Low-altitude logistics distribution unmanned aerial vehicle route optimization method and system based on ACOR-World and RINA*

Abstract

The application discloses a kind of based on ACOR-World and RINA* low-altitude logistics distribution unmanned aerial vehicle route optimization method and system, comprising: obtaining the multi-source space-time data of low-altitude logistics distribution area, and unmanned aerial vehicle operating state and distribution task constraint, constructs low-altitude logistics three-dimensional space-time route map, and defines candidate flight segment action;Call action condition occupation risk world model ACOR-World, deduce the space-time occupation risk field after candidate flight segment execution, and generate dynamic occupation risk pipe gallery;Candidate route is generated in pipe gallery using RINA*, and the comprehensive traffic generation value and flight segment risk contribution degree are calculated;According to risk contribution degree and forecast uncertainty, determine high-risk to be reconstructed flight segment, feed back its local space-time window to ACOR-World to carry out fine-grained risk deduction, and by RINA*, execute local incremental reconstruction, output target route.The application can reduce the number of high-risk flight segment and global re-planning times, improve the safety, real-time performance and stability of low-altitude logistics distribution route.

Description

Technical Field

[0001] This invention belongs to the fields of low-altitude intelligent transportation, UAV route planning and artificial intelligence optimization technology, specifically involving a low-altitude logistics delivery UAV route optimization method and system based on ACOR-World and RINA*. Background Technology

[0002] With the development of the low-altitude economy, on-demand delivery, and urban last-mile logistics demands, low-altitude logistics delivery drones are increasingly being used in scenarios such as park delivery, urban last-mile delivery, and emergency material transfer. Low-altitude logistics delivery offers advantages such as flexible routes, fast response times, and minimal impact from ground road congestion. However, its operating environment typically presents challenges such as dense buildings, limited low-altitude zones, complex dynamic flight targets, significant local weather disturbances, unstable communication coverage, and strict restrictions on no-fly zones. Failure to promptly identify the occupancy and passage risks of candidate flight segments in the future can easily lead to route conflicts, excessive energy consumption, communication interruptions, or unauthorized entry into restricted airspace, thereby affecting delivery efficiency and operational safety.

[0003] Existing UAV route planning methods mostly employ static map modeling, risk map construction, heuristic search, sampling planning, or reinforcement learning to generate routes. While these methods can achieve a certain degree of obstacle avoidance and path optimization, most rely primarily on current or historical conditions for planning, making it difficult to extrapolate future risk changes after a UAV executes a candidate flight segment. Furthermore, obstacle risk, weather risk, energy margin, communication risk, and airspace compliance status are often treated as independent cost items, failing to reflect the coupling relationship between low-altitude environmental conditions, UAV operational status, and delivery task constraints.

[0004] Compared to ground-based autonomous driving scenarios, low-altitude logistics delivery scenarios possess stronger three-dimensional spatiotemporal attributes. Drone routes are not only affected by horizontal obstacle avoidance but also constrained by factors such as altitude layer switching, climb energy consumption, three-dimensional communication obstruction, local wind disturbances, refueling, and temporary airspace control. Therefore, risk prediction and search methods for ground traffic or ordinary two-dimensional path planning are difficult to directly apply to low-altitude logistics delivery scenarios. There is an urgent need for a low-altitude logistics drone route optimization method capable of performing future risk projections for candidate flight segment actions and directly embedding the risk projection results into the route search and local reconstruction process. Summary of the Invention

[0005] Purpose of the invention: This invention aims to provide a method and system for optimizing the flight path of low-altitude logistics delivery UAVs based on ACOR-World and RINA*, in order to solve the problems of insufficient future risk prediction, weak coupling between risk prediction and flight path search, and large computational overhead of global replanning in dynamic environments in existing methods, thereby improving the safety, real-time performance and stability of flight path planning for low-altitude logistics delivery UAVs.

[0006] Technical Solution: This invention discloses a route optimization method for low-altitude logistics delivery drones based on ACOR-World and RINA*, comprising the following steps:

[0007] Step 1: Acquire multi-source spatiotemporal data of the low-altitude logistics delivery area, drone operation status, and delivery task constraints;

[0008] Step 2: Based on the multi-source spatiotemporal data, UAV operating status and delivery task constraints, construct a three-dimensional spatiotemporal route map for low-altitude logistics that includes spatial location, altitude layer, time layer, candidate flight segments and constraint attributes, and define the connection between adjacent time layer nodes that satisfies flight performance and task constraints as candidate flight segment actions;

[0009] Step 3: Call the pre-trained Action Condition Occupation Risk World Model ACOR-World. Based on the current low-altitude environment, UAV operating status, delivery task constraints, and candidate flight segment actions, deduce the spatiotemporal occupancy risk field of candidate flight segments, and generate prediction uncertainty based on the differences in multiple risk deduction results. The spatiotemporal occupancy risk field includes future occupancy risk, weather disturbance risk, energy consumption margin change, communication loss risk, and airspace compliance status.

[0010] Step 4: Construct a risk discrimination vector based on the spatiotemporal occupancy risk field and prediction uncertainty, perform preset flight condition judgment on nodes and candidate flight segments, and determine the continuously connected flightable nodes and candidate flight segments as dynamic occupancy risk corridors;

[0011] Step 5: Construct the risk-aware neural incremental A* search algorithm RINA*, expand the flight segment within the dynamic occupancy risk corridor, introduce the spatiotemporal occupancy risk field and prediction uncertainty into the node expansion constraints, flight segment risk cost and heuristic estimation cost correction, generate candidate routes, and calculate the comprehensive passage cost of the candidate routes.

[0012] Furthermore, during the candidate route generation process, the risk contribution of the route segment is also calculated. Based on the risk contribution of the route segment and the prediction uncertainty, high-risk routes to be reconstructed are identified, and their local spatiotemporal windows are fed back to ACOR-World as new candidate route action conditions for fine-grained risk simulation. RINA* then performs incremental reconstruction of the local dynamic occupancy risk corridor without changing the global risk field outside the window, and finally outputs the target route and the corresponding comprehensive passage value.

[0013] Furthermore, the multi-source spatiotemporal data includes obstacle data, dynamic flight target data, low-altitude meteorological data, communication coverage data, restricted airspace data, and refueling node data; the UAV operating status includes position, altitude, speed, remaining battery power, payload status, and communication status; the delivery task constraints include delivery start point, delivery end point, mission window, endurance constraints, altitude layer constraints, refueling continuity constraints, and airspace compliance constraints.

