Method and system for avoiding mutual interference of multiple unmanned vehicles based on priority and spatiotemporal windows

Abstract

This paper describes a method and system for multi-route autonomous vehicle mutual avoidance based on priority and spatiotemporal windows. This system belongs to the field of intelligent path planning technology, specifically to the field of intelligent collision avoidance technology for autonomous vehicles. It addresses the technical issue that existing autonomous vehicle mutual avoidance methods can lead to collisions in environments where multiple autonomous vehicles are operating simultaneously. By assigning priorities and planning spatiotemporal windows, autonomous vehicles can efficiently avoid each other in a shared workspace, significantly reducing the probability of collisions and improving system safety.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent path planning technology, and specifically relates to the field of intelligent collision avoidance technology for unmanned vehicles. Background Art

[0002] Multi-UGV systems have been widely adopted in various fields. Their core technologies rely on collaborative control, communication networks, and intelligent algorithms. Examples of their application include warehouse cargo handling (e.g., AGV swarms), port container transport, and last-mile delivery. Amazon uses thousands of AGVs to collaboratively complete order sorting, increasing efficiency by 3-4 times. This demonstrates how the application of multi-UGV systems can significantly improve work efficiency in related industries.

[0003] The main technical focus of existing unmanned vehicle mutual avoidance methods is on obstacle detection, emphasizing the use of sensors to perceive the environment around the unmanned vehicle, determine obstacles, and then plan paths to enable the unmanned vehicle to avoid obstacles. This unmanned vehicle obstacle avoidance method may cause path intersection or collision between unmanned vehicles in an environment where multiple unmanned vehicles are operating simultaneously. Summary of the Invention

[0004] In order to solve the technical problem that the existing unmanned vehicle mutual avoidance method may cause path intersection or collision between unmanned vehicles in an environment where multiple unmanned vehicles are operating simultaneously, the present invention provides a multi-unmanned vehicle mutual avoidance method based on priority and space-time window.

[0005] The method comprises the following steps:

[0006] S1. Initialize a map of the multi-unmanned vehicle system, which includes road conditions and the position and speed of each unmanned vehicle in the multi-unmanned vehicle system;

[0007] S2. assigning a priority to each unmanned vehicle in the multi-unmanned vehicle system;

[0008] S3, planning an initial path for each unmanned vehicle in the multi-unmanned vehicle system;

[0009] S4. Based on the initial planned path, determine the predicted future time scale If no unmanned vehicle collision occurs, each unmanned vehicle will follow the initial path. time; if a collision occurs, the unmanned vehicle path is replanned, and then each unmanned vehicle drives according to the updated planned path time;

[0010] S5. Determine whether each unmanned vehicle has reached the destination. If any unmanned vehicle has not reached the destination, the remaining path of the unmanned vehicle is regarded as the initial path, and the process returns to step S4 to continue. If any unmanned vehicle has reached the destination, the unmanned vehicle's journey ends.

[0011] Furthermore, the rules for assigning priorities are as follows:

[0012] First, emergency priority: vehicles approaching the collision threshold have higher priority;

[0013] Secondly, critical tasks are prioritized: vehicles performing rescue and firefighting tasks have higher priority;

[0014] Again, traffic rules take precedence: vehicle priority is determined based on road right rules.

[0015] Furthermore, an initial path is planned for each unmanned vehicle in the multi-unmanned vehicle system. Specifically, the starting point and end point of each unmanned vehicle are set in the multi-unmanned vehicle system map. According to the road conditions in the map, the initial path from the starting point to the end point is planned for each unmanned vehicle with the principle of shortest driving distance and no collision.

[0016] Further, judging the future time scale When determining whether an unmanned vehicle collision will occur, any unmanned vehicle in the unmanned vehicle system is taken as an object and judged one by one whether it has collided with other unmanned vehicles. Specifically:

[0017] S41. Setting the future time scale for predictions , and Divide into share, ,make , predicting that driverless cars will Position at the moment;

[0018] S42, determine whether any two unmanned vehicles are Whether the distance at the moment is less than or equal to the safe distance;

[0019] S43. If the distance is less than or equal to the safety distance, the predicted collision position is recorded, and then the process proceeds to step S45.

[0020] S44: If it is greater than the safety distance, proceed to step S45;

[0021] S45. Judgment and relationship, if , then another , then return to step S42; if If there is a recorded predicted collision location, then it is determined that If If there is no recorded predicted collision location, then it is determined that No collisions between autonomous vehicles will occur.

[0022] Furthermore, the safety distance refers to the minimum distance between the two unmanned vehicles that can continue to travel without colliding, determined based on the driving paths and current speeds of the two unmanned vehicles.

[0023] Furthermore, the re-planning of the unmanned vehicle path is specifically as follows:

[0024] S51. Update the multi-autonomous vehicle system map and add the predicted collision location to the map;

[0025] S52, based on the updated multi-unmanned vehicle system map, using the improved Pied Kingfisher algorithm to regenerate the planned path of the low-priority unmanned vehicle;

[0026] S53. Update the multi-unmanned vehicle system map and delete the predicted collision location.

[0027] Furthermore, the Pied Kingfisher algorithm is divided into three stages, namely the perching and hovering stage, the diving strategy stage, and the symbiotic stage;

[0028] The improved Pied Kingfisher algorithm is specifically as follows:

[0029] During the population initialization process before the perching and hovering phases, a multi-element initial population strategy was introduced;

[0030] In the symbiotic stage, the nonlinear energy attenuation factor E of the improved Golden Leopard optimization algorithm is introduced;

[0031] The joiner strategy of the sparrow search algorithm is introduced in the symbiotic stage.

[0032] Furthermore, after introducing the multivariate initial population strategy, the expression for population initialization becomes: ;

[0033] in, It represents the Individuals in the The specific location in the dimension, Indicates the The number of chaotic sequences, and It represents the upper and lower bounds of the population search space.

