Campus open space room heterogeneous unmanned cluster three-dimensional security patrol method and system

Abstract

The present application relates to the technical field of inspection, and discloses a campus open space room heterogeneous unmanned cluster three-dimensional security inspection method, comprising: a central control platform based on an event type-atomic task-robot capability mapping knowledge base, decomposing a complex event into an atomic task set, and defining a capability demand vector; collecting real-time states of robots, constructing a constrained multi-objective optimization problem, and using a two-stage collaborative solving method combining an improved meme algorithm and conflict-oriented search to output an optimal task allocation and collaborative space-time path plan; issuing instructions through wireless communication, and robots working collaboratively and uploading data in real time; the central control platform fuses and analyzes data, triggers a rapid generation of an adjustment scheme when re-planning, and collects data after the completion of a task; to solve the technical problem that heterogeneous robots lack intelligent task allocation capability in a three-dimensional complex environment, resulting in low inspection collaboration efficiency.

Description

Technical Field

[0001] This invention relates to the field of inspection technology, specifically to a heterogeneous unmanned cluster three-dimensional security inspection method and system for open spaces on campuses. Background Technology

[0002] Campus security patrol scenarios encompass diverse spaces such as open plazas, roads, buildings, green belts, and underground parking garages, with a wide variety of emergencies and high requirements for timely response. Existing solutions have significant shortcomings: Traditional manpower and fixed monitoring methods rely on security patrols and fixed camera surveillance, which have blind spots, limited frequency, and reduced efficiency at night. Fixed cameras have limited field of view, making it difficult to cover inside buildings and complex terrains, and they only have passive recording functions, lacking proactive early warning and real-time response capabilities.

[0003] All types of robots have structural shortcomings: drones have short battery life, are susceptible to weather conditions, and cannot enter indoor environments; unmanned vehicles are limited by roads and cannot handle complex terrains such as lawns and stairs; robot dogs have limited battery life and payload, and are slow, making them unsuitable for large-scale outdoor inspections. A single type of robot cannot simultaneously meet the comprehensive inspection needs of the campus's air, ground, and indoor environments.

[0004] Existing heterogeneous multi-robot collaborative solutions are mostly found in structured environments or specific task scenarios, and have poor adaptability to open and dynamic campus environments. They often adopt static collaborative modes such as master-slave following and fixed area division, lacking intelligent scheduling capabilities based on real-time events. Task allocation is mostly preset or triggered by simple rules, making it difficult to dynamically adjust the capability combination and task division of the three types of devices. Summary of the Invention

[0005] The present invention aims to provide a method and system for three-dimensional security patrol and inspection of heterogeneous unmanned clusters in open spaces on campuses, in order to solve the technical problem that heterogeneous robots lack intelligent task allocation capabilities in complex three-dimensional environments, resulting in low patrol and collaborative efficiency.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campuses, comprising: The central control platform receives inspection instructions or abnormal event alarms, and based on the built-in event type-atomic task-robot capability mapping knowledge base, it breaks down complex events into sets of atomic tasks. And define a capability requirement vector for each atomic task t. ; The real-time states of drones, unmanned vehicles, and robot dogs are collected to form a global state set S. Based on the current task and robot state, a constrained multi-objective optimization problem is constructed. A two-stage collaborative solution method combining an improved meme algorithm and conflict-oriented search is adopted to output the optimal task allocation scheme. and collaborative spatiotemporal path planning ; The central control platform will optimize the task allocation scheme. and collaborative spatiotemporal path planning The data is simultaneously transmitted to the corresponding drones, unmanned vehicles, and robot dogs via wireless communication links. After receiving the instructions, each heterogeneous robot follows the collaborative spatiotemporal path plan. The specified time and space window and path trajectory initiate coordinated motion operations; Each heterogeneous robot is assigned a task according to the optimal task allocation scheme. The system assigns atomic tasks and uploads the task execution status, real-time audio and video, sensor data, location and power information to the central control platform in real time. The central control platform integrates and analyzes the returned data. When the replanning condition is triggered, it uses historical solutions as hot start information, reconstructs and solves the multi-constraint optimization model based on the current global state, generates the adjusted collaborative solution, and sends it out for execution. After the task is completed, the central control platform summarizes all the data from the entire process.

[0007] The principle and advantages of this solution are as follows: the system does not blindly execute tasks, but has a built-in knowledge base mapping "event type - atomic task - robot capability". When a complex event such as a fire or intrusion is received, the system automatically breaks down the event into a specific sequence of atomic tasks (such as location, reconnaissance, and notification) and matches it with the corresponding robot capability requirements, achieving a precise translation from abstract events to executable tasks.