[0014] Furthermore, the construction of the three-dimensional spatiotemporal route map for low-altitude logistics includes the following steps:

[0015] Step 2.1: Based on the delivery origin, delivery destination, relay waypoint, refueling node, flightable altitude layer and low-altitude route node, construct a set of low-altitude route nodes, and perform time extension on the low-altitude route nodes according to a preset time interval to obtain the low-altitude route nodes in the time layer.

[0016] Step 2.2: Based on the spatial distance, altitude layer change, maximum flight speed, maximum climb angle, maximum turn angle, remaining battery power, communication reachability and airspace compliance status between adjacent low-altitude route nodes, determine whether it is permissible to establish a flight connection between adjacent nodes, and construct a candidate flight segment set from the nodes that are permitted to establish a flight connection.

[0017] Step 2.3: The three-dimensional spatiotemporal route map for low-altitude logistics includes a set of low-altitude route nodes, a set of candidate routes, and a set of environmental, meteorological, communication, energy consumption, airspace, and delivery task constraint attributes corresponding to the nodes and candidate routes;

[0018] Step 2.4: Move the drone from the node on time layer t Fly to adjacent time layer nodes The connection is defined as a candidate flight segment action, including the change in altitude layer, and at least one change in the UAV's position, speed, remaining battery power, payload status, communication status, and mission status after executing the candidate flight segment.

[0019] Furthermore, the ACOR-World includes a state encoding unit, an action condition residual derivation unit, a risk decoding unit, and an uncertainty estimation unit;

[0020] The state encoding unit generates a current low-altitude world state representation based on current multi-source spatiotemporal data, a three-dimensional spatiotemporal route map of low-altitude logistics, and the UAV's operational status through state encoding mapping. The state coding mapping is used to extract low-altitude airway node features, candidate segment constraint attributes, environmental disturbance features, and UAV state features, and to fuse the spatial connection relationship, temporal connection relationship, and candidate segment attributes between low-altitude airway nodes.

[0021] The action condition residual derivation unit uses candidate flight segment actions As a condition, based on the action condition residual extrapolation mapping, predict the local state residuals of the occupancy status, weather disturbance impact, energy consumption margin, communication reachability status, and airspace compliance status generated in the corresponding local spatiotemporal neighborhood after executing the candidate flight segment. And based on the current state of the low-altitude world Obtain the future world state representation corresponding to the candidate route. ;

[0022] The risk decoding unit represents the future world state. The risk decoding mapping outputs the spatiotemporal occupancy risk field of candidate flight segments. This includes risks such as future occupancy, weather disturbances, changes in energy consumption margin, communication loss, and airspace compliance status.

[0023] The uncertainty estimation unit generates prediction uncertainty based on the differences between multiple risk decoding results from the risk decoding unit:

[0024] ;

[0025] in, Indicates candidate flight segments The corresponding prediction uncertainty, where M represents the number of risk simulations, This represents the result of the m-th risk simulation. This represents the average of multiple risk simulation results.

[0026] Furthermore, the dynamic occupancy risk corridor is generated based on the spatiotemporal occupancy risk field of candidate flight segments and the prediction uncertainty, including the following steps:

[0027] Step 4.1: Construct the risk discrimination vector corresponding to the candidate flight segment based on the spatiotemporal occupancy risk field and prediction uncertainty:

[0028]

[0029] in, Indicates candidate flight segments The corresponding risk discrimination vector, This represents the prediction uncertainty. The risk discrimination vector includes future occupancy risk, weather disturbance risk, energy consumption margin, communication loss risk, airspace compliance status, and prediction uncertainty.

[0030] Step 4.2: Construct a comprehensive risk discriminant value based on the risk discriminant vector, expressed as:

[0031]

[0032] in, Indicates candidate flight segments The corresponding comprehensive risk assessment value, This represents the risk discrimination mapping, which is used to map the risk discrimination vector consisting of the candidate segment spatiotemporal occupancy risk field and prediction uncertainty into a comprehensive risk discrimination value.

[0033] Construct a flightability discrimination function based on the aforementioned risk discrimination vector:

[0034]

[0035] in, This indicates the flightability determination result for the candidate flight segment. Indicates an indicator function, This represents the risk discrimination vector composed of the candidate flight segment spatiotemporal occupancy risk field and prediction uncertainty. This represents the comprehensive risk discrimination value obtained from the risk discrimination vector through the risk discrimination mapping. Indicates the safe threshold for flight. This indicates that the preset airspace rules are met;

[0036] Step 4.3: [The sentence is incomplete and requires more context to be translated accurately.] The candidate segments are identified as flyable candidate segments, resulting in a set of flyable candidate segments. Those that do not meet the criteria... Candidate routes were removed from the subsequent route search scope;

[0037] Step 4.4: Construct a continuous connected subgraph of the flightable candidate air segments and their connected low-altitude airway nodes, and identify the continuous connected subgraph as a dynamic occupancy risk corridor.

[0038] Furthermore, the RINA* performs segment expansion and candidate route generation within the dynamically occupied risk corridor, including the following steps:

[0039] Step 5.1: Using the node corresponding to the delivery origin as the initial expansion node, extract the node corresponding to the current node within the dynamically occupied risk corridor. Connectable flightable candidate segments And exclude candidate flight segments that do not meet the flightability criteria from the extended set;

[0040] Step 5.2: Extend to nodes Candidate segment establishment node extended evaluation function:

[0041] ;

[0042] in, Indicates expansion to nodes The node evaluation value, This indicates the distance from the delivery origin to the current node. The cumulative value of passage Indicates candidate flight segments The risk-value of the flight segment Indicates the current extension node to the delivery destination. Heuristic cost estimation

[0043] The risk cost of the flight segment is obtained by mapping the risk cost based on the risk discrimination vector:

[0044]

[0045] in, This represents a risk-cost mapping. This represents the risk discrimination vector composed of the candidate flight segment spatiotemporal occupancy risk field and prediction uncertainty;

[0046] Step 5.3: Add candidate nodes that meet the requirements of the node expansion evaluation function to the open set, and expand them in order of increasing evaluation value until the delivery destination is reached or the task termination condition is met. Backtrack the parent node relationship to generate candidate routes. ;

[0047] Step 5.4: Calculate the comprehensive transit cost of the candidate routes, which is the candidate route. The sum of the risk and value of each flight segment.

[0048] Furthermore, the risk contribution of the aforementioned flight segment is calculated as follows:

[0049] Step 6.1: Target candidate routes any candidate segment The risk value of the candidate flight segment Replace with preset benchmark value The risk value of the remaining candidate routes remains unchanged, and a counterfactual passage value is constructed:

[0050] ;

[0051] in, Indicates the candidate flight segment Constructed counterfactual common value, This indicates the baseline value of the candidate flight segment under a preset safety reference state; the preset safety reference state is a state in which the risk of future occupancy is low, the risk of weather disturbance is low, the risk of communication loss is low, the airspace compliance status meets the preset airspace rules and retains basic flight energy consumption.