[0034] Furthermore, the nonlinear energy attenuation factor of the Golden Leopard optimization algorithm Specifically: ;in, Indicates the current iteration number, Indicates the maximum number of iterations;

[0035] After introducing the joiner strategy of the sparrow search algorithm, the updated position of the Pied Kingfisher is obtained by Obtain, among which, Indicates the updated The location of the Pied Kingfisher, and represents the positions of two individuals randomly selected from the population, represents the normal distribution parameter, Indicates the number before the update The location of the Pied Kingfisher, For Random values ​​are uniformly distributed within this interval. Indicates the hunting efficiency of the Pied Kingfisher. Indicates the specific location of the optimal solution at the current time; ,in, Indicates that its elements are randomly assigned or of matrix, for The transposed matrix of Indicates that an element is of matrix.

[0036] The present invention also provides a multi-unmanned vehicle mutual avoidance system based on priority and spatiotemporal windows, the system comprising:

[0037] A module that initializes a map of the multi-route vehicle system: the map includes road conditions and the position and speed of each vehicle in the multi-route vehicle system;

[0038] A module that assigns priorities to each unmanned vehicle in a multi-unmanned vehicle system;

[0039] Path planning module: plans the initial path for each unmanned vehicle in the multi-unmanned vehicle system, and judges the predicted future time scale based on the planned initial path. If no unmanned vehicle collision occurs, each unmanned vehicle will follow the initial path. time; if a collision occurs, the path is replanned, and then each unmanned vehicle drives according to the updated planned path time;

[0040] The module for determining the destination is used to determine whether each unmanned vehicle has reached the destination. If any unmanned vehicle has not reached the destination, the remaining path of the unmanned vehicle is regarded as the initial path, and the module returns to the path planning module to continue. If any unmanned vehicle has reached the destination, the driving of the unmanned vehicle ends.

[0041] The beneficial effects of the method of the present invention are:

[0042] Significantly reduce the risk of collisions between autonomous vehicles: Through priority allocation and spatiotemporal window planning, autonomous vehicles can efficiently avoid each other in a shared workspace, significantly reducing the probability of collision and improving system safety.

[0043] Improve the operational efficiency of multi-unmanned vehicle systems: High-priority unmanned vehicles have priority passing through conflict areas, and low-priority unmanned vehicles avoid conflicts by dynamically adjusting their paths or speeds, reducing waiting time and path redundancy, and improving overall mission execution efficiency.

[0044] Real-time adaptability to dynamic environments: Supports real-time conflict detection and dynamic adjustment, and can quickly respond to environmental changes (such as new obstacles or other changes in the status of other unmanned vehicles), ensuring the robustness of the system in complex environments.

[0045] Optimize resource utilization: Through collaborative planning of priorities and spatiotemporal windows, competition and waste of resources (such as paths and work areas) are avoided, and the resource allocation and utilization efficiency of the multi-unmanned vehicle system are optimized.

[0046] Supports large-scale collaborative work of unmanned vehicles: Suitable for large-scale multi-unmanned vehicle systems, it can efficiently manage the path planning and conflict avoidance of dozens or even hundreds of unmanned vehicles, and is suitable for scenarios such as warehousing and logistics, and intelligent manufacturing.

[0047] Reduce system complexity and computational cost: Through local optimization of priority and spatiotemporal windows, the dependence on global path planning is reduced, the computational complexity and communication overhead are reduced, and it is suitable for distributed computing and real-time control systems.

[0048] Improve mission completion rate: Through efficient conflict avoidance and path optimization, unmanned vehicles can complete tasks more reliably and reduce mission interruptions or failures caused by path conflicts or collisions.

[0049] Enhanced system flexibility and scalability: Supports dynamic priority adjustment and spatiotemporal window optimization, can adapt to different mission requirements and environmental changes, and is easy to expand to more unmanned vehicles or more complex application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is an overall flow chart of the method for avoiding each other of unmanned vehicles based on priority and spatiotemporal windows according to an embodiment of the present invention;

[0051] Figure 2 Schematic diagram of equipment arranged on an unmanned vehicle in an embodiment of the present invention;

[0052] Figure 3 Detailed flow chart of the method for avoiding each other of unmanned vehicles based on priority and spatiotemporal windows in an embodiment of the present invention;

[0053] Figure 4 Schematic diagram of the distribution results of the first 1000 chaotic values ​​for Circle mapping improvement;

[0054] Figure 5 This is a diagram showing the distribution of 1000 chaotic values ​​after Circle mapping improvement;

[0055] Figure 6 This is the flowchart of the improved Pied Kingfisher algorithm;

[0056] Figure 7 It is the convergence curve of each algorithm under the simple path;

[0057] Figure 8 It is the roadmap of each algorithm under the simple path;

[0058] Figure 9 It is the convergence curve of each algorithm under complex path;

[0059] Figure 10 It is the roadmap of each algorithm under complex path;

[0060] Figure 11 The present invention provides a driving route map for performing obstacle avoidance simulation on multiple unmanned vehicles using the method described in the present invention. DETAILED DESCRIPTION

[0061] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Example 1

[0063] This embodiment provides a method for autonomous vehicle mutual avoidance based on priority and spatiotemporal windows. The overall process of the method is as follows: Figure 1 shown.

[0064] Technical issues solved:

[0065] Path conflicts in multi-AV systems: In an environment where multiple AVs operate simultaneously, traditional path planning methods can lead to path intersections or collisions between them. This method effectively addresses this issue by introducing priorities and spatiotemporal windows.

[0066] Real-time avoidance in dynamic environments: In a dynamically changing working environment, unmanned vehicles need to adjust their paths in real time to avoid other moving unmanned vehicles or obstacles. This method achieves real-time avoidance by dynamically adjusting the spatiotemporal window.

[0067] Resource Competition and Efficiency Optimization: In a multi-AV system, competition for resources (e.g., routes and work areas) can lead to inefficiencies. By prioritizing tasks, this approach optimizes resource utilization, ensuring that high-priority tasks are completed first.

[0068] Collaborative work in complex environments: In complex environments, autonomous vehicles need to work together to complete tasks. This method enables autonomous vehicles to work together efficiently while avoiding conflicts through spatial and temporal window planning.

[0069] The core of the invention technology:

[0070] Priority allocation: Assign a priority to each unmanned vehicle based on mission urgency, unmanned vehicle type, or other factors. High-priority unmanned vehicles will pass through the conflict area first.

