Highway emergency-oriented unmanned aerial vehicle cooperative path decision method and system

Abstract

The application belongs to the field of automatic driving, and relates to a method and system for highway emergency event-oriented unmanned aerial vehicle (UAV) cooperative path decision, which converts UAV observation data into structured variables, predicts the evolution trend of the emergency event in a future time window, calculates the opening probability of a lane and a lane recovery traffic window, comprehensively considers the lane opening probability, the lane recovery traffic window and action risk, generates a lane-action-time domain related temporary traffic right graph, and then performs preliminary screening on an original candidate path based on the temporary traffic right graph and the lane recovery traffic window, determines an upper limit of action intensity according to a UAV observation reliability index, performs secondary screening on a candidate path set based on the upper limit of action intensity, and finally selects an optimal path in combination with traffic time, path risk, comfort, task deviation and re-planning jitter, so that the method can improve the path decision foresight, execution stability and operation safety of a vehicle in a complex abnormal road environment.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving and relates to the drive control of autonomous road vehicles, specifically to a method and system for collaborative path decision-making by unmanned aerial vehicles (UAVs) for handling emergencies on highways. Background Technology

[0002] With the rapid development of autonomous driving and intelligent connected vehicle technologies, the perception, planning, and control capabilities of autonomous vehicles in highway environments are constantly improving. Highways are characterized by regular road structures, consistent traffic flow, high vehicle speeds, and strong traffic flow continuity, making them an important scenario for the early deployment of autonomous driving technology. However, once a sudden event occurs on a highway, such as an accident blocking the lane, debris intrusion, fog obscuring the view, or flooding and landslides, the road conditions ahead of the vehicle can change significantly in a short period of time, posing a significant challenge to the autonomous driving system's environmental understanding, path planning, and control execution.

[0003] Current autonomous driving systems primarily rely on information sources such as onboard cameras, millimeter-wave radar, lidar, and high-precision maps to complete environmental perception and path planning. Under normal circumstances, these solutions can adequately meet the operational needs of autonomous vehicles. However, in scenarios involving sudden incidents on highways, it is difficult to obtain timely and complete information on the spatial distribution, evolution trend, and recovery status of the incident area ahead. This results in insufficient forward-looking information when the vehicle makes speed adjustments, lane selections, or path replanning, easily leading to problems such as delayed response, overly conservative planning, or unstable planning decisions. Currently, decision support methods for autonomous vehicles in abnormal road environments can be mainly categorized as follows: (1) Path adjustment method based on vehicle local perception.

[0004] These methods primarily identify obstacles or abnormal areas based on the vehicle's forward perception, and then perform deceleration, following, obstacle avoidance, or lane change control within a localized area. While these methods respond quickly to nearby localized obstacles, they are easily affected by factors such as obstruction from other vehicles, limited visibility, fog, and road gradients in emergency situations on highways. This makes it difficult to obtain a complete picture of the event further ahead in a timely manner, thus hindering the formation of forward-looking decisions for several seconds to tens of seconds into the future.

[0005] (2) Route decision-making method based on fixed roadside facilities or road event broadcasting.

[0006] These methods send warnings of upcoming anomalies to vehicles via roadside cameras, edge computing nodes, traffic event broadcasts, or cloud platforms to assist vehicles in replanning their routes. While these methods can compensate for the limited line-of-sight of onboard perception to some extent, they rely on fixed infrastructure deployment, have limited coverage, and primarily broadcast event results, lacking continuous depiction of the event's evolution and detailed descriptions of specific action permissions.

[0007] (3) Video transmission method based on UAV inspection.

[0008] This type of method utilizes drones to observe target areas ahead, using images, videos, or recognition results to assist human decision-making or autonomous driving control. While this method can expand the vehicle's perception range and improve information acquisition capabilities in obstructed scenarios, existing solutions mostly focus on drone image acquisition, target recognition, and result transmission. They lack a unified predictive model for the evolution of highway emergencies over future periods and also lack mechanisms to directly translate drone observation results into autonomous driving action boundaries, lane time-varying open windows, and executable path constraints.

[0009] Furthermore, existing technologies generally assume that external observation information is always reliable in terms of communication quality, latency, coverage of key areas, perception conflicts, and remaining energy, and rarely consider the changes in UAV observation quality over time. When external observations suffer from delays, packet loss, low confidence in key areas, or air-to-ground perception conflicts, continuing to execute the normal aggressive path can easily lead to unnecessary high-risk actions.

[0010] Therefore, it can be seen that the existing technology has the following three main shortcomings: First, existing technologies focus primarily on identifying the current state of emergencies and issuing anomaly alerts, lacking continuous modeling of future evolution trends such as the extent of road occupancy, queue propagation, formation of temporary lanes, and progress of emergency response. This makes it difficult to support forward-looking path decision-making for autonomous vehicles in emergency scenarios on highways.

[0011] Second, existing external perception information is usually provided to vehicles in the form of event prompts, obstacle locations, passable areas, or risk scores. There is a lack of mechanisms to further transform external observation results into lane restoration windows, action permission states, and candidate path selection conditions, making it difficult for autonomous driving systems to directly utilize external observation information to form executable path constraints.

[0012] Third, existing technologies often assume that external observation information, such as that from drones, is always reliable, and rarely consider the impact of factors such as communication latency, packet loss, key area obstruction, air-to-ground perception conflicts, and insufficient drone battery power on path decision-making. When the reliability of external observations decreases, if vehicles still perform aggressive actions such as continuous lane changes, lane borrowing, high lateral displacement, or high dynamic acceleration and deceleration, it may further amplify road traffic risks.

[0013] Therefore, there is an urgent need to propose a UAV cooperative path decision-making method for highway emergencies, so that the continuous observation results of UAVs can not only be used for event identification and information prompts, but also for event evolution prediction, lane restoration window calculation, temporary passage permission generation, candidate path screening, action intensity gating and graded degradation control of UAV observation reliability, thereby supporting autonomous vehicles to make forward-looking, executable, explainable and degradation-friendly safe path decisions in emergency scenarios. Summary of the Invention

[0014] In view of the shortcomings and deficiencies of the existing technology, the purpose of this invention is to provide a UAV cooperative path decision-making method for highway emergencies. This method enables autonomous vehicles to transform from a passive response mode that relies on onboard local perception and external event prompts in highway emergency scenarios to an active safety decision-making mode based on continuous UAV observation, event evolution prediction, lane restoration window, action permission constraints and observation reliability gating, thereby improving the forward-looking nature, execution stability and operational safety of vehicle path decision-making in complex and abnormal road environments.

[0015] To achieve the above objectives, the present invention adopts the following technical solution: A method for collaborative path decision-making by unmanned aerial vehicles (UAVs) in response to emergencies on highways includes the following steps: Step 1. Acquire multi-source data from both air and ground sources, and determine whether to initiate UAV collaborative observation; if so, proceed to Step 2. Step 2. The UAV continuously observes the target observation area ahead and converts the observation data into structured variables; Step 3. Based on the historical observation sequence of UAVs, the status of the main vehicle, and the road structure characteristics of the event impact area, predict the evolution trend of the sudden event within the future time window through a short-time state-space prediction model; Step 4. Calculate the lane opening probability and lane resumption window based on the emergency evolution prediction results; Step 5. Taking into account lane opening probability, lane resumption window, road rules, action risk, and main vehicle task objective, generate a temporary access permission map with lane-action-time domain association; Step 6. Generate the original candidate route set and filter by action permission and entry time according to the temporary access permission map and lane restoration window to generate candidate routes after the initial screening. If the number of candidate routes meets the requirements, proceed to step 7. Step 7. Assess the reliability of UAV observations. If it meets the requirements, proceed to step 8. Step 8. Calculate the path action intensity level for each candidate path, and determine the upper limit of action intensity based on the UAV observation reliability index. Perform a second screening of the candidate path set based on the upper limit of action intensity. If the number of candidate paths after the second screening meets the requirements, proceed to step 9. Step 9. Select the optimal route by combining travel time, route risk, comfort, task deviation, and replanning jitter.

[0016] As a preferred embodiment of the present invention, the multi-source air-ground data in step 1 includes vehicle status information, road structure information, UAV continuous observation data, vehicle-road cooperative messages, roadside or cloud event records, and communication status information; wherein, the vehicle status information includes the vehicle's longitudinal position, lateral offset, speed, acceleration, jerk, lane number, and vehicle dynamics constraints; the road structure information includes the number of lanes, lane width, emergency lane location, median strip location, ramp entrance / exit location, speed limit information, no-stopping / no-entry rules, emergency lane borrowing rules, and road topology attributes of the event area location; In step 1, based on the constructed emergency event triggering index, it is determined whether to initiate UAV collaborative observation. When the emergency event triggering index is greater than or equal to the emergency event triggering threshold, it is determined that there is a risk of an emergency event ahead, and UAV collaborative observation is initiated. The emergency event triggering index is obtained by weighted summation of the abnormal index of sudden deceleration of traffic flow ahead, the abnormal index of queue overflow, the abnormal index of vehicle-road cooperative event messages, and the abnormal index of vehicle-mounted visual detection.

[0017] As a preferred embodiment of the present invention, in step 2, the original image information or video of the UAV is transformed into event lane length, number of lanes occupied, queue length, queue tail propagation speed, effective width of temporary channel, debris or event boundary expansion, emergency response progress and UAV observation confidence of key areas through target detection, semantic segmentation, lane line recognition, vehicle tracking, map registration and spatiotemporal synchronization.

[0018] As a preferred embodiment of the present invention, in step 3, the continuous observation results of the UAV are constructed into an event evolution state sequence. Combining the current event evolution state, the event state change trend, the main vehicle state, and the road structure characteristics of the event impact area, a short-time state space prediction model is used to predict the event occupancy range, queue propagation trend, temporary channel formation, and emergency response progress within the future time window.

[0019] As a preferred embodiment of the present invention, in step 4, the predicted results of the future event evolution state obtained in step 3 are first mapped to lane-level effective channel width, traffic continuity, queuing encroachment degree and obstacle occupancy intensity. Combined with the emergency response progress, the probability of lane opening at future time is calculated. Then, based on vehicle width, safety distance and opening probability threshold, it is determined whether each lane can be safely entered in the future time slice, and a lane restoration window is generated.

[0020] As a preferred embodiment of the present invention, step 5 integrates the future opening probability of the lane, consistency of road rules, consistency between the action and the main vehicle's task objective, consistency between the action execution time and the lane reopening window, and the intensity of the risk or conflict caused by the action to calculate an access score. Based on the access score, the access status of the temporary access permission map is set to permitted, restricted, or prohibited. The intensity of the risk or conflict caused by the action is expressed as follows: ; In the formula, lane ,action and time slices The corresponding risk or conflict intensity; Indicates the execution of an action The collision time between the subsequent object and the potential conflicting object; It represents a positive number and is used to avoid the denominator being zero; Indicates the first Obstacle occupancy intensity of each lane at future time; Indicates the first The extent of queue encroachment on each lane in future timeframes; This indicates the risk factor of the action itself; to This indicates the weight of each risk item.

