These refer to the lateral position of the vehicle itself and the lateral position of the surrounding vehicles. Minimum lateral distance; Emergency avoidance task predicate function ; Safe parking predicate function ; The predicate function that triggers maximum braking deceleration .

5. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 1, characterized in that, The initial trajectory set is obtained by performing trajectory planning on the vehicle based on the extracted time constraints, state constraints, and vehicle state information of the driving task. Specifically: The spatial location 𝑥 and task completion time 𝑡 are sampled. For a signal time-series logic reduction, a two-dimensional search space with horizontal time and vertical position is obtained. A series of discrete polynomial trajectories are generated in this two-dimensional search space. After preliminary screening by kinematic boundaries, an initial trajectory set is obtained.

6. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 5, characterized in that, The spatial location φ and task completion time φ are sampled. For a signal time-series logic reduction, a two-dimensional search space with horizontal time and vertical position is obtained. A series of discrete polynomial trajectories are generated in this two-dimensional search space. After preliminary screening by kinematic boundaries, an initial trajectory set is obtained, including: Different state variables are selected for different signal timing logic specifications. Where i and j represent the indices for the vertical position and the task completion time, respectively. , This represents the given target velocity and acceleration values. , , T represents the endpoint position, velocity, acceleration, and task completion time; The cost function in the process of generating polynomial trajectories is the smoothing cost. Time cost and endpoint constraints The weighted sum.

7. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 5, characterized in that, The asynchronous temporal robustness risk assessment of the initial trajectory set to select the optimal trajectory includes: Based on vertical costs and horizontal costs Robustness with asynchronous timing Weighted summation to construct joint cost function The optimal trajectory is selected from the initial trajectory set.

8. The real-time planning method for autonomous vehicles based on temporal logic reduction according to claim 1, characterized in that, After selecting the optimal trajectory, the process also includes trajectory rolling planning, specifically: At the start of each planning cycle, the vehicle controller gives instructions to calculate the long-term trajectory and short-term risk assessment. If there is no collision in the short term and a safe parking trajectory exists in the long term, the verification is passed and the feasibility of the current trajectory is recognized. If a collision is likely to occur in the near future, the vehicle will use the safe stopping backup trajectory from the previous moment to decelerate. Once the previous moment passes verification, the vehicle will accelerate back to the original desired speed.

Images