[0071] Space-time window: A space-time window (i.e., the range of occupation in time and space) is planned for each autonomous vehicle to ensure that the autonomous vehicle has exclusive access to the path within a specific time and space, avoiding conflicts with other autonomous vehicles.

[0072] Dynamic adjustment mechanism: According to environmental changes and the status of the unmanned vehicle, the priority and time and space windows are adjusted in real time to ensure the flexibility and robustness of the system.

[0073] Through this approach, multi-unmanned vehicle systems can achieve efficient and safe collaborative work in complex and dynamic environments, significantly improving the overall performance and reliability of the system.

[0074] Furthermore, based on the method described in the embodiment of the present invention, the relevant system architecture design can be further developed, as follows:

[0075] Multi-sensor data fusion

[0076] Sensor fusion: e.g. Figure 2 As shown, laser radar 3, ultrasonic sensor 4, and visual sensor 5 are used on the unmanned vehicle to collect environmental data, including dynamic information such as the position, speed, and direction of other vehicles, as well as static information such as road signs and traffic lights.

[0077] At the same time, infrared sensors, communication antennas and other equipment can be further arranged according to specific needs.

[0078] An onboard computer 1 is synchronously set on the unmanned vehicle to process visual and radar data.

[0079] The navigation control computer 2 is synchronously set on the unmanned vehicle to handle path planning and related collision judgment issues.

[0080] Map service: Integrates high-precision maps to provide vehicles with accurate road information, lane information, speed limit information, etc.

[0081] Navigation control computer decision-making

[0082] Priority allocation module: Sets the corresponding priority for each vehicle based on factors such as vehicle type (such as ordinary vehicle, emergency vehicle), driving status (such as acceleration, deceleration), and preset rules.

[0083] The Spatiotemporal Window Management Module defines time windows and spatial windows. A time window is the time period during which a vehicle is allowed to enter a specific area, while a spatial window is the spatial range reserved to maintain a safe distance. This module is responsible for monitoring and adjusting the spatiotemporal windows of each vehicle to avoid collisions.

[0084] Path planning and obstacle avoidance algorithm: Based on the information of priority and spatiotemporal window, the optimal driving path and avoidance strategy are calculated, while taking into account the reduction of delay time and ensuring traffic safety.

[0085] Power execution and status feedback

[0086] Control instruction generation: The planning results of the decision-making layer are converted into specific control instructions, such as steering angle, acceleration / deceleration, etc., and sent to the vehicle's actuators.

[0087] State feedback adjustment: Receives real-time feedback from multiple sensors and fine-tunes control instructions to ensure the vehicle drives safely along the planned trajectory.

[0088] Link communication

[0089] Link communication: supports communication between vehicles and between vehicles and infrastructure, and is used to exchange information such as location, speed, and intention, improving the system's coordination and response speed.

[0090] Network security: Ensure the security of the communication process and prevent malicious attacks or interference.

[0091] Example 2

[0092] This embodiment is a further limitation of the embodiment 1, and provides a detailed description of the method for autonomous vehicle mutual avoidance based on priority and spatiotemporal window. Figure 3 Shown is a specific flow chart of the method described in this embodiment.

[0093] The method comprises the following steps:

[0094] S1. Initialize a map of the multi-unmanned vehicle system, which includes road conditions and the position and speed of each unmanned vehicle in the multi-unmanned vehicle system;

[0095] S2. assigning a priority to each unmanned vehicle in the multi-unmanned vehicle system;

[0096] The rules for assigning priorities are as follows:

[0097] First, emergency priority: vehicles approaching the collision threshold have higher priority;

[0098] Secondly, critical tasks are prioritized: vehicles performing rescue and firefighting tasks have higher priority;

[0099] Again, traffic rules take precedence: vehicle priority is determined based on road right rules.

[0100] S3, planning an initial path for each unmanned vehicle in the multi-unmanned vehicle system;

[0101] Specifically: set the starting point and end point of each unmanned vehicle in the multi-unmanned vehicle system map, and plan the initial path from the starting point to the end point for each unmanned vehicle based on the road conditions in the map, with the principle of shortest driving distance and no collision.

[0102] S4. Based on the initial planned path, determine the predicted future time scale If no unmanned vehicle collision occurs, each unmanned vehicle will follow the initial path. time; if a collision occurs, the unmanned vehicle path is replanned, and then each unmanned vehicle drives according to the updated planned path time;

[0103] Judging in the predicted future time scale When determining whether an unmanned vehicle collision will occur, any unmanned vehicle in the unmanned vehicle system is taken as an object and judged one by one whether it has collided with other unmanned vehicles. Specifically:

[0104] S41. Setting the future time scale for predictions , and Divide into share, ,make , predicting that driverless cars will Position at the moment;

[0105] S42, determine whether any two unmanned vehicles are Whether the distance at the moment is less than or equal to the safe distance;

[0106] S43. If the distance is less than or equal to the safety distance, the predicted collision position is recorded, and then the process proceeds to step S45.

[0107] S44: If it is greater than the safety distance, proceed to step S45;

[0108] S45. Judgment and relationship, if , then another , then return to step S42; if If there is a recorded predicted collision location, then it is determined that If If there is no recorded predicted collision location, then it is determined that No collisions between autonomous vehicles will occur.

[0109] The safety distance refers to the minimum distance between the two unmanned vehicles that can continue to travel without colliding, determined based on the driving paths and current speeds of the two unmanned vehicles.

[0110] The re-planning of the unmanned vehicle path is specifically as follows:

[0111] S401, updating the multi-unmanned vehicle system map and adding the predicted collision location to the map;

[0112] S402: Based on the updated multi-unmanned vehicle system map, using the improved A* algorithm or the improved Pied Kingfisher algorithm to regenerate the planned path of the low-priority unmanned vehicle;

[0113] S403: Update the multi-unmanned vehicle system map and delete the predicted collision location.

[0114] The improved A* algorithm is specifically as follows: the evaluation function of the A* algorithm is improved, and the evaluation function of the A* algorithm is Change to ;

[0115] ;

[0116] in, is the distribution function of the standard normal distribution, , is the Euclidean distance from the current node to the end point; is the Euclidean distance from the starting point to the end point;

[0117] ;

[0118] in, .

[0119] The Pied Kingfisher algorithm is divided into three stages: the perching and hovering stage, the diving strategy stage, and the symbiotic stage.