[0021] As a preferred embodiment of the present invention, in step 6, an original candidate path set is first generated based on the current state of the master vehicle, the road topology, the candidate target lanes, the master vehicle's target speed, and the path planning task; wherein, the candidate target lanes refer to the lanes that the master vehicle continues to maintain or is preparing to enter in the current prediction time domain; then, the action permission filter and entry time filter are performed using the temporary access permission map and the lane restoration access window to eliminate paths that contain prohibited actions or whose entry times do not meet the restoration window requirements; if the candidate path set after the initial screening is an empty set or the candidate path set is not empty but the number is lower than the minimum candidate path number threshold, then hierarchical degradation control is triggered.

[0022] As a preferred embodiment of the present invention, in step 7, the high confidence coverage of key areas, communication latency, packet loss rate, air-to-ground perception conflict level and UAV remaining battery power are considered to calculate the UAV observation reliability index; if the UAV observation reliability is lower than the minimum reliability threshold, or the air-to-ground perception conflict level exceeds the air-to-ground perception conflict threshold, then graded degradation control is triggered.

[0023] As a preferred embodiment of the present invention, the path motion intensity level in step 8 The expression is: ; in, This indicates the number of left and right lane changes in the path. Indicates whether the path includes a conservative detour. Indicates the peak acceleration along the path. Indicates the peak jerk along the path. Indicates the maximum lateral displacement of the path. to These are the weighting coefficients; Action intensity limit expression:

[0024] in, This indicates the lower limit of the permissible intensity of the action. This indicates the maximum allowed intensity of the action. Indicates the gating shape parameters, Indicates time The reliability indicators of UAV observation; In step 8, if the final candidate path set after secondary screening is empty, or the number of final candidate paths is lower than the minimum candidate path number threshold, then hierarchical degradation control is triggered.

[0025] As a preferred embodiment of the present invention, in step 9, the comprehensive cost is calculated for each of the final candidate paths after the secondary screening, and the candidate path with the smallest comprehensive cost is selected as the target execution path, and the target speed, target acceleration, target lateral trajectory and action sequence are output; wherein, the comprehensive cost is obtained by weighted summation of travel time cost, path risk cost, comfort cost, task deviation cost and replanning jitter cost.

[0026] As a preferred embodiment of the present invention, the graded degradation control is divided into three levels. When the reliability of UAV observation is only lower than the normal working threshold, or when the candidate path set after initial screening is not empty but the number is lower than the minimum candidate path number threshold, the first-level degradation mode is executed. The system restricts the action set, allowing only lane keeping, deceleration and holding, and one left lane change, one right lane change, or one conservative lane change. Continuous left / right lane changes and continuous conservative lane changes are prohibited. Within the conservative action set, the path is selected according to the comprehensive cost function. When the reliability of UAV observation is lower than the minimum working threshold but not exceeding the set amount, the degree of air-to-ground perception conflict exceeds the air-to-ground perception conflict threshold but not exceeding the set amount, or the number of final candidate paths after secondary screening is lower than the set threshold, the second-level degradation mode is executed, and the system only allows lane keeping and deceleration to be maintained; if the safe distance ahead is lower than the set threshold, it switches to stop moving forward or minimum risk parking. If the initial candidate path set is empty, the final candidate path set after secondary screening is empty, the UAV observation reliability is lower than the minimum working threshold and exceeds the set amount, or the air-to-ground perception conflict level exceeds the air-to-ground perception conflict threshold and exceeds the set amount, the system will execute a stop-movement, minimum-risk stop, or manual takeover request, and will no longer output paths that include lateral lane changes or detours.

[0027] The present invention also provides a path decision system for implementing the above-described method. The path decision system includes an air-ground multi-source data acquisition module, a UAV observation data structuring module, a sudden event evolution prediction module, a lane restoration and traffic window calculation module, a temporary traffic permission map generation module, a candidate path generation and permission filtering module, a UAV observation reliability assessment and gating module, a path optimization and control execution module, a graded degradation control module, and a rolling update and closed-loop decision module.

[0028] Advantages and beneficial effects of the present invention: (1) The present invention adopts the UAV working mechanism of “normal low-frequency cruise / standby monitoring + high-frequency collaborative observation after event triggering”, which does not require the UAV to fly continuously in a high-power mode. By forming an emergency event triggering index through vehicle vision, V2X messages, roadside equipment, traffic flow anomalies and cloud platform, the observation frequency of the UAV can be quickly increased when anomalies occur, taking into account response speed, observation continuity and energy utilization efficiency.

[0029] (2) This invention proposes a method for predicting the evolution of emergencies based on continuous observation by UAVs. This method constructs an event state sequence by considering the event's lane occupancy length, number of lanes occupied, queue length, queue tail propagation speed, effective width of temporary passages, event boundary expansion, and emergency response progress. Preferably, a short-time state-space prediction model with historical trend terms, vehicle status terms, and road structure terms of the event's impact area is used for prediction. Compared to methods that only identify the current obstacle location, this prediction method can utilize continuous observation differentials to suppress single-frame noise and output lane opening trends for multiple future time slices. It has the advantages of strong foresight, low computational cost, and ease of deployment on vehicles or at the edge.

[0030] (3) This invention proposes a method for calculating lane restoration passage windows, which further maps UAV observations and event evolution prediction results into lane-level effective channel width, traffic continuity, queue encroachment degree, and obstacle occupancy intensity, and combines the main vehicle width, lateral safety distance, and lane opening probability threshold to generate an accessible time window. This method extends the "whether passage is possible" from a static judgment to a temporal constraint of "when it is safe to enter," which can reduce premature lane changes, accidental entry into unrestored lanes, and frequent replanning problems.

[0031] (4) This invention proposes a temporary access permission map generation method based on lane-action-time domain association, which uniformly encodes lane restoration windows, highway traffic rules, emergency response status, candidate lane sets, and action risks into three types of action permissions: permitted, restricted, and prohibited. Through this permission map, UAV observation results are no longer merely used as alarm prompts or risk scores, but are directly transformed into candidate path selection conditions for autonomous vehicles, improving the interpretability and executability of the connection between external perception information and vehicle execution decisions.

[0032] (5) This invention proposes a hierarchical degradation control and rolling update mechanism. When there is no solution in the permission filtering, insufficient number of candidate paths, insufficient reliability, or excessive air-ground perception conflict, it does not directly execute aggressive replanning. Instead, it outputs requests for deceleration and maintenance, lane keeping, conservative lane change, stop, minimum risk parking, or manual takeover according to the severity of the anomaly: mild, moderate, and severe. At the same time, the event prediction results are continuously refreshed through rolling updates and decision lag mechanisms, and frequent lateral switching is reduced, thereby improving the stability, safety, and engineering feasibility of path decision-making.

[0033] (6) In view of the problem that existing external observation information is mostly limited to event prompts, obstacle recognition or risk scoring, and is difficult to directly constrain the action selection of autonomous vehicles, this invention proposes a method for generating lane restoration access windows and temporary access permission maps. This method calculates the restoration access window of each target lane in the future time window based on the event evolution prediction results, and generates a temporary access permission map with lane-action-time domain association by combining highway road rules, event handling status and master vehicle task objectives. It assigns allowed, restricted or prohibited states to actions such as lane keeping, lane changing, deceleration and keeping, conservative lane borrowing, stopping and minimum risk parking, thereby transforming UAV observation results into action boundaries and path constraints that can be directly used for candidate path screening.

[0034] (7) In view of the problem that existing technologies ignore the dynamic changes in the reliability of UAV observation, which may lead to vehicles still performing aggressive path actions when external observation is unreliable, this invention proposes a UAV observation reliability gating and graded degradation control method. This method integrates the coverage of key areas, communication delay, packet loss rate, air-to-ground perception conflict degree and UAV remaining power to construct a UAV observation reliability index, and dynamically limits the intensity of candidate path actions according to the index. When the UAV observation reliability is high, medium intensity paths such as single left / right lane change or conservative lane borrowing can be retained. When the UAV observation reliability is lower than the threshold, the air-to-ground perception conflict exceeds the threshold, or there is no solution for the candidate path, the system automatically restricts continuous lane change, conservative lane borrowing, high acceleration, high jerk and large lateral displacement paths, and enters graded degradation control modes such as deceleration maintenance, prohibition of continuous lateral actions, only allowing one conservative lane change, minimum risk stopping or manual takeover request, so as to avoid vehicles still performing high-risk actions when external observation is unreliable, and improve the safety, stability and executability of path decision-making in highway emergency scenarios.

[0035] (8) This invention enables autonomous vehicles to change from a passive response mode that relies on onboard local perception and external event prompts in the event of a sudden incident on the highway to an active safety decision-making mode based on continuous observation by UAVs, event evolution prediction, lane restoration window, action permission constraints and observation reliability gating, thereby improving the forward-looking nature of the vehicle's path decision, execution stability and operational safety in complex and abnormal road environments. Attached Figure Description

[0036] Figure 1 This invention provides a flowchart of a UAV collaborative path decision-making method for highway emergencies; Figure 2 This invention provides a block diagram of a drone collaborative path decision-making system for highway emergencies. Detailed Implementation

[0037] To enable those skilled in the art to better understand the technical solutions and advantages of the present invention, the present application will be described in detail below with reference to the accompanying drawings, but this is not intended to limit the scope of protection of the present invention.

[0038] like Figure 1 As shown in the figure, this embodiment provides a UAV cooperative path decision-making method for highway emergencies, which includes the following steps: Step 1. Acquire multi-source air and ground data and establish a unified spatiotemporal reference: In highway emergency scenarios, the system first acquires multi-source data through vehicle-mounted sensors, vehicle controllers, positioning and navigation modules, vehicle-to-everything (V2X) terminals, high-precision map modules, roadside facilities, cloud-based event platforms, and UAV observation platforms. The UAV operates in two modes: when no emergency is triggered, it performs low-frequency patrols, remains stationary, or receives low-frequency data transmitted back from the platform; when the emergency trigger index reaches a threshold, it switches to high-frequency continuous observation of the affected area. Therefore, the subsequent "historical observation sequence" can come from continuous high-frequency observations after the trigger, or initially from low-frequency observations before the trigger, roadside / platform records, or a combination of the two most recent high-frequency observation points.