[0052] Step 6.2: Calculate the risk contribution of candidate routes based on the difference between the comprehensive transit value and the counterfactual transit value of the original candidate routes.

[0053] ;

[0054] in, Indicates the risk contribution of candidate flight segments. As a stable term, Indicate candidate routes The comprehensive value of accessibility;

[0055] Alternatively, the risk contribution of the aforementioned flight segment can be determined using a normalized method, expressed as:

[0056]

[0057] in, This indicates that candidate segment e is on the candidate route. Normalized risk contribution in This represents the segment risk value of candidate segment e. It is a stable term.

[0058] Furthermore, the locally dynamically occupied risk corridor is incrementally reconstructed, as follows:

[0059] Step 7.1: When the risk contribution of the candidate flight segment is greater than the risk contribution threshold, or the prediction uncertainty corresponding to the candidate flight segment is greater than the uncertainty threshold, the candidate flight segment is identified as a high-risk flight segment to be reconstructed.

[0060] Step 7.2: Construct a local spatiotemporal window based on the spatial extent of high-risk flight segments to be reconstructed, the expected transit time layer, and the scope of risk propagation:

[0061] ;

[0062] in, Represents a local spatiotemporal window. Represents a node The spatial distance of the highest-risk flight segments awaiting reconstruction Indicates a local spatial range. Indicates a local time range;

[0063] Step 7.3: Feed back the local spatiotemporal window as a new candidate flight segment action condition to ACOR-World, and only perform fine-grained updates on the risk field within the local spatiotemporal window, while keeping the global risk field outside the window unchanged;

[0064] Step 7.4: RINA* performs incremental reconstruction of the dynamically occupied risk corridor within the local spatiotemporal window to obtain a set of local alternative flight segments. It is then connected with the preceding and subsequent segments outside the window in the original candidate routes to generate the target route. .

[0065] This invention also discloses a low-altitude logistics delivery UAV route optimization system based on ACOR-World and RINA*, comprising:

[0066] The multi-source spatiotemporal data acquisition module is used to acquire multi-source spatiotemporal data of the low-altitude logistics delivery area, the drone's operating status, and delivery task constraints;

[0067] The three-dimensional spatiotemporal route map construction module is used to construct a three-dimensional spatiotemporal route map for low-altitude logistics based on the multi-source spatiotemporal data, UAV operating status and delivery task constraints, and define candidate flight segment actions;

[0068] The Action Condition Risk Inference Module is used to call the pre-trained Action Condition Occupation Risk World Model ACOR-World. Based on the current low-altitude environment, UAV operating status, delivery task constraints and candidate flight segment actions, it infers the spatiotemporal occupancy risk field of candidate flight segments and generates prediction uncertainty based on the differences in multiple risk inference results.

[0069] The dynamic occupancy risk corridor generation module is used to construct a risk discrimination vector based on the spatiotemporal occupancy risk field and prediction uncertainty of the candidate flight segment, to judge the flightability of nodes and candidate flight segments, and to generate a dynamic occupancy risk corridor.

[0070] The risk perception route search module is used to call the risk perception neural incremental A* search algorithm RINA* to expand the route within the dynamically occupied risk corridor, generate candidate routes, and calculate the comprehensive passage cost and route risk contribution.

[0071] The local incremental reconstruction module is used to identify high-risk segments to be reconstructed based on the risk contribution and prediction uncertainty of the segments. Its local spatiotemporal window is used as the new candidate segment action conditions and fed back to ACOR-World for fine-grained risk simulation. RINA* then performs incremental reconstruction of the local dynamic occupancy risk corridor without changing the global risk field outside the window.

[0072] The results output module is used to output the target route and its corresponding comprehensive transit value.

[0073] Beneficial effects

[0074] 1. This invention addresses the shortcomings of existing methods in predicting future risks. It uses ACOR-World to extrapolate future local risk states based on candidate flight segment actions. It maps future occupancy risks, weather disturbance risks, energy margin changes, communication loss risks, and airspace compliance status into a spatiotemporal occupancy risk field for candidate flight segments. Furthermore, it generates prediction uncertainty based on the differences in multiple risk extrapolation results, improving the ability of low-altitude logistics delivery UAVs to predict dynamic flight targets, sudden weather events, and temporary airspace changes. ACOR-World is not used alone for risk prediction; rather, it establishes a mapping relationship between low-altitude environmental conditions, UAV operating conditions, and flight segment risks based on candidate flight segment actions. RINA* does not perform a conventional search on a fixed risk map; instead, it uses a risk discrimination vector composed of the candidate flight segment spatiotemporal occupancy risk field and prediction uncertainty in the correction of node expansion constraints, flight segment risk costs, and heuristic estimation costs. When a high-risk segment requiring reconstruction appears in the candidate route, its local spatiotemporal window is fed back to ACOR-World as a new candidate segment action condition for fine-grained risk simulation, thus forming a closed-loop route optimization process of "action condition risk simulation - risk discrimination vector construction - dynamic risk corridor generation - risk embedding search - local feedback reconstruction".

[0075] 2. This invention constructs a risk discrimination vector based on the spatiotemporal occupancy risk field and prediction uncertainty of candidate routes, and generates a dynamic occupancy risk corridor based on the risk discrimination vector. At the same time, the risk discrimination vector is embedded in the node expansion constraints, route risk costs and heuristic estimation cost corrections of RINA*, so that the risk inference results directly participate in the route search process, thereby reducing ineffective expansion in high-risk areas and improving the safety and efficiency of route search.

[0076] 3. To address the issue of high global replanning overhead in existing methods under dynamic environments, this invention identifies high-risk segments to be reconstructed by integrating risk discrimination values, segment risk cost values, and prediction uncertainties. The local spatiotemporal window of these segments is then fed back to ACOR-World as a new candidate segment action condition for fine-grained risk extrapolation. Local incremental reconstruction is performed while keeping the global risk field outside the window unchanged, thereby reducing the computational overhead of replanning and improving route continuity and mission completion stability. Attached Figure Description

[0077] Figure 1 This is a flowchart illustrating the overall process of the low-altitude logistics delivery drone route optimization method of the present invention.