[0120] The improved Pied Kingfisher algorithm is specifically as follows:

[0121] During the population initialization process before the perching and hovering phases, a multi-element initial population strategy was introduced;

[0122] In the symbiotic stage, the nonlinear energy attenuation factor E of the improved Golden Leopard optimization algorithm is introduced;

[0123] The joiner strategy of the sparrow search algorithm is introduced in the symbiotic stage.

[0124] S5. Determine whether each unmanned vehicle has reached the destination. If any unmanned vehicle has not reached the destination, the current path of the unmanned vehicle is regarded as the initial path, and the process returns to step S4 to continue. If any unmanned vehicle has reached the destination, the unmanned vehicle's journey ends.

[0125] Example 3

[0126] This example specifically illustrates the improved A* algorithm from Example 2. The A* (A-Star) algorithm is a heuristic search algorithm widely used in graph search and path planning, particularly for finding the shortest path from a start node to a target node in large-scale graphs. It combines the best-first search strategy of the Dijkstra algorithm with a heuristic estimation method, significantly improving search efficiency while ensuring the optimal solution is found.

[0127] The A* algorithm evaluates the evaluation function of each node To determine the search direction, where:

[0128] ;

[0129] From the starting point to the current node the actual cost; From the current node The estimated cost to the target node (heuristic function). The A* algorithm always prioritizes expansion The smallest node, thereby improving search efficiency while ensuring optimality.

[0130] Introduction to Common Improved A* Algorithms

[0131] Bidirectional A*: This algorithm searches simultaneously from the starting point and the target until the two search paths intersect. Its advantages include reducing the search space and improving search efficiency, making it particularly suitable for situations where both the starting point and the target are clearly defined. Its disadvantages include requiring additional data structures to manage the search in both directions and potentially failing to guarantee optimality in dynamic environments.

[0132] Incremental A*: This algorithm reuses previous search information to avoid replanning when the environment changes dynamically. This improved method is suitable for dynamic environments, reduces the computational overhead of replanning, and improves real-time performance. However, its implementation is more complex and requires additional memory to store historical information.

[0133] Hierarchical A*: This algorithm divides the map into multiple layers, performing rough planning at the higher levels before refining it at the lower levels. This improved approach significantly reduces the search space and improves computational efficiency, but the resulting path may not be globally optimal, and hierarchical design requires additional work.

[0134] Jump Point Search (JPS): This method exploits the symmetry of the map, skipping unnecessary nodes and directly searching for key nodes (jump points). This method can significantly reduce the number of search nodes in a grid map, but is only applicable to uniform grid maps.

[0135] Dynamic Weighted A* algorithm: Add a dynamic coefficient before h(n) and g(n) in the evaluation function, as follows:

[0136] ;

[0137] By adjusting the weight and , which can strike a balance between search speed and path optimality.

[0138] In the existing dynamic weighting algorithm, the weight is adjusted dynamically and segmentedly by the starting node, the target node, and the position of the current node between the starting node and the target node. and , such as the Chinese invention patent application "A multi-target point path planning method based on improved A* algorithm and particle swarm algorithm" (publication number: CN117739985A), or the weight and Replace it with a method defined by the starting node, target node, and current node, such as the Chinese invention patent application "An improved AGV path planning method based on A* algorithm" (publication number: CN111708364A).

[0139] Step-by-step weight adjustment and The main disadvantage of the method is that the segmented weight adjustment is prone to search oscillation in the area where the environmental characteristics suddenly change, resulting in path jitter; and Replacing it with a method that defines starting nodes, target nodes, and current nodes is relatively more flexible, but the weight generation mechanism based on preset rules is difficult to cope with unstructured scenarios such as dynamic obstacles and sudden changes in terrain.

[0140] In essence, the core contradiction of the existing dynamic weighting mechanism is that the weight function lacks quantitative modeling of the probabilistic characteristics of the search process. Therefore, the present invention proposes a new dynamic weighted A* algorithm, which introduces the normal function into the weight adjustment process of the A* algorithm, providing a probability-driven weight generation mechanism, so that the weight function can respond to environmental changes in real time and improve the accuracy of path planning.

[0141] The evaluation function for this application is: ;

[0142] ;

[0143] in, is the distribution function of the standard normal distribution, , is the Euclidean distance from the current node to the end point; is the Euclidean distance from the starting point to the end point.

[0144] ;

[0145] in, .

[0146] When the distribution function of the general normal distribution Parameters in When , it is the distribution function of the standard normal distribution.

[0147] It is easy to see from the distribution function graph of the standard normal distribution that we can control and The changing trend of and 's changing trend.

[0148] In this embodiment, and The value range is (-5, 0), and changes with the distance between the current node and the starting point and the end point. and The definition of The change from the starting point to the end point is from 1 to 0, Decrease from 0 to -5, Decrease from 1 to 0, From -5 to 0, Increase from 0 to 1.

[0149] Table 1 shows 、 scope of change and trend of change.

[0150] Table 1:

[0151]

[0152] Table 2 shows that 、 scope of change and trend of change.

[0153] Table 2:

[0154]

[0155] Example 4

[0156] This embodiment specifically describes the improved Pied Kingfisher algorithm in Example 2.

[0157] Path planning is a key issue in the design of unmanned vehicle navigation systems. The Pied Kingfisher Optimizer (PKO) is a novel intelligent optimization algorithm that simulates the hunting behavior and symbiotic relationships of the Pied Kingfisher. It features a simple structure and fast convergence. To address the shortcomings of the Pied Kingfisher algorithm, such as insufficient optimization accuracy and low search precision, a multi-strategy improved Pied Kingfisher algorithm (IPKO) is proposed. To enhance population diversity and optimize population quality, IPKO employs a multi-initialization strategy based on quantum computing. By incorporating a quantum gate dynamic mutation mechanism, this scheme generates an initial population set with greater randomness and diversity. The nonlinear energy decay factor E from the improved Golden Leopard Optimizer and the joiner strategy from the Sparrow Search Algorithm are introduced into the symbiotic phase of the Pied Kingfisher. Through role division and dynamic competition, the algorithm improves group collaboration efficiency and accelerates population convergence to high-quality areas. To further demonstrate the practical application potential of the IPKO algorithm, it was successfully applied to a robot path planning problem. Experimental results show that the proposed IPKO algorithm can effectively handle practical application problems, further verifying the effectiveness and reliability of IPKO in solving practical problems.