[0039] Furthermore, in this embodiment, the air-ground multi-source data includes master vehicle status information, road structure information, UAV continuous observation data, vehicle-road cooperative messages, roadside or cloud event records, and communication status information, etc.; wherein, the master vehicle refers to an autonomous vehicle equipped with the path decision system of this invention, which needs to perform path planning, speed adjustment and lane selection according to the status of sudden events on the highway ahead, and can also be understood as the vehicle currently being controlled or assisted in decision-making.

[0040] Furthermore, in this embodiment, the vehicle state information includes the vehicle's longitudinal position, lateral offset, speed, acceleration, jerk, lane number, target lane, vehicle width, vehicle dynamics constraints, current path planning state, and autonomous driving system operating state. Considering that highway path decision-making mainly includes longitudinal speed adjustment and lateral lane selection, this invention preferably uses the Frenet coordinate system under the road reference line to describe the vehicle state. Vehicle State Vector It can be represented as:

[0041] in, Indicates the longitudinal position of the main vehicle along the road reference line. This indicates the lateral offset of the main vehicle relative to the road reference line. Indicates the speed of the main vehicle. Indicates the longitudinal acceleration of the main vehicle. Indicates the longitudinal acceleration of the main vehicle. Indicates the lane number where the main vehicle is located. The road curvature at time t represents the vehicle's position (a vehicle dynamics constraint used to describe the current position of the main vehicle). This indicates the current speed limit for that section of road.

[0042] Road structure information includes the number of lanes, lane width, emergency lane location, central divider location, ramp entrance and exit locations, curve and slope combination information, speed limit information, no-stopping and no-driving rules, emergency lane borrowing rules, and road topology attributes of the location of the event area.

[0043] Furthermore, to determine whether to initiate UAV collaborative observation, this invention constructs a sudden event triggering index:

[0044] In the formula, Indicates time The sudden event trigger index; This indicates an abnormal index of sudden deceleration of traffic flow ahead; This indicates the abnormal index of queue backflow; This indicates the anomaly index of vehicle-road cooperative event messages; Indicates the abnormality index of vehicle-mounted visual detection; , , and This represents the weighting coefficient corresponding to each anomaly index, used to characterize the importance of different information sources in the judgment of the triggering of emergencies.

[0045] The weighting coefficients satisfy:

[0046]

[0047] In the formula, Indicates the first The weighting coefficients corresponding to the anomaly index; This represents the weight index, with a value range of [value range missing]. to The summation constraint is used to ensure that the weights of the four types of outlier indices are normalized.

[0048] The abnormal index of sudden deceleration of traffic flow ahead can be expressed as:

[0049] In the formula, Indicates time The index of sudden deceleration of traffic flow ahead; Indicates time The average longitudinal acceleration of the vehicle or traffic flow ahead; Indicates the preset sudden deceleration threshold; This represents a truncation function, used to restrict the calculation result to a specific range. to Within the interval. When When the speed is significantly lower than the preset sudden deceleration threshold, the system determines that there is a risk of abnormal deceleration in the traffic flow ahead.

[0050] The queuing backflow anomaly index can be expressed as:

[0051] In the formula, Indicates time The queuing backflow anomaly index; This indicates the current queue length. Indicates the queue length at the previous observation time; Indicates the time interval between two consecutive observations; This indicates the preset maximum queue growth rate; This indicates the maximum allowed queue growth length within an observation period. When the queue length growth approaches or exceeds the allowed range, the system determines that there is a risk of queue overflow ahead.

[0052] The anomaly index of vehicle-road cooperative event messages can be expressed as:

[0053] In the formula, Indicates time The abnormality index of vehicle-road cooperative event messages; Indicates the credibility of vehicle-road cooperative event messages; This indicates the vehicle-to-infrastructure (V2I) event message reception flag. When a message about a sudden event is received ahead... Values Otherwise, the value is When the message has high credibility and the message reception flag is... At that time, the system determines that there is an emergency ahead based on the vehicle-road cooperative information.

[0054] The vehicle-mounted visual detection anomaly index can be expressed as:

[0055] In the formula, Indicates time The abnormality index of vehicle-mounted visual detection; This indicates the probability that the vehicle's vision model recognizes an obstacle or scattered object. Indicates the probability of lane anomalies; Indicates the probability of smoke, fog, or limited visibility; This indicates the probability of an abnormally stopped or abnormally slow-moving vehicle ahead; the maximum value function is used to extract the most significant visual anomaly information.

[0056] The system activates the UAV collaborative observation module when the following equation is satisfied:

[0057] In the formula, This indicates the threshold for triggering a sudden event. When the sudden event trigger index... Greater than or equal to the threshold When the system determines that there is a risk of an emergency ahead, it switches the UAV from low-frequency patrol / standby observation to high-frequency collaborative observation. If the UAV is already on the inspection route, it increases the observation frequency of the target area and adjusts the viewing angle. Thus, the historical observation sequence required for subsequent prediction can be obtained by continuously connecting the low-frequency observation before the trigger with the high-frequency observation after the trigger. When historical data is insufficient, the system uses the differential trend of the two most recent observation times or the prior of historical samples for short-term prediction.

[0058] To ensure consistency in subsequent calculations, the system unifies vehicle-mounted data, roadside data, UAV data, and platform data to the same time and road coordinate reference. The time reference can use a second-level or sub-second UTC timestamp, and the road coordinate reference can use the Frenet coordinate system or a high-precision map lane coordinate system. For target positions in UAV image coordinates, they can be converted to positions in the road coordinate system through camera calibration, homography transformation, georegistration, or map matching methods.

[0059] Step 2. Structured processing of continuous UAV observation data: The drone continuously observes the target observation area ahead according to a preset observation cycle. The target observation area is the entire road range that the drone needs to cover, including at least the accident core area, the tail end of the queue, the temporary passage area, the lane boundary area, the emergency response area, and the candidate lane area that the main vehicle may enter. The continuous observation sequence of the drone can be represented as:

[0060] in, Indicates the length of historical observations used for modeling. Indicates the drone at a certain time Structured observation vectors.

[0061] The raw observation data from UAVs includes video frames, images, target detection results, semantic segmentation results, lane recognition results, target tracking results, UAV pose information, shooting angle, flight altitude, communication status, and battery status. Through target detection, semantic segmentation, lane recognition, vehicle tracking, map registration, and spatiotemporal synchronization, the raw image information or video is transformed into structured event observations.

[0062] Single-moment UAV observation vector Defined as:

[0063] in, Indicates the length of the road occupied by the event. Indicates the number of lanes occupied. Indicates the queue length. Indicates the propagation speed at the tail of the queue. Indicates the effective width of the temporary passage. Indicates the extent of the expansion of the debris or event boundary. Indicates the progress of emergency response. This indicates the confidence level of the UAV's observation of a key area. The key area is the sub-area in the target observation area that has the greatest impact on vehicle traffic decisions (the core road area directly related to the handling of highway emergencies and vehicle traffic decisions). It can be weighted and divided according to the accident location, the tail of the queue, the candidate route entry point, and the emergency response area. It includes the core impact area of ​​the accident, the tail of the queue area (the road section affected by the queued vehicles), the temporary passage area (the range of the temporary passage), the lane boundary area, the candidate route entry area, and the emergency response area (emergency control area).

[0064] Specifically, in this embodiment, the above-mentioned observations are obtained in the following ways: For drones at any time After registering the obtained emergency area to the road coordinate system, the road occupancy length due to the emergency can be expressed as:

[0065] In the formula, Indicates time The length of the road occupied by the event; Indicates time The area of ​​the emergency identified by the drone; Represents any point in the event region; Point The longitudinal coordinates along the road reference line; the difference between the maximum and minimum values ​​is used to characterize the length of the emergency zone occupied in the longitudinal direction of the road.

[0066] The number of lanes occupying a road can be expressed as:

[0067] In the formula, Indicates time The number of lanes occupying the road; This indicates the total number of lanes on the current road. Indicates the lane number; Indicates the first One lane; Indicates the first [item] within the vertical range of the event's impact. Single lane area; Represents the area function of a region; Indicates the event area and the first The overlapping area of ​​the lanes; This indicates the threshold for determining lane occupancy. Indicator functions are used to determine the expression when a certain condition is met. Otherwise take When the overlap between the event area and a certain lane reaches or exceeds [a certain percentage]... At that time, the system determines that the lane is occupied by an event.

[0068] Queue length can be expressed as:

[0069] In the formula, Indicates time Queue length; Indicates the longitudinal position of the end of the queue; This indicates the position of the head of the queue or the upstream boundary of the event; the difference between the two is used to characterize the length of the queue of vehicles in the longitudinal direction of the road.

[0070] The propagation speed at the tail of the queue can be expressed as:

[0071] In the formula, Indicates time The propagation speed of the queue tail; Indicates the current position at the end of the queue; Indicates the position of the tail of the queue at the previous observation time; This represents the time interval between adjacent observation times; this formula is used to calculate the change in the position of the tail of the queue per unit time.

[0072] The effective width of a temporary passage can be expressed as:

[0073] In the formula, Indicates time The effective width of the temporary passage; This indicates the longitudinal road section affected by the incident. This section is preferably determined by superimposing the minimum and maximum longitudinal coordinates of the incident area with a safety buffer distance (determined by the upstream and downstream boundaries of the incident and the safety buffer distance, which can cover the core area of ​​the accident, the area affected by debris, and the temporary control area). When there is a queue tail or emergency response area, the upstream boundary of the section can be extended to the queue tail position, and the downstream boundary can be extended to the boundary affected by the response vehicles or debris. Represents the longitudinal coordinates of the road; Indicates time Longitudinal position The horizontal coordinates of the right boundary of the passable area; Indicates time Longitudinal position The horizontal coordinates of the left boundary of the passable area; the minimum value function is used to ensure that the temporary passage meets the passage width requirements throughout the entire event-affected area (ensuring that the narrowest position also meets the passage conditions).

[0074] The extent of the debris or event boundary expansion can be expressed as:

[0075] In the formula, Indicates time The event boundary expansion amount; This represents the Hausdorff distance function, used to measure the maximum offset distance between two boundary sets; Indicates the boundary of the event region at the current moment; It represents the boundary of the event area at the previous observation time; this formula is used to characterize the degree of dispersion of debris, expansion of the accident area, or change in the temporary control area.

[0076] The progress of emergency response can be expressed as:

[0077] In the formula, Indicates time Emergency response progress; Indicates the extent to which accident vehicles or obstructed vehicles have been removed; This indicates the arrival status of rescue vehicles, road clearing vehicles, or emergency personnel. Indicates the completion status of the deployment of traffic cones, warning signs, or temporary control facilities; Indicates the degree of lane restoration; to This represents the weighting coefficient of each sub-indicator of the progress of the disposal.