[0078] Figure 2 This is a schematic diagram of the low-altitude logistics three-dimensional spatiotemporal route map and the generation of dynamically occupied risk pipe corridors according to the present invention;

[0079] Figure 3This is a schematic diagram of the ACOR-World action condition occupancy risk world model structure of the present invention;

[0080] Figure 4 This is a schematic diagram of the RINA* risk perception neural incremental A* search and local feedback reconstruction of the present invention;

[0081] Figure 5 This is a comparison chart of simulation results between the present invention and different route planning methods. Detailed Implementation

[0082] The invention will now be described in detail with reference to specific examples.

[0083] This invention discloses a route optimization method for low-altitude logistics delivery drones based on ACOR-World and RINA*, such as... Figures 1 to 5 As shown, the specific steps include:

[0084] Step 1: Obtain multi-source spatiotemporal data of the low-altitude logistics delivery area, drone operation status, and delivery task constraints.

[0085] Multi-source spatiotemporal data includes obstacle data, dynamic flight target data, low-altitude meteorological data, communication coverage data, restricted airspace data, and refueling node data; UAV operating status includes location, altitude, speed, remaining battery power, payload status, and communication status; delivery mission constraints include delivery origin, delivery destination, mission window, endurance constraints, altitude constraints, refueling continuity constraints, and airspace compliance constraints.

[0086] Step 2: Based on the multi-source spatiotemporal data, UAV operating status and delivery task constraints, construct a three-dimensional spatiotemporal route map for low-altitude logistics and define candidate flight segment actions.

[0087] Step 2.1: Construct a set of low-altitude airway nodes based on the delivery origin, delivery destination, relay waypoint, refueling node, flightable altitude layer and low-altitude airway node, and extend the low-altitude airway nodes in time layer according to a preset time interval to obtain the low-altitude airway nodes in time layer.

[0088] Step 2.2: Based on the spatial distance, altitude layer change, maximum flight speed, maximum climb angle, maximum turn angle, remaining battery power, communication reachability and airspace compliance status between adjacent low-altitude route nodes, determine whether it is permissible to establish a flight connection between adjacent nodes, and form a candidate flight segment set from the node pairs that are allowed to establish a flight connection.

[0089] Step 2.3: Represent the three-dimensional spatiotemporal route map of the low-altitude logistics as follows: ,in, This represents a three-dimensional spatiotemporal flight path map for low-altitude logistics. This represents the set of low-altitude airway nodes after time-layer expansion. Represents the set of candidate flight segments. This represents the set of environmental, meteorological, communication, energy consumption, airspace, and delivery task constraint attributes corresponding to nodes and candidate flight segments.

[0090] Step 2.4: Move the drone from the node on time layer t Fly to adjacent time layer nodes The connection is defined as the candidate segment action, expressed as:

[0091]

[0092] in, Indicates the action of the candidate flight segment. Indicates the amount of change in altitude. This indicates the change in the UAV's state variables after executing the candidate flight segment. The UAV state variables include at least one of the following: position, speed, remaining battery power, payload status, communication status, and mission status.

[0093] Step 3: Call the pre-trained Action Condition Occupation Risk World Model (ACOR-World) to deduce the spatiotemporal occupancy risk field of candidate flight segments, and generate prediction uncertainty based on the differences in multiple risk deduction results. ACOR-World is an Action Condition Occupation Risk World Model constructed for low-altitude logistics 3D spatiotemporal route maps and candidate flight segment actions. It is used to deduce the local low-altitude state changes after executing candidate flight segments based on the current low-altitude environmental state, UAV operating status, delivery task constraints, and candidate flight segment actions, and output the spatiotemporal occupancy risk field and prediction uncertainty corresponding to the candidate flight segments.

[0094] The "Action Condition Occupied Risk World Model ACOR-World" includes a state encoding unit, an action condition residual derivation unit, a risk decoding unit, and an uncertainty estimation unit.

[0095] Step 3.1: The state coding unit generates a representation of the current low-altitude world state based on the current multi-source spatiotemporal data, the three-dimensional spatiotemporal route map of low-altitude logistics, and the UAV's operational status.

[0096]

[0097] in, This represents the current low-altitude world state at time level t. This represents the current multi-source spatiotemporal data. This represents a three-dimensional spatiotemporal flight path map for low-altitude logistics. Indicates the operating status of the drone. This represents a state coding mapping.

[0098] Step 3.2: The residual derivation unit for action conditions uses candidate flight segment actions. Given the condition, predict the local state residuals after executing the candidate flight segment:

[0099]

[0100] And obtain the future world state representation corresponding to the candidate flight segment:

[0101]

[0102] in, This represents the local state residual caused by the actions of the candidate flight segment. This represents the mapping of action condition residuals. This represents the future world state after executing the candidate flight segment.

[0103] Step 3.3: The risk decoding unit outputs the spatiotemporal occupancy risk field of the candidate flight segment based on the future world state representation corresponding to the candidate flight segment:

[0104]

[0105] in, Indicates candidate flight segments The corresponding spacetime occupancy risk field, This represents the risk decoding mapping. The spatiotemporal occupancy risk field consists of future occupancy risk, weather disturbance risk, changes in energy consumption margin, communication loss risk, and airspace compliance status, and is represented as:

[0106]

[0107] in, This indicates the risk of future occupancy. Indicates the risk of weather disturbances. Indicates changes in energy margin. This indicates a risk of communication failure. This indicates the airspace compliance status.

[0108] Step 3.4: The uncertainty estimation unit calculates the difference between multiple risk decoding results for the same candidate flight segment, generating the prediction uncertainty:

[0109]

[0110] in, Indicates candidate flight segments The corresponding prediction uncertainty, where M represents the number of risk simulations, This represents the result of the m-th risk simulation. This represents the average of multiple risk simulation results.

[0111] Step 3.5: ACOR-World is pre-trained using historical low-altitude flight data, simulated low-altitude environment data, and mission execution data. Training samples include the current low-altitude environment state, UAV operating status, delivery mission constraints, candidate flight segment actions, and risk status labels after candidate flight segment execution. Risk status labels include future occupancy labels, weather disturbance labels, energy margin labels, communication loss labels, and airspace compliance labels. The training loss is expressed as:

[0112]

[0113] in, This represents the local state residual prediction error. Indicates risk decoding error. This indicates the classification error of airspace compliance status. This represents the prediction uncertainty calibration error. After training, ACOR-World is used for forward inference during route optimization, and retraining is not required for each route optimization.

[0114] Step 4: Generate a dynamic occupancy risk corridor based on the spatiotemporal occupancy risk field and prediction uncertainty of candidate flight segments.