[0158] In 2024, Anas Bouaouda proposed the Pied Kingfisher algorithm (PKO), inspired by the social behavior of the Pied Kingfisher in nature. PKO is inspired by the hunting behavior and symbiotic relationships of Pied Kingfishers in nature and consists of three phases: perching and hovering, diving strategy, and symbiotic phase.

[0159] To address issues such as insufficient algorithm optimization accuracy and search capabilities, as well as path planning for mobile unmanned vehicles in both simple and complex environments, this paper proposes an Improved Pied Kingfisher Optimizer (IPKO) algorithm to further enhance the performance of Pied Kingfisher (PKO). A multi-population initialization strategy is introduced in the initial stages of PKO. This strategy avoids the problem of uneven initial population distribution and enhances global exploration capabilities by expanding the coverage of the feasible solution space. The PKO algorithm incorporates the nonlinear energy decay factor of the Golden Leopard Optimizer and the joiner strategy of the Sparrow Search Algorithm. This combination enables the algorithm to establish a dynamic balance during the search process, allowing for extensive exploration in the early stages while focusing on local optimization in the later stages, improving accuracy and convergence speed. This strategy guides the Pied Kingfisher algorithm to conduct a more efficient global search at each stage, avoiding being trapped in local optimal solutions and accelerating convergence.

[0160] This article introduces the original Pied Kingfisher algorithm. The PKO algorithm consists of three phases: perching and hovering, diving strategy, and symbiotic phase. Within the Pied Kingfisher algorithm, these unique behaviors embody sophisticated logic, significantly improving the Pied Kingfisher's search efficiency. Perching and hovering frequently shift postures across a vast search domain, uncovering more potential high-quality solutions; diving behavior allows timely penetration into key areas to accurately capture the optimal solution; and symbiotic behavior connects individuals and promotes cooperation. By leveraging the coordinated operation of these behavioral phases, the algorithm efficiently searches complex environments, directly reaching the optimal solution.

[0161] Before the perching and hovering phases, the population needs to be initialized. The PKO optimization algorithm, like many other algorithms, starts the search process by randomly generating a set of initial solutions from the search space. The following is the expression for population initialization:

[0162] (1)

[0163] in, It represents the i Individual in the j The specific position in the dimension, rand represents a random number between (0, 1), and It represents the upper and lower bounds of the population search space.

[0164] After the initial population is generated, the fitness function evaluates each individual Pied Kingfisher to determine their actual problem-solving ability. The best performing individuals are selected based on their outstanding evaluation values ​​to breed the next generation. The following is the formula for the exploration phase:

[0165] (2)

[0166] (3)

[0167] in, and Representing the Second and In the iteration The location of the Pied Kingfisher, Representative In the iteration The position and parameters of the Pied Kingfisher From the normal distribution, Represents a 1-row Dim-column matrix, each element is The dimension Dim is used to reflect the complexity of the solution. T The value will be dynamically adjusted according to the current strategy ("perching" strategy or "hovering" strategy).

[0168] When perched, parameters T It is expressed as the following formula (4):

[0169] (4)

[0170] (5)

[0171] in, is the current iteration number, It represents the maximum number of iterations. is the constant 8. It represents a random value that is uniformly distributed in the interval (0, 1).

[0172] When hovering, the parameters T It can be expressed by the following formula (6):

[0173] (6)

[0174] (7)

[0175] in, and Respectively represent Individuals and The fitness value of an individual.

[0176] Diving strategy stage. The diving behavior of the Pied Kingfisher has a significant impact on its predator, enabling it to thrive in the aquatic environment. The following formula (8) is the formula for the Pied Kingfisher's diving strategy:

[0177] (8)

[0178] (9)

[0179] (10)

[0180] (11)

[0181] in, This parameter represents the hunting ability of the Pied Kingfisher. It refers to the best fitness value obtained in the iterative process. It acts as a scaling factor, and its main purpose is to control the search range of individuals. Continuously increasing, parameters The value of will show a trend of gradually decreasing. It is an intermediate variable that plays an important role in the equilibrium search process. It means the specific location of the optimal solution at the moment.

[0182] Symbiotic stage. A typical example of this particular stage is the symbiotic relationship between the Pied Kingfisher and the Otter. The essence of this symbiotic relationship is that the two different species can each benefit from the other without causing harm to the other. The specific behavior can be expressed mathematically, as shown in the following equation (12):

[0183] (12)

[0184] (13)

[0185] in and represents the positions of two individuals randomly selected from the population, Indicates the hunting efficiency of the Pied Kingfisher. and The values ​​of are set to 0.5 and 0 respectively.

[0186] The improved Pied Kingfisher algorithm is introduced in detail.

[0187] During the population initialization process before the perching and hovering stages, a multi-element initial population strategy was introduced.

[0188] Like most heuristic algorithms, the PKO algorithm generates the initial population in a pseudo-random manner, which may lead to the aggregation of initial individuals and uneven population distribution. In order to obtain a higher-quality initial population, a multi-initial population strategy is introduced. Chaos is a natural phenomenon that exhibits highly nonlinear characteristics. Because the chaotic sequence it generates has extremely strong randomness and ergodic characteristics, it has been widely used in the field of optimization search-related problems. Circle mapping is one of them, and its expression is shown as follows:

[0189] (14)

[0190] in, Indicates the The distribution of 1000 chaotic values ​​is as follows: Figure 4 As shown, from Figure 4 It can be seen that most of the chaotic values ​​are in the range of [0.3, 0.7], and the frequency distribution is uneven. Therefore, the following improvement is made:

[0191] (15)

[0192] After improvement, the distribution of 1000 chaotic values ​​is as follows Figure 5 As shown, compared Figure 4 The improved chaotic value distribution is more uniform, and the sequences generated within the parameter range traverse the entire state space, indicating good uniform distribution. As can be seen from the scatter plot, Figure 5 The scattered distribution in appears more random, with no obvious clustering areas. Figure 4 Although the distribution of scattered points in is also random, there is a clear clustering of scattered points at certain horizontal axis positions (such as 200-300 and 700-800). This indicates that the improved circle chaotic map has a significantly improved uniform distribution compared to the original. Then, equations (15) and (16) are used for population initialization:

[0193] (16)

[0194] Quantum computing is one of the cutting-edge directions in the field of computing. The qubit is the basic unit of quantum computing. Unlike bits in classical computers, which can only represent two states, 0 or 1, qubits can be in a superposition state of 0 and 1. This superposition state can be expressed as Equation (18), where , represents the probability between two states. The two basic states of the qubit are represented by and They constitute the basic storage unit , as shown in Equation (17). In a quantum computer, the state of a qubit evolves based on the operation of a quantum gate (similar to the logic gates in a classical computer). A quantum gate is a basic operation performed on one or more qubits. Complex quantum algorithms can be implemented through the operation of a series of quantum gates.