[0078] The weighting coefficients satisfy:

[0079]

[0080] In the formula, Indicates the first The weights of each sub-indicator of emergency response progress are used to ensure that the emergency response progress is obtained by weighting multiple normalized sub-indicators.

[0081] The confidence level of UAV observations of key areas can be expressed as:

[0082] In the formula, Indicates time Confidence level of observations in key areas; Indicates the image sharpness of the drone; This represents the average confidence level of the target detection. Indicates coverage rate in key areas; Indicates the occlusion percentage of key areas; Indicates the degree of air-to-ground perception conflict; Used to indicate the degree of unobstructedness; Used to indicate the degree of consistency in air-to-ground perception; to This represents the weighting coefficient corresponding to each observation quality indicator.

[0083] The weighting coefficients satisfy:

[0084]

[0085] In the formula, Indicates the first The weights of the observation quality indicators; this constraint is used to ensure that the confidence level of observations in key areas is obtained by weighting five normalized observation quality indicators.

[0086] Through the above processing, the UAV observation results are transformed from image or video information into structured state variables that can be directly used for event evolution prediction and path decision-making.

[0087] Step 3. Prediction of the evolution of sudden events based on continuous observation by UAVs: To enable autonomous vehicles not only to recognize the current state of an emergency but also to predict how the event will develop over a future period, this invention constructs a sequence of event evolution states from continuous observations by unmanned aerial vehicles (UAVs). Event Evolution State Vector Defined as:

[0088] By combining historical observation trends, vehicle status, and road structure information in the event impact area, the event status in the future prediction time domain is predicted:

[0089] In the formula, Indicates future time Predicted event states; Indicates the current time The state of event evolution; It indicates the trend of event status changes at the current moment; Indicates the current status of the main vehicle; This indicates the road structure characteristics of the event's affected area or the area into which candidate paths enter; This represents a prediction function for the evolution of emergencies, used to map the current event evolution state, historical trends, vehicle status, and road structure characteristics to the future event state; This indicates the future prediction step size.

[0090] The predicted step size satisfies:

[0091] In the formula, Indicates the length of the prediction time domain; This indicates that the system predicts the future number of... The time slice to the first The event state of a time slice.

[0092] The mapping relationship of the prediction function can be expressed as:

[0093] In the formula, This represents the event evolution prediction mapping function; the four variables in the left-hand set represent the model input; This represents the model output; the mapping relationship is used to explain how the system obtains predictions of future event states from the current state and structural constraints.

[0094] The historical observation state change trend can be obtained by weighted sum of state differences from the most recent multiple observation periods:

[0095] In the formula, It indicates the trend of event status changes at the current moment; This indicates the length of historical observations used to calculate trends of change; Indicates the historical observation difference sequence number; Indicates the first The time weights of each historical difference term; Indicates the state of a more recent historical event; Indicates the state of an older historical event; This formula represents the time interval between adjacent observation times; it is used to calculate the most recent observation time. The adjacent difference results are weighted and summed to reduce the impact of single-frame observation noise on trend judgment.

[0096] The weighting coefficients satisfy:

[0097]

[0098] In the formula, Indicates the first The weights of each historical difference term; this constraint is used to ensure that the sum of the weights in the historical difference trend calculation is [value missing]. .

[0099] When the historical observation length is two observation times, the above equation degenerates into:

[0100] In the formula, Indicates the current event status; Indicates the event state at the previous observation time; This represents the time interval between the current moment and the previous observation moment; this formula is used to represent the average rate of change of the event state when only two observation moments are used.

[0101] The road structure characteristics of the event-affected area or candidate path entry area can be represented as follows:

[0102] In the formula, Indicates the number of road lanes corresponding to the event's affected area or the candidate path entry area; Indicates lane width; Indicates the availability of the emergency lane; Indicates the speed limit in the area affected by the event; This represents the road curvature at the event center or candidate entry point (used to describe the road structure constraints of the event's influence area or candidate entry point). Indicates the longitudinal slope of the area affected by the event; Indicates the type of emergency; Indicates the lane where the emergency occurred; Indicates the vertical location of the center of the emergency; , , and These indicate whether ramps, bridges, tunnels, and curves exist within the event's affected area; superscript... This represents the transpose of a vector.

[0103] In this embodiment, a short-time state-space prediction model with exogenous input is preferably used for event evolution prediction. This model is based on the current event evolution state, uses historical differential trends to represent queue propagation, debris expansion, and changes in handling progress, and uses the main vehicle state and road structure of the event impact area as exogenous constraints. Model parameters are obtained through regularized training. Its advantages are: first, it can utilize continuous observation from UAVs to form time trends, avoiding the direct determination of the path by single-frame recognition results; second, the model structure is clear and computationally intensive, suitable for continuous operation in vehicle-mounted computing units or roadside edge computing units; and third, the output results still retain interpretable variables such as lane occupancy length, queue length, temporary lane width, and handling progress, facilitating subsequent recovery window and permission map calculations.

[0104] Specifically, the short-time event evolution prediction model (the trained short-time state-space prediction model) can be represented as:

[0105] In the formula, Indicates future time Predicted event states; Indicates the current state of the event's evolution; Indicates the trend of event status changes; Represents the main vehicle state vector; This indicates the road structure characteristics of the event's affected area or the area into which candidate paths enter; This represents the coefficient matrix that preserves the current state. Represents the influence matrix of historical trends; This represents the main vehicle state coupling matrix; This represents the coupling matrix of road structure and event structure features.

[0106] The above matrix parameters can be obtained through training with historical samples, and their optimization objective is:

[0107] In the formula, Represented by matrix , , and To optimize variables; Indicates the training sample number; Indicates the total number of training samples; Indicates the first The true or labeled event state of a sample at a future time; , , and They represent the first The current event status, status change trend, vehicle status, and road structure characteristics corresponding to each sample; This represents the squared L2 norm, used to measure prediction error. Represents the regularization coefficient; This represents the squared Frobenius norm of a matrix, used to suppress excessively large matrix parameters.

[0108] The model input can be represented as:

[0109] In the formula, Indicates time The model input set; to It represents the sequence of event evolution states from the initial observation time in history to the current time; This indicates the length of historical observations; the model input is used to support predictions of future event states.

[0110] The model output can be represented as:

[0111] In the formula, Indicates time The model output set; , and These represent the event states in different predicted time slices in the future; This indicates the predicted time domain length; the output is used to represent the evolution of events over multiple future time slices.

[0112] The future event state vector can be represented as:

[0113] In the formula, Indicates future time The event state vector; Indicates the length of road occupancy for future events; Indicates the number of lanes that will be occupied in the future; Indicates the future queue length; Indicates the future propagation speed of the queue tail; Indicates the effective width of the temporary passageway in the future; Indicates the extent to which the boundary of future events expands; This indicates the progress of future emergency response.

[0114] This step allows us to obtain the lane occupancy range, queue length, queue tail position, temporary lane width, boundary expansion trend, and handling progress trend of events within the future prediction time domain, providing a basis for subsequent lane restoration window calculations.

[0115] Step 4. Calculate the lane opening probability and the reopening window: After obtaining the future event state prediction results, this invention further maps the future event evolution state vector output in step 3 into lane-level effective channel width, traffic continuity, queuing encroachment degree and obstacle occupancy intensity, and calculates the opening probability of each target lane in the future time window accordingly.

[0116] No. The probability that a lane will be open at a future time can be expressed as:

[0117] In the formula, Indicates the first Lane in the future The probability of opening; Indicates the lane number; This represents the Sigmoid activation function; Indicates the bias term; to This indicates the weighting coefficients corresponding to the factors influencing the opening of each lane; Indicates the predicted effective channel width; Indicates predicted traffic continuity; This indicates the predicted degree of queue encroachment; Indicates the predicted obstacle occupancy intensity; This indicates the predicted progress of emergency response. Positive terms increase the probability of lane opening, while negative terms decrease the probability of lane opening.

[0118] The Sigmoid function can be represented as:

[0119] In the formula, This represents the output value of the Sigmoid function; This represents the overall score entered. Represents the natural constant; this function is used to map any real number to... to Interval.

[0120] The effective lane width is obtained by mapping the predicted event state:

[0121] In the formula, Indicates the first The effective lane width of each lane at future time; This represents the effective channel width mapping function; Indicates future time The event state; this mapping is used to convert the event prediction results into lane-level width parameters.

[0122] Lane continuity is obtained by mapping the predicted event state:

[0123] In the formula, Indicates the first The continuity of traffic flow for each lane in future timeframes; This represents the traffic continuity mapping function; the mapping is used to determine whether the corresponding lane has continuous traffic conditions in future time slices.

[0124] The degree of queue encroachment is obtained by mapping the predicted event state:

[0125] In the formula, Indicates the first The extent of queue encroachment on each lane in future timeframes; This represents a mapping function for the degree of queue encroachment; this mapping is used to convert predicted queue length and queue spatial distribution into lane-level queue impact indicators.

[0126] Obstacle occupancy intensity is obtained by mapping the predicted event state:

[0127] In the formula, Indicates the first Obstacle occupancy intensity of each lane at future time; This represents the obstacle occupancy intensity mapping function; this mapping is used to convert the event's lane occupancy range and boundary expansion trend into a lane-level obstacle impact index.

[0128] Furthermore, in this embodiment, the predicted lane-level effective channel width, traffic continuity, queue encroachment degree, and obstacle occupancy intensity are respectively represented as follows: (1) The effective lane width at the lane level can be obtained from the lateral overlap relationship of the predicted event area, the predicted queuing area, and the lane area, and is preferably expressed as:

[0129] In the formula, Indicates the first Lane in the future The effective channel width corresponds to the aforementioned mapping result. ; This represents the vertical interval affected by future events, which is determined by the future event states predicted in step 3. With road structural features To be determined jointly; and They represent the first The lanes are in the longitudinal position The right and left boundaries of the location; Indicates the region of predicted events Predicting queuing areas The lateral width of the lane is occupied by the combined effect of the debris area and the emergency response area.

[0130] (2) Lane traffic continuity can be obtained from the continuous longitudinal length within the event impact zone that meets the minimum traffic width requirement, preferably expressed as:

[0131]

[0132] In the formula, Indicates the first Lane in the future The continuity of passage corresponds to the aforementioned mapping result. ; Indicates the first The lanes must meet the minimum traffic width within the longitudinal section affected by the incident. The continuous passable longitudinal length; Indicates the vertical range of event impact. The total length; The vehicle width and lateral safety clearance parameters should be kept consistent with those in the restored traffic window. .