[0115] Step 4.1: Construct the risk discrimination vector corresponding to the candidate flight segment based on the spatiotemporal occupancy risk field and prediction uncertainty:

[0116]

[0117] in, Indicates candidate flight segments The corresponding risk discrimination vector, This represents the uncertainty of the forecast. The risk discrimination vector includes future occupancy risk, weather disturbance risk, energy consumption margin, communication loss risk, airspace compliance status, and forecast uncertainty.

[0118] Among them, the future occupancy risk is determined based on the probability of overlap between the predicted trajectory of dynamic flight targets or obstacles and the spatiotemporal occupancy area of ​​the candidate flight segment; the weather disturbance risk is determined based on the wind speed, wind direction, rainfall intensity, and the wind resistance capability of the UAV in the area traversed by the candidate flight segment; the energy consumption margin is determined based on the relationship between the expected remaining power after executing the candidate flight segment and the energy required for subsequent tasks; the communication loss risk is determined based on the communication coverage strength, obstruction status, and link interruption probability; the airspace compliance status is determined based on whether the candidate flight segment meets the no-fly, restricted-fly, and temporary control rules; and the prediction uncertainty is determined based on the differences in the results of multiple simulations by ACOR-World.

[0119] Step 4.2: Construct a comprehensive risk discriminant value based on the risk discriminant vector, expressed as:

[0120]

[0121] in, Indicates candidate flight segments The corresponding comprehensive risk assessment value, This represents the risk discrimination mapping, which is used to map the risk discrimination vector consisting of the candidate segment spatiotemporal occupancy risk field and prediction uncertainty into a comprehensive risk discrimination value.

[0122] Construct a flightability discriminant function based on the risk discriminant vector:

[0123]

[0124] in, This indicates the flightability determination result for the candidate flight segment. Indicates an indicator function, Indicates the safe threshold for flight. This indicates that the preset airspace rules are met.

[0125] Step 4.3: [The sentence is incomplete and requires more context to be translated accurately.] The candidate segments are identified as flyable candidate segments, resulting in a set of flyable candidate segments. Those that do not meet the criteria... Candidate routes were removed from the subsequent route search scope:

[0126]

[0127] in, This represents the set of flyable candidate flight segments at time level t. Indicates that by node To node Candidate flight segments.

[0128] Step 4.4: Construct a continuous connected subgraph from the flyable candidate air segments and their connected low-altitude airway nodes, and identify the continuous connected subgraph as the dynamic occupancy risk corridor:

[0129]

[0130] in, This indicates the risk of dynamic occupancy of the utility tunnel. This represents the set of low-altitude airway nodes that connect to flyable candidate air segments. This represents the set of flightable candidate routes. The dynamic occupancy risk corridor is used to limit the node expansion range of subsequent RINA*, ensuring that route searches are conducted only within connected areas that meet the conditions of risk safety, communication reachability, energy reachability, and airspace compliance.

[0131] Step 5: Generate candidate routes within the dynamically occupied risk corridor based on RINA*.

[0132] Step 5.1: Using the node corresponding to the delivery origin as the initial expansion node, in the dynamically occupied risk corridor... Internal extraction and current node Connectable flightable candidate segments And will not satisfy the flightability criterion function. Candidate segments are excluded from the extended set.

[0133] Step 5.2: Extend to nodes Candidate segment establishment node extended evaluation function:

[0134]

[0135] in, Indicates expansion to nodes The node evaluation value, This indicates the distance from the delivery origin to the current node. The cumulative value of passage This represents the risk value of a flight segment obtained by mapping the risk discrimination vector. Indicates the current extension node to the delivery destination. Heuristic cost estimation.

[0136] Among them, the risk value of flight segments The risk cost mapping is obtained from the risk discrimination vector and is used to characterize the passage risk of candidate air segments under the combined effects of future occupancy, weather disturbances, energy consumption margin, communication loss, airspace compliance and prediction uncertainty.

[0137] The heuristic cost estimation is determined based on the shortest reach time from the current expansion node to the delivery destination and the minimum remaining energy requirement:

[0138]

[0139] in, This represents the shortest reachable time from the current extended node to the delivery destination under the maximum allowable flight speed constraint. Indicates the remaining allowable time for the delivery task. This represents the minimum remaining energy consumption requirement from the current expansion node to the delivery destination under a preset safety reference state. This indicates the drone's current remaining available energy.

[0140] Step 5.3: Add candidate nodes that meet the requirements of the node expansion evaluation function to the open set, and expand them in order of increasing evaluation value; when the search reaches the delivery destination or the task termination condition is met, backtrack the parent node relationship to generate candidate routes. .

[0141] Step 5.4: Calculate the comprehensive transit cost of the candidate routes; wherein, the segment risk cost of the candidate route is obtained by mapping the risk cost according to the risk discrimination vector, and is expressed as:

[0142]

[0143] in, Indicates candidate flight segments The risk-value of the flight segment This represents a risk-cost mapping. This represents the risk discrimination vector composed of the candidate flight segment spatiotemporal occupancy risk field and prediction uncertainty;

[0144] Candidate routes The comprehensive transit value is expressed as:

[0145]

[0146] in, Indicate candidate routes The comprehensive passage value is used to characterize the overall passage risk of a candidate route under the combined constraints of future occupancy risk, weather disturbance risk, changes in energy margin, communication loss risk, airspace compliance status, and forecast uncertainty.

[0147] Step 6: Calculate the risk contribution of flight segments and perform local incremental reconstruction.

[0148] Step 6.1: Target candidate routes any candidate segment The risk value of the candidate flight segment Replace with preset benchmark value The risk value of the remaining candidate routes remains unchanged, and a counterfactual passage value is constructed:

[0149]

[0150] in, Indicates the candidate flight segment Constructed counterfactual common value, This indicates the baseline value of the candidate flight segment under the preset safety reference state.

[0151] The preset safety reference state is a state where the risk of future occupancy is low, the risk of weather disturbance is low, the risk of communication loss is low, and the airspace compliance status meets the preset airspace rules while retaining basic flight energy consumption.

[0152] Step 6.2: Calculate the risk contribution of candidate routes based on the difference between the comprehensive transit value and the counterfactual transit value of the original candidate routes.

[0153]

[0154] in, Indicates the risk contribution of candidate flight segments. It is a stable term.

[0155] Alternatively, in another implementation, the risk contribution of a flight segment can be determined using a normalization method:

[0156]

[0157] in, This indicates that candidate segment e is on the candidate route. The normalized risk contribution in the data, e′ represents the candidate route. Any candidate segment in the, It is a stable term.

[0158] Step 6.3: When the risk contribution of a candidate segment is greater than the risk contribution threshold, or the prediction uncertainty corresponding to the candidate segment is greater than the uncertainty threshold, the candidate segment is identified as a high-risk segment to be reconstructed.