[0195] (17)

[0196] (18)

[0197] When an observation is performed on a quantum state in superposition, the quantum state collapses, with a certain probability of collapsing to the zero state and a corresponding probability of collapsing to the 1 state. Leveraging this property of probability amplitude, qubits can be represented in the form of a matrix constructed from sine and cosine, as shown in the following formula:

[0198] (19)

[0199] in, = , = , represents a transpose. The aforementioned quantum correlation properties are applied during the initialization of the group PKO. Specifically, qubits are used to construct individuals. In other words, probability amplitudes are used to represent the positions held by different pied kingfishers. Based on this, the following correspondence exists between the position of an individual in quantum space and its state matrix:

[0200] (20)

[0201] Among them, and are the cosine and sine positions, respectively, is the rotation angle. The mapping relationship between quantum space and solution space is established through linear transformation. If there is such a quantum bit, its representation is , Represents the quantum bit angle, where the value range of the quantum bit is At the same time, the corresponding solution can be expressed as , its value range is [ , ].

[0202] On this basis, the following series of conversion operations can be carried out, as shown below:

[0203] (twenty one)

[0204] (twenty two)

[0205] in, and are the upper and lower limits of the population range. Since the position of each Pied Kingfisher is determined by the improved circular chaos map, this process is actually like the corresponding quantum state collapsing into a deterministic state after observation. To restore it to the superposition state, assuming that its current state is a sine or cosine position, the conversion can be performed according to the following formula: Indicates the rotation angle:

[0206] (twenty three)

[0207] (twenty four)

[0208] In addition, mutation can be used to obtain a more diverse initial population. In the field of quantum computing, quantum devices used to implement quantum state logic conversion operations are called quantum logic gates. Among them, there are some common quantum gates, such as Pauli gates, Hadamard gates, and quantum rotation gates. In this invention, a dynamic quantum rotation gate is used, which is specifically expressed as follows:

[0209] (25)

[0210] in, Represents the rotation matrix of the dynamic quantum rotation gate, π / 4≤ ≤2π, represents the rotation angle of the dynamic quantum revolving gate. The update formula using the dynamic quantum revolving gate is as follows:

[0211] (26)

[0212] The position of the population after the mutation needs to be calculated using equations (21) and (22). Finally, the individuals at the initial sine and cosine positions after the revolving door mutation are compared with those originally at the sine and cosine positions, and the best individuals are selected from them, thus selecting the final initial population.

[0213] In the symbiotic stage, the nonlinear energy attenuation factor E of the improved Golden Leopard optimization algorithm is introduced; and the joiner strategy of the Sparrow search algorithm is introduced.

[0214] The symbiotic stage of PKO lacks a clear role division mechanism, resulting in limited resource utilization efficiency, especially in complex problems, which makes it easy to fall into local optimality. This paper introduces the joiner strategy of the sparrow search algorithm into the PKO symbiotic stage, and improves the efficiency of group collaboration through role division and dynamic competition. Specifically, some Pied Kingfisher individuals actively follow the position of the current optimal kingfisher, which accelerates the algorithm's convergence to potential high-quality areas in the solution space; when multiple candidate Pied Kingfisher individuals compete for the same high fitness target, the system will screen out dominant individuals with higher optimization potential through a direct comparison mechanism of fitness values. In addition, this paper introduces the nonlinear energy attenuation factor E of the Golden Leopard Optimization Algorithm to balance local and global search capabilities:

[0215] (27)

[0216] in, Indicates the current iteration number, Indicates the maximum number of iterations.

[0217] And introduce the joiner strategy of the sparrow search algorithm so that the updated position of the Pied Kingfisher is achieved through:

[0218] (28) obtain, wherein, Indicates the updated The location of the Pied Kingfisher, and represents the positions of two individuals randomly selected from the population, represents the normal distribution parameter, Indicates the number before the update The location of the Pied Kingfisher, For Random values ​​are uniformly distributed within this interval. Indicates the hunting efficiency of the Pied Kingfisher. Indicates the specific location of the optimal solution at the current time; ,in, Indicates that its elements are randomly assigned or of matrix, for The transposed matrix of Indicates that an element is of matrix.

[0219] The flowchart of the improved Pied Kingfisher algorithm is shown in Figure 6 , the pseudo code is as follows:

[0220] [Enter the maximum number of iterations M, the population number Pop, and the beating factor BF

[0221] Use formulas (15) and (16) to initialize the population

[0222] The rotation angle is obtained from formulas (23) and (24):

[0223] According to formulas (21) and (22), another position is obtained

[0224] Perform mutation operations according to formulas (25) and (26)

[0225] Compare the individuals in the population listed above and select the best one as the initial population

[0226] Calculate the fitness value of each individual and record the current individual optimal solution and optimal position

[0227] while t <M do

[0228] Calculate E and E by equations (10) and (27)

[0229] if rand()<0.8 then

[0230] if rand()>0.5 then

[0231] Calculate the habitat parameter T according to formula (4):

[0232] The position of the Pied Kingfisher when roosting is calculated according to formula (2).

[0233] else

[0234] According to formula (6), the hovering parameter T is calculated

[0235] The position of the Pied Kingfisher when roosting is calculated according to formula (2).

[0236] end

[0237] else

[0238] According to formula (8), the position of the Pied Kingfisher when diving for fish is calculated.