[0133] (3) The degree of queuing encroachment can be obtained by the area overlap ratio between the predicted queuing area and the target lane's affected area, preferably expressed as:

[0134] In the formula, Indicates the first Lane in the future The degree of queuing encroachment corresponds to the aforementioned mapping results. ; This represents the predicted queue vehicle coverage area obtained by mapping the future queue length and queue tail propagation speed predicted in step 3. Indicates the first The number of lanes in the queue. The larger this value, the more severe the queuing vehicles are occupying the target lane.

[0135] (4) The obstacle occupancy intensity can be obtained by combining the area overlap ratio between the predicted event area and the target lane's affected area and the event boundary expansion trend, preferably expressed as:

[0136] In the formula, Indicates the first Lane in the future The obstacle occupancy intensity corresponds to the aforementioned mapping result. ; This indicates that the event state is predicted by step 3. The region occupied by the predicted event obtained through mapping; This represents the amount of future event boundary expansion in the predicted event state in step 3; The weight representing the impact of boundary expansion is preferably within a certain range. ; This indicates that the obstruction intensity is limited to the range of 0 to 1. Therefore, in the lane reopening window... , , , All are the future event states output in step 3 and road structural features A clear correspondence is formed.

[0137] No. The definition of the lane reopening window is:

[0138] in, This represents the threshold for the probability of lanes being open. Indicates the width of the main vehicle. Indicates the lateral safety distance between vehicles. This indicates the queuing encroachment threshold. This indicates the threshold for obstacle occupancy.

[0139] This lane reopening window indicates not only whether a lane is passable, but also which time slots within the future forecast time domain allow vehicles to safely enter that lane. Therefore, path planning no longer relies solely on local obstacle avoidance based on the current obstacle location, but can make forward-looking decisions by incorporating future lane reopening trends.

[0140] Step 5. Generate a temporary access permission diagram with lane-action-time domain association: After obtaining the lane restoration access window, this invention generates a temporary access permission map. This temporary access permission map is not a typical risk scoring table, but rather a permission constraint structure that simultaneously includes three dimensions: lane, action, and time window. It is used to directly convert UAV observations into action permission boundaries for autonomous vehicles.

[0141] Suppose the current analysis region has a total of Lanes, the lane set is:

[0142] The candidate action set is as follows:

[0143] in, Indicates lane keeping action. Indicates a left lane change maneuver. Indicates a right lane change maneuver. Indicates deceleration and hold action. This indicates a conservative approach to using the passage. This indicates a halt to the forward movement. "Minimum risk parking action" indicates the action with the lowest risk; "continuous lateral action" refers to the continuous execution of left lane change, right lane change or conservative lane change; "high dynamic action" refers to the combination of actions in the path where the target acceleration, jerk or lateral displacement exceeds the preset comfort threshold.

[0144] The temporary access permission diagram is defined as follows:

[0145] in, Indicates the first Lane, Action Predicted time slices The corresponding permission status. Permission status includes three categories: allowed, restricted, and prohibited.

[0146] The permission scoring function can be expressed as:

[0147] In the formula, Indicates the first Lane, Action Predicted time slices Corresponding permission rating; Indicates the first The probability of each lane opening in the future; This indicates consistency in road rules; This indicates that the action is consistent with the main vehicle's mission objective; This indicates the consistency between the timing of the action execution and the window for restoring lane traffic flow; Indicates the intensity of risk or conflict arising from the action; to This indicates the weight of each scoring item. The first four items are used to increase the permission score, while the risk / conflict items are used to decrease the permission score.

[0148] The weighting coefficients satisfy:

[0149]

[0150] In the formula, Indicates the first The weights of each permission scoring factor; this constraint is used to ensure that the permission score is obtained by normalized weighting of each factor.

[0151] The consistency between actions and the main vehicle's mission objectives can be expressed as:

[0152] In the formula, This indicates that the action is consistent with the main vehicle's mission objective; Indicates the lane number; Indicates the lane number of the target lane; lane Deviation from the target lane (the desired lane given in advance by the navigation task, exit direction, road rules, current driving intention, or passage task); Indicates the lane deviation scale parameter; Indicates the execution of an action The corresponding target speed; This indicates the target speed of the main vehicle's mission (which is constrained by road speed limits, event zone control speeds, following safety speeds, and mission cruise speeds). This represents the speed deviation scale parameter. The smaller the lane deviation and speed deviation, the larger the consistency index.

[0153] The consistency between the timing of action execution and the restoration of the passage window can be represented as:

[0154] In the formula, Indicates the time of action execution and the number of... The consistency of the passage windows for each lane has been restored; Indicates characteristic functions; Indicates the time when a future action will be performed; Indicates the first The window for restoring traffic to each lane. When belong At that time, the indicator was taken Otherwise take .

[0155] The intensity of risk or conflict arising from an action can be expressed as:

[0156] In the formula, Indicates the intensity of risk or conflict corresponding to lane, action, and time slice; Indicates the execution of an action The collision time between the subsequent object and the potential conflicting object; It represents a very small positive number and is used to avoid the denominator being zero; Indicates the intensity of obstacle occupancy; Indicates the degree of encroachment on the queue; This indicates the risk factor of the action itself; to This indicates the weight of each risk item. The shorter the collision time, the more severe the obstacle occupation or queue encroachment, the greater the intensity of the risk conflict.

[0157] Permission status can be divided according to the scoring threshold as follows:

[0158] In the formula, Indicates the first Lane, Action Time slice The corresponding permission status; This indicates the corresponding permission level rating; Indicates the prohibition threshold; Indicates the allowable threshold; numerical value Indicates prohibition, numerical value Indicates a restriction, numerical value This indicates permission. When the permission score is below the prohibition threshold, the system determines that the action is prohibited; when the permission score reaches the allowance threshold, the system determines that the action is allowed.

[0159] For example, when a candidate lane has a clear window for reopening traffic within the next 5 to 10 seconds, and the road rules allow entry, left lane change, right lane change, or conservative lane borrowing actions can be set to permitted; when a lane is continuously occupied by accident vehicles or debris and the probability of reopening is low, left lane change, right lane change, or conservative lane borrowing actions can be set to prohibited; when an emergency lane does not meet the legal or safe conditions for borrowing, conservative lane borrowing actions can be set to prohibited; when the probability of lane opening is in the intermediate state, the corresponding action can be set to restricted, and subsequent reliability gating can be required for further judgment.

[0160] Step 6. Generate the original candidate path set and perform action permission filtering: At the path planning layer, the system generates an original set of candidate paths based on the current state of the master vehicle, the road topology, the set of candidate target lanes, the master vehicle's target speed, and the path planning task. The candidate target lanes refer to the lanes that the master vehicle can continue to maintain or prepare to enter in the current prediction time domain, including the master vehicle's current lane, adjacent variable entry lanes, and lanes allowed to enter through the temporary access permission map, and not just lanes other than the master vehicle's current driving lane.

[0161] The set of candidate target lanes can be represented as:

[0162] In the formula, Indicates time The target lane set; Indicates the first One lane; Indicates the first Lane, Action Time slice The corresponding permission status; This indicates that the lane is not prohibited under the corresponding action and time slice; Indicates the first This set represents the candidate lanes that can be entered or maintained during the current route planning.

[0163] The original set of candidate paths can be represented as:

[0164] In the formula, Indicates time The generated set of original candidate paths; , to Indicate different candidate paths; This indicates the number of candidate paths. Each candidate path corresponds to a driving plan that the main vehicle may execute within a certain period of time.

[0165] Each candidate path can be represented as:

[0166] In the formula, Indicates the first Candidate paths; Indicates the discrete point number of the path; This indicates the number of discrete points contained in a single candidate path; Indicates vertical position; Indicates lateral offset; Indicates the target speed; Indicates the target acceleration; Indicates the target jerk; Indicates the corresponding time.

[0167] The lateral acceleration constraint can be expressed as:

[0168] In the formula, Indicates the first Path number Lateral acceleration at discrete points; This represents the maximum permissible lateral acceleration; this constraint is used to ensure that the candidate path meets the requirements for vehicle lateral stability and comfort.

[0169] The jerk constraint can be expressed as:

[0170] In the formula, Indicates the first Path number Target accelerometer at discrete points; This indicates the maximum permissible jerk; this constraint is used to ensure the smoothness of the path execution process and the comfort of the ride.

[0171] The velocity constraint can be expressed as:

[0172] In the formula, Indicates the first Path number The target velocity at a discrete point; This indicates the current speed limit for the road segment; this constraint is used to ensure that the speed of the candidate path is not negative and does not exceed the road speed limit.

[0173] Furthermore, based on the temporary access permission map generated in step 5 and the restored access window calculated in step 4, the original candidate path set is filtered by action permission and entry time to obtain the candidate path set after permission filtering.

[0174] The set of candidate paths after permission filtering can be represented as:

[0175] In the formula, This represents the set of candidate routes after filtering by the temporary access permission map and the lane restoration window; Indicates the first Candidate paths; Represents the original set of candidate paths; Indicates the corresponding permission status; Indicate candidate path Entering the The timing of each lane; Indicates the first The resumption window for each lane. This set is used to retain candidate paths where the action is not prohibited and the entry into the target lane meets the resumption window requirements.

[0176] The screening process includes two levels of constraints: First, action permission constraints. If a candidate path contains an action that is deemed prohibited by the temporary access permission graph, the candidate path is eliminated.

[0177] Second, time-permission constraints. Even if an action is permitted, if the time when the path enters the target lane is not within the lane's reopening window, the candidate path will still be eliminated.

[0178] Therefore, this step allows the UAV observation results and event evolution prediction results to no longer remain at the information display level, but to directly affect the generation and pruning process of the candidate path set.

[0179] In this embodiment, if the candidate path set after permission filtering If the set is empty, hierarchical degradation control is triggered, and step 10 is executed; otherwise, step 7 is executed.

[0180] Step 7. Construct UAV observation reliability indicators: External observations by drones do not always maintain the same reliability. In scenarios involving sudden incidents on highways, factors such as communication obstruction, image compression, rain, snow, fog, drone attitude jitter, obstruction of key areas, transmission delays, packet loss, insufficient battery power, and air-to-ground perception conflicts can all reduce the drone's ability to support autonomous driving decisions. Therefore, this invention constructs a drone observation reliability index and uses it for subsequent action intensity gating.

[0181] UAV observation reliability indicators It can be represented as:

[0182] In the formula, Indicates time The reliability indicators of UAV observation; This indicates high-confidence coverage in key areas; Indicates normalized communication delay; This represents the normalized packet loss rate; Indicates the normalized air-to-ground perception conflict level; This indicates the normalized remaining battery power of the drone; Used to indicate communication latency reliability; Used to indicate the integrity of communication data; Used to represent air-to-ground awareness consistency; to This indicates the weight of each reliability factor.