[0159] Step 6.4: Construct a local spatiotemporal window based on the spatial range of high-risk flight segments to be reconstructed, the expected passage time layer, and the scope of risk propagation:

[0160]

[0161] in, Represents a local spatiotemporal window. Represents a node The spatial distance of the highest-risk flight segments awaiting reconstruction. Indicates a local spatial range. This indicates a local time range. The local spatial range and local time range are dynamically adjusted based on the flight segment length, risk level, and UAV flight speed, and are used to cover high-risk flight segments to be reconstructed and their potentially affected adjacent candidate flight segments.

[0162] Step 6.5: Feed the local spatiotemporal window as a new candidate flight segment action condition back to ACOR-World, and only perform fine-grained updates on the risk field of the candidate flight segment within the local spatiotemporal window, while keeping the global risk field outside the window unchanged.

[0163] Step 6.6: RINA* performs incremental reconstruction of the dynamically occupied risk corridors within the local spatiotemporal window to obtain a set of local alternative flight segments. Then, connect it with the preceding and subsequent segments outside the window in the original candidate route to generate the target route:

[0164]

[0165] in, Indicates the target route. This represents the set of preceding segments outside the window. This represents the set of local alternative flight segments obtained from the reconstruction. This indicates the set of subsequent flight segments outside the window.

[0166] To ensure the feasibility of the target route, a set of local alternative routes is needed. The entry node is consistent with the end node of the preceding segment in terms of spatial location, altitude layer, and time layer, and the exit node is consistent with the beginning node of the subsequent segment in terms of spatial location, altitude layer, and time layer, thereby ensuring the temporal continuity and flight continuity of the target route after reconstruction.

[0167] In one embodiment, the present invention can also be implemented through a low-altitude logistics scheduling platform, an edge computing node, or a UAV ground control station. The system includes a multi-source spatiotemporal data acquisition module, a three-dimensional spatiotemporal route map construction module, an action condition risk deduction module, a dynamic occupancy risk corridor generation module, a risk perception route search module, a local incremental reconstruction module, and a result output module.

[0168] To verify the effectiveness of the method of this invention, simulation experiments were conducted in a set of typical urban last-mile low-altitude logistics delivery scenarios. The simulation area was set as an urban low-altitude delivery area with a horizontal range of 2km × 2km and an altitude range of 0m to 120m, divided into altitude layers at 10m intervals. The maximum horizontal speed of the UAV was set to 15m / s, the maximum climb angle to 25°, the maximum turning angle to 45°, the initial battery level to 100%, the payload to 1.5kg, and the time step to [missing information]. Take 5 seconds.

[0169] The number of dynamic flight targets is set to 20 to 40, with their initial positions uniformly distributed within the delivery area, random movement directions, and uniformly sampled speeds ranging from 3 m / s to 12 m / s; weak communication coverage areas are randomly deployed as circular areas with a radius of 50 m to 100 m, with their center positions uniformly distributed within the area; temporary no-fly zones are randomly generated as rectangular areas, accounting for 5% to 15% of the total delivery area; three power replenishment nodes are set, and the dynamic meteorological disturbance area includes conventional wind fields and local gust areas, with a maximum wind speed set to 10 m / s.

[0170] The comparison methods include the traditional A* algorithm, the static risk map A* algorithm, the neural A* algorithm, and the method of this invention. Evaluation metrics include comprehensive transit cost, initial planning time, average adjustment time after dynamic disturbance, number of high-risk segments, number of global replanning attempts, and mission completion rate. The average adjustment time after dynamic disturbance represents the average time of a single global replanning attempt after a dynamic disturbance is triggered for the traditional A* algorithm, the static risk map A* algorithm, and the neural A* algorithm; for the method of this invention, it represents the average time of a single local incremental reconstruction attempt after a dynamic disturbance is triggered. The results in the table are the average values ​​from 50 independent Monte Carlo simulation runs.

[0171] Table 1 Comparison of Simulation Results for Different Route Planning Methods

[0172] Traditional A* algorithm 1.00 2.18 1.45 9.6 4.6 82.4 Static Risk Map A* Algorithm 0.83 2.74 1.52 6.8 3.2 87.6 Neural A* Algorithm 0.76 3.21 1.38 5.9 2.4 90.8 Method of the present invention 0.61 2.45 0.74 3.2 0.3 96.2

[0173] As shown in Table 1, the traditional A* algorithm has a shorter initial planning time, but it does not consider future risk changes after candidate segment actions, resulting in a higher number of high-risk segments and global replanning times. The static risk map A* algorithm can reduce some risk segments, but its risk cost is mainly based on the current state, requiring frequent replanning under dynamic flight targets, sudden weather changes, and temporary airspace control changes. The neural A* algorithm can improve search performance, but lacks a mechanism for action condition feedback and reconstruction for locally high-risk segments. The initial planning time of the method in this invention is slightly higher than that of the traditional A* algorithm, but the average adjustment time after dynamic disturbances is significantly reduced, and the number of global replanning times is significantly decreased. This indicates that dynamic occupancy risk corridor constraint search and local incremental reconstruction can reduce the replanning overhead in dynamic environments.

[0174] Table 2 Simulation results of the method of the present invention under different dynamic disturbance intensities

[0175] low disturbance 20 0.48 1.7 0.8 0.56 98.1 medium disturbance 30 0.61 3.2 1.4 1.04 96.2 High disturbance 40 0.74 5.8 2.2 1.85 94.8

[0176] As shown in Table 2, with the increase in the number of dynamic flight targets and the intensity of environmental disturbances, the average comprehensive transit cost, the number of high-risk flight segments, and the number of local reconstructions all increase. In high-disturbance scenarios, due to the expansion of the local spatiotemporal window as the risk propagation range expands, the average total time for local reconstruction increases to 1.85s. However, the method of this invention can still maintain a task completion rate of 94.8%, indicating that it can locate high-risk flight segments to be reconstructed by the risk contribution of flight segments and the prediction uncertainty, and improve the route resilience in dynamic environments through incremental reconstruction using local spatiotemporal windows.

[0177] Figure 5 This is a comparison chart of simulation results between the present invention and different route planning methods. The horizontal axis represents different route planning methods, and the vertical axis represents the overall transit cost, the number of high-risk segments, the average adjustment time after dynamic disturbances, and the mission completion rate, respectively. Figure 5It can be clearly seen that the method of the present invention significantly reduces the average adjustment time after dynamic disturbances while reducing the overall transit cost and the number of high-risk flight segments, and improves the mission completion rate. This shows that the method of the present invention has good comprehensive performance in terms of safety, dynamic response capability and mission stability.