[0239] end

[0240] if rand()>(1-PE) then

[0241] Update the position of the Pied Kingfisher in the symbiotic stage according to the above formula (28)

[0242] else

[0243] The position of the Pied Kingfisher in the symbiotic stage is updated according to the following formula (28):

[0244] end

[0245] t=t+1

[0246] end

[0247] Returns the optimal position and optimal fitness value]

[0248] In this paper, in order to verify the effectiveness of the improved Pied Kingfisher algorithm in different obstacle environments, two different maps were designed according to the number and sparse distribution of obstacles, namely simple terrain and complex terrain. The number of obstacles in simple terrain is small and the distribution is sparse, while the number of obstacles in complex terrain is large and the distribution is dense.

[0249] The authors also studied the search for the optimal path on two different maps using the Black Kite Optimization Algorithm (BKA), the Sparrow Optimization Algorithm (SSA), the Osprey Optimization Algorithm (OOA), the Chef-Based Optimization Algorithm (CBOA), the Coot Optimization Algorithm (COOT), the Pied Kingfisher Algorithm (PKO), and the Improved Pied Kingfisher Algorithm (IPKO). The algorithm was set to a maximum of 100 iterations, a population of 30, and 10 runs. In these swarm intelligence algorithms, each individual in the population represents a potential solution (for path planning, this can be considered a possible path for the robot). During the iteration process, each individual in the swarm continuously tracks its own optimal position and the optimal position within the group, continuously updating its own speed and position accordingly. After multiple rounds of iteration, the algorithm gradually converges towards the optimal solution, ultimately finding the optimal path.

[0250] Figure 7 is the convergence curve of each algorithm under the simple path, Figure 8 is the roadmap of each algorithm under the simple path, Figure 9 is the convergence curve of each algorithm under complex path, Figure 10 Table 3 is the roadmap of each algorithm under complex paths. Table 3 shows the optimal solutions obtained by optimizing simple path planning of unmanned vehicles with different algorithms. Table 4 shows the optimal solutions obtained by optimizing complex path planning of unmanned vehicles with different algorithms. From the above comparison, it can be seen that the optimal path obtained by the improved Kingfisher algorithm in the present invention is superior to other algorithms in both simple and complex paths, showing significant advantages.

[0251] Table 3:

[0252]

[0253] Table 4:

[0254]

[0255] Based on the original Pied Kingfisher (PKO) algorithm, an improved Pied Kingfisher (IPKO) algorithm was proposed to address issues such as insufficient optimization accuracy and low search precision in the Pied Kingfisher (PKO) algorithm. This algorithm addresses issues such as the relatively small number of population categories in the original PKO algorithm, which results in weak global exploration capabilities, a lack of information exchange between individuals leading to blind search, and a tendency to fall into local optimality in late iterations. Specific improvements include:

[0256] A multi-population initialization strategy was introduced to improve population diversity;

[0257] The nonlinear energy attenuation factor E of the improved Golden Leopard optimization algorithm is introduced to balance the local and global search capabilities; (Formula 27)

[0258] By introducing the joiner strategy of the sparrow search algorithm into the symbiotic phase of the pied kingfisher, we can accelerate convergence by guiding some individuals to track the optimal solution through role division. At the same time, we use a dynamic competition mechanism to select individuals with high fitness, effectively improving the algorithm's global optimization efficiency and accuracy. (Formula 28)

[0259] The test results above demonstrate that IPKO achieves more stable convergence and higher accuracy than other swarm intelligence optimization algorithms. Finally, the algorithm's effectiveness in solving practical problems was demonstrated using path planning for an unmanned vehicle.

[0260] Example 5

[0261] This embodiment provides a driving route map for multi-unmanned vehicle obstacle avoidance simulation using the method of the present invention, such as Figure 11 As shown in the figure, serial numbers 1-6 represent the six autonomous vehicles, with increasing priority. S represents the starting point, T represents the end point, and the "x" symbol on the grid network represents an obstacle. The "x" symbol on the autonomous vehicle's route represents the predicted collision location.

[0262] This embodiment further illustrates specific application examples of the method described in the present invention and lists several important implementation methods.

[0263] Warehousing and logistics:

[0264] In e-commerce warehouses, hundreds of AGVs (automated guided vehicles) use this method to achieve efficient path planning and conflict avoidance, significantly improving cargo sorting and transportation efficiency.

[0265] Smart Manufacturing:

[0266] In factory production lines, multiple collaborative unmanned vehicles work together through priorities and time-space windows to avoid collisions and optimize production processes.

[0267] Autonomous driving:

[0268] In an autonomous driving fleet, vehicles use this method to achieve safe and efficient path planning, reducing traffic congestion and accident risks.

[0269] Serving unmanned vehicles:

[0270] In a hotel or hospital, service autonomous vehicles use this method to avoid interfering with each other, ensuring efficient and seamless service to customers.

[0271] In summary, the method described in the present invention provides an efficient, safe and flexible mutual avoidance method for multiple unmanned vehicle systems through the organic combination of priority and spatiotemporal window, which has significant technical advantages and application value.