[0183] The weighting coefficients satisfy:

[0184]

[0185] In the formula, Indicates the first The weights of the reliability factors; this constraint is used to ensure that the reliability index of UAV observation is obtained by weighting five categories of normalized factors.

[0186] High confidence coverage in key areas can be expressed as:

[0187] In the formula, This indicates high-confidence coverage after integrating all key areas; Indicates the key area number; Indicates the total number of key areas; Indicates the first The importance weight of each key region; Indicates the first Key areas at time A sign indicating whether the area is covered by drones; Indicates the first The formula is used to calculate the weighted average of the coverage status and observation confidence of each key region.

[0188] Normalized communication delay can be expressed as:

[0189] In the formula, Indicates normalized communication delay; This indicates the moment when the vehicle or edge computing unit receives data from the drone. Indicates the moment when the drone generated the observation data; This represents the maximum permissible communication delay; the minimum function is used to normalize the communication delay to no more than [a certain value]. Within the range.

[0190] The normalized packet loss rate can be expressed as:

[0191] In the formula, This represents the normalized packet loss rate; This indicates the number of data frames that were not successfully received in the most recent communication window; This represents the total number of data frames that should be received within the most recent communication window; this formula is used to characterize the proportion of UAV data lost within the most recent communication window.

[0192] The degree of air-to-ground perception conflict can be expressed as:

[0193] In the formula, Indicates the degree of air-to-ground perception conflict; This indicates the positional difference between drone observation results and vehicle-mounted or roadside perception results; This indicates differences in lane occupancy assessment; This indicates a difference in target velocity estimates; , and These represent the maximum position difference, maximum lane determination difference, and maximum speed difference used during normalization, respectively. , and This indicates the weight of the different difference items.

[0194] The normalized index of remaining battery power of a drone can be expressed as:

[0195] In the formula, This represents a normalized indicator of the drone's remaining battery power. Indicates the drone's current remaining battery power; Indicates the minimum safe return battery level; Indicates the full charge level; This represents a truncation function used to restrict the result to a specific range. to Within the range.

[0196] Unlike ordinary confidence level judgments, the reliability index in this invention is not only used to determine whether UAV information is available, but also further involved in determining the upper limit of the action intensity of candidate paths.

[0197] In this embodiment, if the reliability of UAV observation is lower than the minimum working threshold, or the degree of air-to-ground perception conflict exceeds the threshold, then graded degradation control is triggered and step 10 is executed; otherwise, step 8 is executed.

[0198] Step 8. Action intensity gating based on UAV reliability: To prevent vehicles from performing high-risk maneuver combinations even when the observation quality of the UAV deteriorates, this invention defines a path maneuver intensity level function. For a candidate path, its maneuver intensity is determined by both the maneuver type and the kinematic amplitude, and can be expressed as:

[0199] in, This indicates the number of left and right lane changes in the path. Indicates whether the path includes a conservative detour. Indicates the peak acceleration along the path. Indicates the peak jerk along the path. Indicates the maximum lateral displacement of the path. to For weighting coefficients; left lane change, right lane change and conservative lane change correspond to lateral actions, and target acceleration, jerk and maximum lateral displacement correspond to action execution intensity.

[0200] Construct an upper limit for action intensity based on UAV observation reliability indicators. function:

[0201] in, This indicates the lower limit of the permissible intensity of the action. This indicates the maximum allowed intensity of the action. The gating shape parameter is θ∈[2,5].

[0202] Based on the upper limit of action intensity, a second screening is performed on the candidate path set:

[0203] When the reliability of drone observation is high, the upper limit of action intensity With a larger value, the system can retain medium-intensity paths, including single left lane changes, right lane changes, or conservative lane changes. The upper limit of maneuver intensity increases as the reliability of UAV observations decreases. With synchronized deceleration, the system automatically eliminates paths involving continuous left / right lane changes, conservative lane changes, high acceleration, high jerk, or large lateral displacement, retaining only low-intensity paths such as lane keeping, deceleration-based lane keeping, stopping, or minimum-risk parking. Thus, drone reliability is no longer merely an additional consideration, but directly enters the path set selection process, constituting the dynamic safety boundary for autonomous driving path planning.

[0204] In this embodiment, if the final candidate path set after the UAV reliability gating is empty, then hierarchical degradation control is triggered and step 10 is executed; otherwise, step 9 is executed.

[0205] Step 9. Path optimization and target control decision output: The overall cost function can be expressed as:

[0206] In the formula, Indicate candidate path The overall cost; Indicates the cost of travel time; Indicates the cost of path risk; This indicates a trade-off in comfort; This indicates the cost of the task deviating from its intended purpose; This indicates the cost of replanning and jitter. to This represents the weight of each sub-cost, used to comprehensively evaluate the safety, efficiency, comfort, and stability of candidate paths.

[0207] The weighting coefficients satisfy:

[0208]

[0209] In the formula, Indicates the first The weight of each sub-cost; this constraint is used to ensure that the comprehensive cost function is obtained by weighting the five types of normalized sub-costs.

[0210] The time cost of travel can be expressed as:

[0211] In the formula, This represents the travel time cost of the candidate path; Indicate candidate path Time required to complete the planned task; and These represent the shortest and longest travel times in the candidate path set, respectively. This represents a very small positive number, used to prevent the denominator from being zero; this formula is used to normalize the travel time of different candidate paths.

[0212] The cost of comfort can be expressed as:

[0213] In the formula, This represents the comfort cost of the candidate path; Indicates the discrete point number of the path; Indicates the number of discrete points on a single path; Indicates acceleration; This represents jerk; the squared term is used to penalize larger accelerations and jerks to improve path smoothness.

[0214] The cost of mission deviation can be expressed as:

[0215] In the formula, This represents the task deviation cost of the candidate path; Indicates the lane number where the candidate path's endpoint is located; Indicates the lane number of the target lane; Indicates the speed at the end of the candidate path; Indicates the target speed of the main vehicle; and These represent the lane departure cost weight and the speed departure cost weight, respectively.

[0216] The cost of replanning jitter can be expressed as:

[0217] In the formula, This represents the replanning jitter cost of the candidate path; Indicate candidate path The corresponding main action types; Indicates the type of action used in the previous planning cycle; Indicates the path executed or selected in the previous planning cycle; This represents the path difference penalty weight; this formula is used to reduce frequent lane changes, frequent path switching, and lateral control oscillations.

[0218] Step 10. Trigger hierarchical degradation control when there is no solution for the path or the observation is unreliable: This invention no longer uses a single degradation trigger index to uniformly judge all anomalies. Instead, it sets degradation trigger conditions after steps 6, 7, and 8 respectively to maintain a clear execution order.

[0219] To achieve hierarchical conservative control, this invention sets up a three-level degradation mode based on factors such as UAV observation reliability, number of candidate paths, path feasibility, and degree of air-to-ground perception conflict.

[0220] Specifically, when the set of candidate paths for permission filtering obtained in step 6 is empty, it directly enters the third-level degradation mode; when the set of candidate paths for permission filtering is not empty but the number is lower than the minimum candidate path number threshold, it enters the first-level degradation mode and continues to execute steps 7 to 9 only in low-risk candidate paths. The minimum number of candidate paths threshold is denoted as: , which represents the minimum number of candidate paths required to ensure the stability of path optimization.

[0221] The determination that the number of candidate paths after permission filtering is insufficient is as follows:

[0222] In the formula, This represents the set of candidate paths after permission filtering. This indicates the number of paths in the candidate path set after permission filtering; This represents the minimum threshold for the number of candidate paths. This formula is used to indicate that the set of candidate paths is not empty, but the number is below the minimum required for stable path optimization.

[0223] When step 7 determines that the reliability of UAV observation is lower than the minimum reliability threshold or the degree of air-to-ground perception conflict exceeds the air-to-ground perception conflict threshold, it enters the second or third level degradation mode. In this embodiment, the minimum reliability threshold can be expressed as:

[0224] In the formula, This indicates the minimum reliability threshold for UAV observations; Indicates the basic reliability threshold; This represents the weighting coefficient for the speed factor; Indicates the current speed of the main vehicle; Indicates the current road speed limit; This represents the weighting coefficient of the action intensity factor; This represents the intensity index of the current path or planned action. This formula is used to dynamically change the reliability threshold with vehicle speed and action intensity; the higher the vehicle speed and the more aggressive the action, the higher the reliability threshold required by the system for the UAV.

[0225] The air-to-ground sensing conflict threshold can be expressed as:

[0226] In the formula, Indicates the air-to-ground sensing conflict threshold; Indicates the allowable positional error; Indicates the permissible lane determination error; Indicates the allowable speed error; , and These represent the maximum position error, maximum lane determination error, and maximum speed error used during normalization, respectively. , and This represents the weight of different error terms.

[0227] When the number of final candidate paths after reliability gating in step 8 is insufficient, the system enters the second-level degradation mode; when the set of final candidate paths after gating in step 8 is empty, the system enters the third-level degradation mode.

[0228] The determination of insufficient candidate paths after UAV reliability gating is as follows:

[0229] In the formula, Represents the set of candidate paths after reliability gating of the UAV; Indicates the number of paths in the candidate path set after gating; This represents the minimum threshold for the number of candidate paths. This formula is used to indicate that after reliability gating, there are still candidate paths, but their number is insufficient to support stable path optimization.

[0230] In this embodiment, the minimum candidate path number threshold is used to determine whether there are enough alternative solutions for path optimization, and is not used to replace the comprehensive cost function. When the number of candidate paths is sufficient, the system performs normal path optimization; when the number of candidate paths is insufficient but low-risk candidate paths still exist, the system continues to select paths according to the comprehensive cost function within the conservative action set; when the candidate paths are empty or the key observations are seriously unreliable, normal path optimization is no longer performed, but a higher-level degradation strategy is directly output.

[0231] Furthermore, in this embodiment, the set of low-risk candidate paths includes lane-keeping, deceleration-keeping, stop-and-go, and minimum-risk stopping paths; in cases of minor anomalies and where temporary passage permission allows, it may include one left lane change, one right lane change, or one conservative detour. The feasibility of a low-risk candidate path requires simultaneous satisfaction of vehicle dynamics constraints, road rule constraints, temporary passage permission constraints, resumption of passage window constraints, and minimum safe distance constraints; if any constraint is not satisfied, the path is deemed infeasible.

[0232] Furthermore, in this embodiment, the first-level degradation mode is used to execute under mildly abnormal conditions. When the reliability of UAV observation is slightly lower than the normal working threshold, or when the number of candidate paths after permission filtering is insufficient but low-risk candidate paths still exist (the set of candidate paths after permission filtering is not empty but the number is lower than the minimum candidate path number threshold), the system restricts the action set, allowing only lane keeping, deceleration and holding, and one left lane change, one right lane change, or one conservative lane change; at the same time, it reduces the target speed and increases the replanning hysteresis threshold, prohibiting continuous left / right lane changes and continuous conservative lane changes.