[0178] The experimental results show that the method of the present invention is applicable to low-altitude logistics delivery scenarios with dynamic flight targets, weather disturbances, weak communication coverage, and temporary airspace control. Through candidate segment action condition risk deduction, risk discrimination vector construction, dynamic occupancy risk corridor generation, RINA* risk embedding search, and local incremental reconstruction, it can effectively reduce the number of high-risk segments, the number of global replanning times, and the comprehensive passage cost, thereby improving the safety and mission completion rate of low-altitude logistics delivery UAV route optimization.

[0179] The above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent substitutions, conventional modifications, or combinations made based on the technical concept of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for optimizing the flight path of low-altitude logistics delivery drones based on ACOR-World and RINA*, characterized in that, Includes the following steps: Step 1: Acquire multi-source spatiotemporal data of the low-altitude logistics delivery area, drone operation status, and delivery task constraints; Step 2: Based on the multi-source spatiotemporal data, UAV operating status and delivery task constraints, construct a three-dimensional spatiotemporal route map for low-altitude logistics that includes spatial location, altitude layer, time layer, candidate flight segments and constraint attributes, and define the connection between adjacent time layer nodes that satisfies flight performance and task constraints as candidate flight segment actions; Step 3: Call the pre-trained Action Condition Occupation Risk World Model ACOR-World. Based on the current low-altitude environment, UAV operating status, delivery task constraints, and candidate flight segment actions, deduce the spatiotemporal occupancy risk field of candidate flight segments, and generate prediction uncertainty based on the differences in multiple risk deduction results. The spatiotemporal occupancy risk field includes future occupancy risk, weather disturbance risk, energy consumption margin change, communication loss risk, and airspace compliance status. Step 4: Construct a risk discrimination vector based on the spatiotemporal occupancy risk field and prediction uncertainty, perform preset flight condition judgment on nodes and candidate flight segments, and determine the continuously connected flightable nodes and candidate flight segments as dynamic occupancy risk corridors; Step 5: Construct the risk-aware neural incremental A* search algorithm RINA*, expand the flight segment within the dynamic occupancy risk corridor, introduce the spatiotemporal occupancy risk field and prediction uncertainty into the node expansion constraints, flight segment risk cost and heuristic estimation cost correction, generate candidate routes, and calculate the comprehensive passage cost of the candidate routes.

2. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 1, characterized in that, The risk contribution of flight segments is also calculated during the candidate route generation process. Based on the risk contribution and prediction uncertainty of the aforementioned flight segments, high-risk flight segments to be reconstructed are identified. Their local spatiotemporal windows are used as new candidate flight segment action conditions and fed back to ACOR-World for fine-grained risk simulation. RINA* then performs incremental reconstruction of the local dynamic occupancy risk corridor without changing the global risk field outside the window, and finally outputs the target route and the corresponding comprehensive passage value.

3. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 1, characterized in that, The multi-source spatiotemporal data includes obstacle data, dynamic flight target data, low-altitude meteorological data, communication coverage data, restricted airspace data, and refueling node data; the UAV operating status includes position, altitude, speed, remaining battery power, payload status, and communication status; the delivery task constraints include delivery start point, delivery end point, mission window, endurance constraints, altitude layer constraints, refueling continuity constraints, and airspace compliance constraints.

4. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 1, characterized in that, The construction of the three-dimensional spatiotemporal route map for low-altitude logistics includes the following steps: Step 2.1: Based on the delivery origin, delivery destination, relay waypoint, refueling node, flightable altitude layer and low-altitude route node, construct a set of low-altitude route nodes, and perform time extension on the low-altitude route nodes according to a preset time interval to obtain the low-altitude route nodes in the time layer. Step 2.2: Based on the spatial distance, altitude layer change, maximum flight speed, maximum climb angle, maximum turn angle, remaining battery power, communication reachability and airspace compliance status between adjacent low-altitude route nodes, determine whether it is permissible to establish a flight connection between adjacent nodes, and construct a candidate flight segment set from the nodes that are permitted to establish a flight connection. Step 2.3: The three-dimensional spatiotemporal route map for low-altitude logistics includes a set of low-altitude route nodes, a set of candidate routes, and a set of environmental, meteorological, communication, energy consumption, airspace, and delivery task constraint attributes corresponding to the nodes and candidate routes; Step 2.4: Move the drone from the node on time layer t Fly to adjacent time layer nodes The connection is defined as a candidate flight segment action, including the change in altitude layer, and at least one change in the UAV's position, speed, remaining battery power, payload status, communication status, and mission status after executing the candidate flight segment.

5. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 1, characterized in that, The ACOR-World includes a state encoding unit, an action condition residual derivation unit, a risk decoding unit, and an uncertainty estimation unit; The state encoding unit extracts low-altitude route node features, candidate segment constraint attributes, environmental disturbance features, and UAV state features based on current multi-source spatiotemporal data, low-altitude logistics 3D spatiotemporal route maps, and UAV operational status. It then fuses and encodes the spatial connectivity, temporal connectivity, and candidate segment attributes between the low-altitude route nodes to generate a current low-altitude world state representation. ; The action condition residual derivation unit uses candidate flight segment actions Given the conditions, based on the origin and destination nodes, flight time layer, altitude layer changes, expected energy consumption changes, and constraint attributes of the candidate flight segment, predict the local state residuals of the occupancy status, weather disturbance impact, energy consumption margin, communication reachability, and airspace compliance status generated in the corresponding local spatiotemporal neighborhood after executing the candidate flight segment. The local state residual is then compared with the current low-altitude world state. By merging, we obtain the future world state representation corresponding to the candidate flight segments. ; The risk decoding unit represents the future world state. The risk decoding mapping outputs the spatiotemporal occupancy risk field of candidate flight segments. This includes risks such as future occupancy, weather disturbances, changes in energy consumption margin, communication loss, and airspace compliance status. The uncertainty estimation unit generates the prediction uncertainty based on the differences between the multiple risk decoding results of the risk decoding unit: ； in, Indicates candidate flight segments The corresponding prediction uncertainty, where M represents the number of risk simulations, This represents the result of the m-th risk simulation. This represents the average of multiple risk simulation results.

6. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 5, characterized in that, The dynamic occupancy risk corridor is generated based on the spatiotemporal occupancy risk field of candidate flight segments and the prediction uncertainty, including the following steps: Step 4.1: Construct the risk discrimination vector corresponding to the candidate flight segment based on the spatiotemporal occupancy risk field and prediction uncertainty. The risk discrimination vector includes future occupancy risk, weather disturbance risk, energy consumption margin, communication loss risk, airspace compliance status, and prediction uncertainty. Step 4.2: Construct a comprehensive risk discrimination value based on the risk discrimination vector, and construct a flyability discrimination function based on the comprehensive risk discrimination value: in, This indicates the flightability determination result for the candidate flight segment. Indicates an indicator function, This represents the comprehensive risk discrimination value obtained from the risk discrimination vector through the risk discrimination mapping. Indicates the safe threshold for flight. This indicates that the preset airspace rules are met; Step 4.3: [The sentence is incomplete and requires more context to be translated accurately.] The candidate segments are identified as flyable candidate segments, thus obtaining a set of flyable candidate segments; Step 4.4: Construct a continuous connected subgraph of the flightable candidate air segments and their connected low-altitude airway nodes, and identify the continuous connected subgraph as a dynamic occupancy risk corridor.

7. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 6, characterized in that, The RINA* performs segment expansion and candidate route generation within the dynamically occupied risk corridor, including the following steps: Step 5.1: Using the node corresponding to the delivery origin as the initial expansion node, extract the node corresponding to the current node within the dynamically occupied risk corridor. Connectable flightable candidate segments And exclude candidate flight segments that do not meet the flightability criteria from the extended set; Step 5.2: Extend to nodes Candidate segment establishment node extended evaluation function: ； in, Indicates expansion to nodes The node evaluation value, This indicates the distance from the delivery origin to the current node. The cumulative value of passage Indicates candidate flight segments The risk-value of the flight segment Indicates the current extension node to the delivery destination. Heuristic cost estimation The risk cost of the flight segment is obtained by mapping the risk cost based on the risk discrimination vector: in, This represents a risk cost mapping, where the risk cost of a flight segment is used to characterize the passage risk of a candidate flight segment under the combined effects of future occupancy risk, weather disturbance risk, changes in energy margin, communication loss risk, airspace compliance status, and prediction uncertainty. Step 5.3: Add candidate nodes that meet the requirements of the node expansion evaluation function to the open set, and expand them in order of increasing evaluation value until the delivery destination is reached or the task termination condition is met. Backtrack the parent node relationship to generate candidate routes. ; Step 5.4: Calculate the comprehensive transit cost of the candidate routes, which is the candidate route. The sum of the risk and value of each flight segment.

8. The method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 2, characterized in that, The risk contribution of the aforementioned flight segment is calculated as follows: Step 6.1: Target candidate routes Any candidate segment The risk value of the candidate flight segment Replace with preset benchmark value The risk value of the remaining candidate routes remains unchanged, and a counterfactual passage value is constructed: ； in, Indicates the candidate flight segment Constructed counterfactual common value, This indicates the baseline value of the candidate flight segment under a preset safety reference state; the preset safety reference state is a state in which the risk of future occupancy is low, the risk of weather disturbance is low, the risk of communication loss is low, the airspace compliance status meets the preset airspace rules and retains basic flight energy consumption. Step 6.2: Calculate the risk contribution of candidate routes based on the difference between the comprehensive transit value and the counterfactual transit value of the original candidate routes. ； in, Indicates the risk contribution of candidate flight segments. As a stable term, Indicate candidate routes The comprehensive value of accessibility; Alternatively, the risk contribution of the aforementioned flight segment can be determined using a normalized method, expressed as: in, This indicates that candidate segment e is on the candidate route. Normalized risk contribution in This represents the segment risk value of candidate segment e. It is a stable term.

9. A method for optimizing the flight path of a low-altitude logistics delivery UAV based on ACOR-World and RINA* as described in claim 8, characterized in that, The locally dynamically occupied risk corridor will undergo incremental reconstruction, as detailed below: Step 7.1: When the risk contribution of the candidate flight segment is greater than the risk contribution threshold, or the prediction uncertainty corresponding to the candidate flight segment is greater than the uncertainty threshold, the candidate flight segment is identified as a high-risk flight segment to be reconstructed. Step 7.2: Construct a local spatiotemporal window based on the spatial extent of high-risk flight segments to be reconstructed, the expected transit time layer, and the scope of risk propagation: ； in, Represents a local spatiotemporal window. Represents a node The spatial distance of the highest-risk flight segments awaiting reconstruction Indicates a local spatial range. Indicates a local time range; Step 7.3: Feed back the local spatiotemporal window as a new candidate flight segment action condition to ACOR-World, and only perform fine-grained updates on the risk field within the local spatiotemporal window, while keeping the global risk field outside the window unchanged; Step 7.4: RINA* performs incremental reconstruction of the dynamically occupied risk corridor within the local spatiotemporal window to obtain a set of local alternative flight segments. It is then connected with the preceding and subsequent segments outside the window in the original candidate routes to generate the target route. .

10. A system based on the low-altitude logistics delivery UAV route optimization method based on ACOR-World and RINA* as described in any one of claims 1 to 9, characterized in that, include: The multi-source spatiotemporal data acquisition module is used to acquire multi-source spatiotemporal data of the low-altitude logistics delivery area, the drone's operating status, and delivery task constraints; The three-dimensional spatiotemporal route map construction module is used to construct a three-dimensional spatiotemporal route map for low-altitude logistics based on the multi-source spatiotemporal data, UAV operating status and delivery task constraints, and define candidate flight segment actions; The Action Condition Risk Inference Module is used to call the pre-trained Action Condition Occupation Risk World Model ACOR-World. Based on the current low-altitude environment, UAV operating status, delivery task constraints and candidate flight segment actions, it infers the spatiotemporal occupancy risk field of candidate flight segments and generates prediction uncertainty based on the differences in multiple risk inference results. The dynamic occupancy risk corridor generation module is used to construct a risk discrimination vector based on the spatiotemporal occupancy risk field and prediction uncertainty of the candidate flight segment, to judge the flightability of nodes and candidate flight segments, and to generate a dynamic occupancy risk corridor. The risk perception route search module is used to call the risk perception neural incremental A* search algorithm RINA* to expand the route within the dynamically occupied risk corridor, generate candidate routes, and calculate the comprehensive passage cost and route risk contribution. The local incremental reconstruction module is used to identify high-risk segments to be reconstructed based on the risk contribution and prediction uncertainty of the segments. Its local spatiotemporal window is used as the new candidate segment action conditions and fed back to ACOR-World for fine-grained risk simulation. RINA* then performs incremental reconstruction of the local dynamic occupancy risk corridor without changing the global risk field outside the window. The results output module is used to output the target route and its corresponding comprehensive transit value.

Images