Claims

1. A multi-unmanned vehicle mutual avoidance method based on priority and spatiotemporal windows, characterized by: The method comprises the following steps: S1. Initialize a map of the multi-unmanned vehicle system, which includes road conditions and the position and speed of each unmanned vehicle in the multi-unmanned vehicle system; S2. assigning a priority to each unmanned vehicle in the multi-unmanned vehicle system; S3, planning an initial path for each unmanned vehicle in the multi-unmanned vehicle system; S4. Based on the initial planned path, determine the predicted future time scale If no unmanned vehicle collision occurs, each unmanned vehicle will follow the initial path. time; if a collision occurs, the unmanned vehicle path is replanned, and then each unmanned vehicle drives according to the updated planned path time; Replanning the unmanned vehicle path is as follows: S51. Update the multi-autonomous vehicle system map and add the predicted collision location to the map; S52, based on the updated multi-unmanned vehicle system map, using the improved Pied Kingfisher algorithm to regenerate the planned path of the low-priority unmanned vehicle; S53, updating the multi-unmanned vehicle system map and deleting the predicted collision location; The improved Pied Kingfisher algorithm is specifically as follows: During the population initialization process before the perching and hovering phases, a multi-element initial population strategy was introduced; In the symbiotic stage, the nonlinear energy attenuation factor E of the improved Golden Leopard optimization algorithm is introduced; Introducing the joiner strategy of the sparrow search algorithm in the symbiotic stage; The nonlinear energy attenuation factor E of the Golden Leopard optimization algorithm is specifically: ;in, Indicates the current iteration number, Indicates the maximum number of iterations; After introducing the joiner strategy of the sparrow search algorithm, the updated position of the Pied Kingfisher is obtained by Obtain, among which, Indicates the updated The location of the Pied Kingfisher, and represents the positions of two individuals randomly selected from the population, represents the normal distribution parameter, Indicates the number before the update The location of the Pied Kingfisher, For Random values ​​are uniformly distributed within this interval. Indicates the hunting efficiency of the Pied Kingfisher. Indicates the specific location of the optimal solution at the current time; ,in, Indicates that its elements are randomly assigned or of matrix, for The transposed matrix of Indicates that an element is of matrix; S5. Determine whether each unmanned vehicle has reached the destination. If any unmanned vehicle has not reached the destination, the remaining path of the unmanned vehicle is regarded as the initial path, and the process returns to step S4 to continue. If any unmanned vehicle has reached the destination, the unmanned vehicle's journey ends.

2. The method for avoiding mutual avoidance of multiple unmanned vehicles based on priority and spatiotemporal windows according to claim 1 is characterized in that: The rules for assigning priorities are as follows: First, emergency priority: vehicles approaching the collision threshold have higher priority; Secondly, critical tasks are prioritized: vehicles performing rescue and firefighting tasks have higher priority; Again, traffic rules take precedence: vehicle priority is determined based on road right rules.

3. The method for avoiding mutual avoidance of multiple unmanned vehicles based on priority and spatiotemporal windows according to claim 2 is characterized in that: Plan the initial path for each unmanned vehicle in the multi-unmanned vehicle system. Specifically, set the starting point and end point of each unmanned vehicle in the multi-unmanned vehicle system map. According to the road conditions in the map, with the principle of shortest driving distance and no collision, plan the initial path from the starting point to the end point for each unmanned vehicle.

4. The method for avoiding mutual avoidance of multiple unmanned vehicles based on priority and spatiotemporal windows according to claim 3 is characterized in that: Judging in the predicted future time scale When determining whether an unmanned vehicle collision will occur, any unmanned vehicle in the unmanned vehicle system is taken as an object and judged one by one whether it has collided with other unmanned vehicles. Specifically: S41. Setting the future time scale for predictions , and Divide into share, ,make , predicting that driverless cars will Position at the moment; S42, determine whether any two unmanned vehicles are Whether the distance at the moment is less than or equal to the safe distance; S43. If the distance is less than or equal to the safety distance, the predicted collision position is recorded, and then the process proceeds to step S45. S44: If it is greater than the safety distance, proceed to step S45; S45. Judgment and relationship, if , then another , then returns to step S42; if If there is a recorded predicted collision location, then it is determined that If If there is no recorded predicted collision location, then it is determined that No collisions between autonomous vehicles will occur.

5. The method for avoiding mutual avoidance of multiple unmanned vehicles based on priority and spatiotemporal windows according to claim 4 is characterized in that: The safety distance refers to the minimum distance between the two unmanned vehicles that can continue to travel without colliding, determined based on the driving paths and current speeds of the two unmanned vehicles.

6. The method for avoiding mutual avoidance of multiple unmanned vehicles based on priority and spatiotemporal windows according to claim 5 is characterized in that: After introducing the multi-initial population strategy, the expression for population initialization becomes: ; in, Indicates the Individuals in The specific location in the dimension, Indicates the The number of chaotic sequences, and They represent the upper and lower bounds of the population search space respectively.

7. A multi-unmanned vehicle mutual avoidance system based on priority and spatiotemporal windows, characterized by: The system comprises: A module that initializes a map of the multi-route vehicle system: the map includes road conditions and the position and speed of each vehicle in the multi-route vehicle system; A module that assigns priorities to each unmanned vehicle in a multi-unmanned vehicle system; Path planning module: plans the initial path for each unmanned vehicle in the multi-unmanned vehicle system, and judges the predicted future time scale based on the planned initial path. If no unmanned vehicle collision occurs, each unmanned vehicle will follow the initial path. time; if a collision occurs, the path is replanned, and then each unmanned vehicle drives according to the updated planned path Time; re-planning the unmanned vehicle path is as follows: S51. Update the multi-autonomous vehicle system map and add the predicted collision location to the map; S52, based on the updated multi-unmanned vehicle system map, using the improved Pied Kingfisher algorithm to regenerate the planned path of the low-priority unmanned vehicle; S53, updating the multi-unmanned vehicle system map and deleting the predicted collision location; The improved Pied Kingfisher algorithm is specifically as follows: During the population initialization process before the perching and hovering phases, a multi-element initial population strategy was introduced; In the symbiotic stage, the nonlinear energy attenuation factor E of the improved Golden Leopard optimization algorithm is introduced; Introducing the joiner strategy of the sparrow search algorithm in the symbiotic stage; The nonlinear energy attenuation factor E of the Golden Leopard optimization algorithm is specifically: ;in, Indicates the current iteration number, Indicates the maximum number of iterations; After introducing the joiner strategy of the sparrow search algorithm, the updated position of the Pied Kingfisher is obtained by Obtain, among which, Indicates the updated The location of the Pied Kingfisher, and represents the positions of two individuals randomly selected from the population, represents the normal distribution parameter, Indicates the number before the update The location of the Pied Kingfisher, For Random values ​​are uniformly distributed within this interval. Indicates the hunting efficiency of the Pied Kingfisher. Indicates the specific location of the optimal solution at the current time; ,in, Indicates that its elements are randomly assigned or of matrix, for The transposed matrix of Indicates that an element is of matrix; The module for determining the destination is used to determine whether each unmanned vehicle has reached the destination. If any unmanned vehicle has not reached the destination, the remaining path of the unmanned vehicle is regarded as the initial path, and the module returns to the path planning module to continue. If any unmanned vehicle has reached the destination, the driving of the unmanned vehicle ends.

Images