[0233] The second-level degradation mode is used to execute under moderate abnormal conditions. When there is insufficient coverage in critical areas, high communication latency or packet loss rate, significant air-to-ground perception conflicts, or insufficient number of candidate paths after reliability gating (below the set threshold), the system only allows lane keeping and deceleration; if the safe distance ahead is insufficient (below the set threshold), it further switches to stopping or minimum risk parking.

[0234] The third-level degradation mode is used to execute under severe abnormal conditions. When the candidate path set is empty after permission filtering, the candidate path set is empty after reliability gating, the reliability of UAV observation is significantly lower than the minimum reliability threshold, the air-to-ground perception conflict significantly exceeds the air-to-ground perception conflict threshold, or information in key areas is severely lacking, the system will execute a stop-from-go, minimum-risk stop, or manual takeover request, and will no longer output paths that include lateral lane changes or detours.

[0235] This invention employs the aforementioned hierarchical degradation control to process situations such as unsolvable paths, unreliable observations, and excessive air-to-ground perception conflicts in a sequential manner. This avoids mixing normal path optimization with degradation control logic and ensures that vehicles do not continue to perform high-risk lateral maneuvers when external observations are insufficient. Furthermore, this degradation control is not a typical safe parking strategy, but a hierarchical conservative control mechanism jointly triggered by UAV observation reliability, temporary access permission maps, candidate path gating results, and air-to-ground perception conflicts.

[0236] In this embodiment, the degradation target control velocity can be expressed as:

[0237] In the formula, This indicates the target control speed output to the vehicle longitudinal controller in degradation mode; This represents the reference speed obtained from normal path planning; Indicates the degradation rate adjustment gain; Indicates the reliability index of UAV observations; This indicates the degree of unreliability of UAV observations; this formula is used to reduce the target control speed based on the reliability of UAV observations. The lower the reliability of UAV observations, the lower the degraded target control speed.

[0238] With further restrictions, the degenerate target control velocity can be expressed as:

[0239] In the formula, This represents the degenerate target control velocity after being constrained by upper and lower limits. Indicates the minimum safe control speed; Used to ensure that the degradation rate is not lower than the minimum safe rate; This formula is used to ensure that the degradation rate does not exceed the normal reference speed; it is also used to avoid excessive speed correction that could lead to unstable vehicle control.

[0240] The degradation rate adjustment gain can be expressed as:

[0241] In the formula, Indicates the degradation rate adjustment gain; Indicates the maximum allowable rate of degradation; This indicates the maximum comfortable deceleration allowed in degradation mode; This indicates the time window for controlling response time or speed adjustment; This indicates the reference speed for normal path planning; This represents a very small positive number, used to prevent the denominator from being zero; this formula is used to determine the speed reduction based on the vehicle's comfort braking capability and reference speed.

[0242] In this embodiment, the degradation target control speed is not a fixed value, but is dynamically determined based on the executable reference speed of the previous cycle, the road speed limit, the comfort deceleration, and the reliability of UAV observations. When the normal candidate path is empty, the reference speed can be taken as the target speed of the path executed in the previous cycle, the current vehicle speed, or the safe speed under the road speed limit, and gradually reduced through comfort deceleration constraints to avoid sudden braking that could lead to control instability. This speed is used by the longitudinal controller to perform deceleration hold, stop forward, or minimum risk parking.

[0243] Step 11. Rolling Updates and Decision Lag: During the vehicle's execution of the target path or degradation strategy, the UAV continuously refreshes the observation data according to a preset cycle. Based on the latest observation data, the system re-executes steps 2 to 10, including event evolution prediction, lane restoration window calculation, temporary access permission map generation, candidate path screening, reliability gating, path optimization, and degradation control judgment.

[0244] To avoid vehicle motion jitter caused by continuous replanning, this invention introduces a decision hysteresis mechanism. When the cost difference between the old and new paths is less than the hysteresis threshold, and the change in the action category of the new path may lead to unnecessary lateral switching, the system prioritizes maintaining the decision from the previous cycle. The output decision is defined as follows:

[0245] in, Indicates the hysteresis threshold. This indicates the action category corresponding to the path.

[0246] By using rolling updates and decision lag, this invention can adapt to the continuous changes in the scope of lane occupation, queue length, temporary lane formation and handling progress of highway emergencies, enabling autonomous vehicles to make decisions based on the latest event evolution results and the latest UAV observation reliability, while avoiding control instability caused by frequent lateral movements.

[0247] The technical logic of this invention can form the following closed loop: First, continuous UAV observation transforms images, videos, and communication status into structured event state variables. Then, the system predicts future event evolution trends based on continuous observation results, vehicle status, and road structure characteristics. On this basis, the system further calculates lane opening probabilities and lane restoration windows, and generates a temporary access permission map with lane-action-time domain association. Finally, the system uses the temporary access permission map and UAV observation reliability indicators to screen, gate, and optimize candidate paths. When candidate paths are empty, lack sufficient reliability, or experience excessive air-to-ground perception conflicts, the system enters a tiered degradation control mode, thereby ensuring the foresight, feasibility, and safety of path decision-making in highway emergency scenarios.

[0248] Furthermore, such as Figure 2 As shown, this embodiment also provides a path decision system for implementing the above-described method, including an air-ground multi-source data acquisition module, a UAV observation data structuring module, a sudden event evolution prediction module, a lane restoration window calculation module, a temporary access permission map generation module, a candidate path generation and permission filtering module, a UAV observation reliability assessment and gating module, a path optimization and control execution module, a graded degradation control module, and a rolling update and closed-loop decision module; The multi-source air-to-ground data acquisition module collects vehicle operating status, road structure information, continuous UAV observation data, and communication status information through vehicle-mounted sensors, vehicle controllers, positioning and navigation modules, vehicle-to-everything (V2X) terminals, high-precision maps, roadside facilities, cloud event platforms, and UAV platforms. Then, the aforementioned multi-source data is uniformly transmitted to the vehicle-mounted computing unit, roadside edge computing unit, or cloud platform. Through timestamp synchronization, coordinate transformation, and map matching methods, the vehicle-mounted data, roadside data, platform data, and UAV observation data are unified to the same time reference and road coordinate reference. The UAV observation data structuring module retrieves information such as UAV video, images, pose, flight altitude, and observation angle from the air-to-ground multi-source data acquisition module. Through target detection, semantic segmentation, lane line recognition, vehicle tracking, and road coordinate registration methods, it performs structured analysis on the area of ​​the sudden incident ahead. It extracts the status information of the accident core area, the tail of the queue area, the temporary passage area, the lane boundary area, and the emergency response area, and forms a UAV observation vector that includes the length of the lane occupied by the incident, the number of lanes occupied, the queue length, the propagation speed of the queue tail, the effective width of the temporary passage, the expansion of the event boundary, the progress of emergency response, and the observation confidence of key areas.

[0249] The emergency event evolution prediction module predicts the short-term evolution of an emergency event based on the continuous observation sequence output by the UAV observation data structuring module, combined with the main vehicle operating status, number of road lanes, lane width, emergency lane availability, event type, and topological attributes of the event location. By analyzing changes in the event's occupied area, the trend of the queue tail moving, changes in temporary lane width, and changes in emergency response progress, it predicts the degree of impact of the event on each target lane within the future time window and outputs the opening probability of each target lane. The lane resumption window calculation module retrieves the prediction results from the emergency evolution prediction module. Based on the target lane opening probability, predicted effective lane width, predicted queue encroachment degree, and predicted obstacle occupancy intensity, it judges the resumption conditions for each target lane in the future time window. When the target lane simultaneously meets the following conditions: opening probability reaches a threshold, effective lane width meets vehicle passage requirements, queue encroachment degree is below a threshold, and obstacle occupancy intensity is below a threshold, it is determined that the lane has the conditions to resume traffic in the corresponding time slice, and a lane resumption window is generated. The temporary access permission map generation module, based on the access restoration window output by the lane restoration window calculation module, and combined with highway road rules, event handling status, main vehicle task objectives, and action risk constraints, constructs a temporary access permission map with lane-action-time domain association. Simultaneously, it determines the permissions for candidate actions such as lane keeping, left lane change, right lane change, deceleration and maintenance, conservative lane changing, stopping, and minimum risk parking, classifying the state corresponding to each lane, each action, and each predicted time slice as permitted, restricted, or prohibited. Through this module, UAV observation results are no longer merely external prompts but are transformed into action boundaries and path constraints that can be directly executed by autonomous vehicles.

[0250] The candidate path generation and permission filtering module generates an initial set of candidate paths based on the current status of the main vehicle, road topology, candidate target lanes, target speed, and path planning task. Subsequently, it retrieves permission constraint information output by the temporary access permission map generation module and time window information output by the lane restoration access window calculation module to filter the candidate paths. If a candidate path contains an action deemed prohibited by the permission map, the path is eliminated. Similarly, if a candidate path enters the target lane at a time outside the lane restoration access window, the path is also eliminated. After this filtering process, a set of executable candidate paths that satisfy both action permission constraints and time permission constraints is obtained.

[0251] The UAV observation reliability assessment and gating module comprehensively considers the coverage of key areas such as the accident core area, the tail of the queue area, the lane boundary area, the temporary passage area, and the candidate path entry area by the UAV, as well as factors such as communication latency, packet loss rate, air-to-ground perception conflict level, and the remaining battery power of the UAV to calculate the UAV observation reliability index. Based on the reliability index, it determines the upper limit of the action intensity of the candidate path and restricts high-risk action combinations such as continuous left / right lane changes, conservative lane changes, high acceleration, high jerk, and large lateral displacement in the candidate path. When the UAV observation reliability is high, the system allows the retention of medium-intensity paths, including single left lane changes, right lane changes, or conservative lane changes; when the UAV observation reliability decreases, the system automatically eliminates high-risk lateral actions and high-dynamic paths, retaining only low-intensity paths such as lane keeping, deceleration and holding, stopping, or minimum-risk stopping.

[0252] The path optimization and control execution module calculates the comprehensive path cost from the candidate path set output by the UAV reliability gating module based on travel time cost, path risk cost, comfort cost, mission deviation cost, and replanning jitter cost, and selects the candidate path with the lowest cost as the target execution path. If there are multiple paths with similar costs, the path with fewer lateral movements, lower movement intensity, higher path smoothness, and better consistency with the previous cycle decision is selected first. After determining the target execution path, the module outputs the target speed, target acceleration, target lateral trajectory, and movement sequence, and controls the vehicle to execute according to the target path.

[0253] The graded degradation control module is activated when the candidate path set after permission filtering is empty or the number of candidate paths is insufficient, the reliability of UAV observation is lower than the minimum reliability threshold, the degree of air-to-ground perception conflict exceeds the air-to-ground perception conflict threshold, or the candidate path set after gated access is empty or the number of candidate paths is insufficient. It then triggers degradation judgments in the execution order of steps 6, 7, and 8, and performs three levels of conservative control—mild, moderate, and severe—based on the degree of anomaly. For mild anomalies, only lane keeping, deceleration and hold, and one left lane change, right lane change, or conservative lane change are allowed; for moderate anomalies, only lane keeping and deceleration and hold are allowed; for severe anomalies, a stop, minimum risk parking, or manual takeover request is executed.

[0254] The rolling update and closed-loop decision-making module receives the latest observation data from the UAV at preset intervals during the vehicle's execution of the target path or degradation strategy, and re-triggers the processes of event evolution prediction, lane restoration window calculation, temporary access permission map generation, candidate path screening, reliability gating, and path optimization. Simultaneously, this module introduces a decision-holding mechanism. When the cost difference between the old and new paths is small and changes in action categories may lead to unnecessary lateral switching, the previous cycle's decision is prioritized to reduce frequent path switching and lateral control oscillations, ensuring the continuity of vehicle decision-making and control stability in highway emergency scenarios.

[0255] The present invention also provides an electronic device, comprising: one or more processors and a memory; wherein the memory is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described method for UAV cooperative path decision-making for highway emergencies.

[0256] The present invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for unmanned aerial vehicle (UAV) cooperative path decision-making for highway emergencies.

[0257] Those skilled in the art will understand that all or part of the functions of the various methods / modules in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved.

[0258] In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the programs can also be stored in storage media such as servers, other computers, disks, optical discs, flash drives, or portable hard drives. They can be downloaded or copied to the memory of the local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be implemented.

[0259] The above describes specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included 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 method for collaborative path decision-making by unmanned aerial vehicles (UAVs) for highway emergencies, characterized in that: Includes the following steps: Step 1. Acquire multi-source data from both air and ground sources, and determine whether to initiate UAV collaborative observation; if so, proceed to Step 2. Step 2. The UAV continuously observes the target observation area ahead and converts the observation data into structured variables; Step 3. Based on the historical observation sequence of UAVs, the status of the main vehicle, and the road structure characteristics of the event impact area, predict the evolution trend of the sudden event within the future time window through a short-time state-space prediction model; Step 4. Calculate the lane opening probability and lane resumption window based on the emergency evolution prediction results; Step 5. Taking into account lane opening probability, lane resumption window, road rules, action risk, and main vehicle task objective, generate a temporary access permission map with lane-action-time domain association; Step 6. Generate the original candidate route set and filter by action permission and entry time according to the temporary access permission map and lane restoration window to generate candidate routes after the initial screening. If the number of candidate routes meets the requirements, proceed to step 7. Step 7. Assess the reliability of UAV observations. If it meets the requirements, proceed to step 8. Step 8. Calculate the path action intensity level for each candidate path, and determine the upper limit of action intensity based on the UAV observation reliability index. Perform a second screening of the candidate path set based on the upper limit of action intensity. If the number of candidate paths after the second screening meets the requirements, proceed to step 9. Step 9. Select the optimal route by combining travel time, route risk, comfort, task deviation, and replanning jitter.

2. The UAV cooperative path decision-making method for highway emergencies according to claim 1, characterized in that, The multi-source air-ground data mentioned in step 1 includes vehicle status information, road structure information, continuous UAV observation data, vehicle-road cooperative messages, roadside or cloud event records, and communication status information. Among them, the vehicle status information includes the vehicle's longitudinal position, lateral offset, speed, acceleration, jerk, lane number, and vehicle dynamics constraints. The road structure information includes the number of lanes, lane width, emergency lane location, median strip location, ramp entrance and exit locations, speed limit information, no-stopping and no-entry rules, emergency lane borrowing rules, and road topology attributes of the event area location. In step 1, based on the constructed emergency event triggering index, it is determined whether to initiate UAV collaborative observation. When the emergency event triggering index is greater than or equal to the emergency event triggering threshold, it is determined that there is a risk of an emergency event ahead, and UAV collaborative observation is initiated. The emergency event triggering index is obtained by weighted summation of the abnormal index of sudden deceleration of traffic flow ahead, the abnormal index of queue overflow, the abnormal index of vehicle-road cooperative event messages, and the abnormal index of vehicle-mounted visual detection.

3. The UAV cooperative path decision-making method for highway emergencies according to claim 2, characterized in that, In step 2, through target detection, semantic segmentation, lane line recognition, vehicle tracking, map registration, and spatiotemporal synchronization, the original image information or video of the UAV is transformed into event lane length, number of lanes occupied, queue length, queue tail propagation speed, effective width of temporary passage, debris or event boundary expansion, emergency response progress, and UAV observation confidence of key areas.

4. The UAV cooperative path decision-making method for highway emergencies according to claim 3, characterized in that, In step 3, the continuous observation results of the UAV are constructed into an event evolution state sequence. Combining the current event evolution state, the event state change trend, the main vehicle state, and the road structure characteristics of the event impact area, a short-time state space prediction model is used to predict the event occupancy range, queue propagation trend, temporary channel formation, and emergency response progress within the future time window.

5. The UAV cooperative path decision-making method for highway emergencies according to claim 4, characterized in that, In step 4, the predicted results of future event evolution states obtained in step 3 are first mapped to lane-level effective channel width, traffic continuity, queuing encroachment degree and obstacle occupancy intensity. Combined with the emergency response progress, the probability of lane opening at future moments is calculated. Then, based on vehicle width, safety distance, and opening probability threshold, it is determined whether each lane can be safely entered in the future time slice, and a lane restoration window is generated.

6. The UAV cooperative path decision-making method for highway emergencies according to claim 5, characterized in that, Step 5 integrates the probability of lane opening at future times, consistency of road rules, consistency of actions with the master vehicle's task objectives, consistency of action execution time with the lane reopening window, and the intensity of risks or conflicts caused by the action to calculate a permission score. Based on the permission score, the permission status of the temporary access permission map is set to allowed, restricted, or prohibited. The intensity of risks or conflicts caused by the action is expressed as follows: ； In the formula, lane ,action and time slices The corresponding risk or conflict intensity; Indicates the execution of an action The collision time between the subsequent object and the potential conflicting object; It represents a positive number and is used to avoid the denominator being zero; Indicates the first Obstacle occupancy intensity of each lane at future time; Indicates the first The extent of queue encroachment on each lane in future timeframes; This indicates the risk factor of the action itself; to This indicates the weight of each risk item.

7. The UAV cooperative path decision-making method for highway emergencies according to claim 6, characterized in that, In step 6, an initial set of candidate paths is generated based on the current state of the main vehicle, the road topology, candidate target lanes, the target speed of the main vehicle, and the path planning task. The candidate target lanes refer to the lanes that the main vehicle will continue to maintain or prepare to enter in the current prediction time domain. Then, the action permission map and the lane restoration window are used to filter the action permission and entry time, and paths containing prohibited actions or whose entry times do not meet the restoration window requirements are eliminated. If the candidate path set after the initial screening is empty or the candidate path set is not empty but the number is lower than the minimum candidate path number threshold, then hierarchical degradation control is triggered. In step 7, the high-confidence coverage of key areas, communication latency, packet loss rate, air-to-ground perception conflict level and UAV remaining battery power are considered to calculate the UAV observation reliability index. If the UAV observation reliability is lower than the minimum reliability threshold, or the air-to-ground perception conflict level exceeds the air-to-ground perception conflict threshold, then graded degradation control is triggered. Path motion intensity level in step 8 The expression is: ； in, This indicates the number of left and right lane changes in the path. Indicates whether the path includes a conservative detour. Indicates the peak acceleration along the path. Indicates the peak jerk along the path. Indicates the maximum lateral displacement of the path. to These are the weighting coefficients; Action intensity limit expression: ； in, This indicates the lower limit of the permissible intensity of the action. This indicates the maximum allowed intensity of the action. Indicates the gating shape parameters, Indicates time The reliability indicators of UAV observation; In step 8, if the final candidate path set after secondary screening is empty, or the number of final candidate paths is lower than the minimum candidate path number threshold, then hierarchical degradation control is triggered.

8. The UAV cooperative path decision-making method for highway emergencies according to claim 7, characterized in that, In step 9, the comprehensive cost is calculated for each of the final candidate paths after the second screening, and the candidate path with the smallest comprehensive cost is selected as the target execution path. The target speed, target acceleration, target lateral trajectory and action sequence are output. The comprehensive cost is obtained by weighted summation of travel time cost, path risk cost, comfort cost, task deviation cost and replanning jitter cost.

9. The UAV cooperative path decision-making method for highway emergencies according to claim 8, characterized in that, The graded degradation control is divided into three levels. When the reliability of UAV observation is only lower than the normal working threshold, or when the candidate path set after initial screening is not empty but the number is lower than the minimum candidate path number threshold, the first-level degradation mode is executed. The system restricts the action set, allowing only lane keeping, deceleration and holding, and one left lane change, one right lane change, or one conservative lane change. Continuous left / right lane changes and continuous conservative lane changes are prohibited. Within the conservative action set, the path is selected according to the comprehensive cost function. When the reliability of UAV observation is lower than the minimum working threshold but not exceeding the set amount, the degree of air-to-ground perception conflict exceeds the air-to-ground perception conflict threshold but not exceeding the set amount, or the number of final candidate paths after secondary screening is lower than the set threshold, the second-level degradation mode is executed, and the system only allows lane keeping and deceleration to be maintained; if the safe distance ahead is lower than the set threshold, it switches to stop moving forward or minimum risk parking. If the initial candidate path set is empty, the final candidate path set after secondary screening is empty, the UAV observation reliability is lower than the minimum working threshold and exceeds the set amount, or the air-to-ground perception conflict level exceeds the air-to-ground perception conflict threshold and exceeds the set amount, the system will execute a stop-movement, minimum-risk stop, or manual takeover request, and will no longer output paths that include lateral lane changes or detours.

10. A path decision system for implementing the UAV cooperative path decision method according to any one of claims 1 to 9, characterized in that, The path decision system includes an air-ground multi-source data acquisition module, a UAV observation data structuring module, an emergency evolution prediction module, a lane restoration window calculation module, a temporary access permission map generation module, a candidate path generation and permission filtering module, a UAV observation reliability assessment and gating module, a path optimization and control execution module, a graded degradation control module, and a rolling update and closed-loop decision module.

Images