[0008] The first stage, serving as global scheduling, utilizes an improved meme algorithm to comprehensively consider time, energy consumption, risk, and the robot's real-time state, seeking the optimal task allocation scheme from a massive number of combinations. This solves the problem that traditional genetic algorithms are prone to getting stuck in local optima and cannot take multiple objectives into account in task allocation.

[0009] The second phase is used for conflict-free execution, utilizing Conflict-Oriented Search (CBS) combined with customization. The algorithm plans kinematically feasible and spatiotemporally conflict-free fine-grained paths for three completely different types of robots: drones, unmanned vehicles, and robot dogs. This method ensures that multiple robots do not collide and can cooperate efficiently, solving the problem of path conflict among multiple agents. The combination of two stages guarantees both the optimality of global task allocation and the feasibility of conflict-free local execution paths. This combined algorithm of macro-scheduling and micro-collision avoidance is highly innovative.

[0010] The task execution process is not static. When new high-priority events occur, the robot runs out of power, or a malfunction occurs, the system uses historical solutions as a warm start, quickly re-solves the optimization model, and dynamically adjusts the scheduling strategy to ensure uninterrupted inspection tasks.

[0011] This solution eliminates blind spots in all scenarios, including plazas, buildings, and underground parking garages, through a combination of drones, unmanned vehicles, and robotic dogs; it possesses proactive early warning and real-time response capabilities. Leveraging the complementary capabilities of these three types of robots, a three-dimensional collaborative system of "air-ground-indoor" is constructed. Drones handle large-scale rapid reconnaissance, robotic dogs handle detailed indoor evidence collection, and unmanned vehicles handle material transportation; the three work together to compensate for each other's shortcomings. Moving away from static models, a multi-objective optimization model is built based on real-time events. The system dynamically adjusts which robot is responsible for which task based on its current location, battery level, and the urgency of the event, achieving optimal resource allocation.

[0012] Preferably, as an improvement, in dynamic task allocation and collaborative planning, the objective function of the multi-objective optimization problem is to minimize the global cost J:

[0013] Where A is the set of available robots; For a set of atomic tasks, Let be a binary decision variable, representing whether atomic task t is assigned to the robot. ; , , Robots The estimated execution time, energy consumption, and risk factor for task t; and These are the normalization coefficients; These are adjustable weighting coefficients.

[0014] The beneficial effects of this improvement are: by constructing a global cost function with the goal of minimizing the weighted average of time, energy consumption, and risk, and by introducing a normalization coefficient to eliminate the numerical imbalance caused by different dimensions, the optimization objective becomes more scientific and reasonable; at the same time, by flexibly adapting the adjustable weight coefficient to the needs of different scenarios such as daily inspections and emergency responses, a dynamic balance between efficiency, energy consumption, and safety is achieved, significantly improving the rationality and global optimality of task allocation.

[0015] Preferably, as an improvement, in dynamic task allocation and collaborative planning, the multi-objective optimization problem needs to satisfy the following constraints: Task completion constraint: Each atomic task must be assigned one and only once; Capability matching constraint, if Then the robot's capability vector The task requirement vector must be satisfied. ; Energy constraint: The total energy consumption of all tasks assigned to a single robot must not exceed its remaining power. Spatiotemporal conflict-free constraints ensure that the paths assigned to different robots do not overlap in space and time, which is guaranteed by assigning a unique spatiotemporal window to each path segment.

[0016] The beneficial effects of this improvement are: by setting four hard constraints—full task coverage, capability matching, power safety, and no spatiotemporal conflicts—it ensures from a mechanism perspective that no task is missed, robot capabilities are precisely matched with tasks, energy consumption is not exceeded, and multi-machine movement does not collide. This avoids risks such as task failure, robot overload, and path conflicts from the source, and significantly improves the safety, stability, and reliability of the system operation.

[0017] Preferably, as an improvement, the two-stage collaborative solution method includes: In the first stage, the task allocation is solved based on the improved meme algorithm to obtain the optimal binary decision variable matrix. ; In the second stage, a multi-agent conflict-free path planning based on conflict-oriented search (CBS) is performed, using the optimal task allocation scheme output from the first stage. As input, it plans precise, conflict-free spatiotemporal paths for each robot that satisfy kinematic constraints, and outputs a cooperative spatiotemporal path plan. .

[0018] The beneficial effects of this improvement are: it decomposes the complex heterogeneous cluster collaboration problem into two stages: global task allocation and local conflict-free path planning, realizing hierarchical collaboration of "optimal allocation first, then safe execution"; it ensures both the global optimization of task scheduling and the physical feasibility of path execution, breaking through the technical bottleneck that traditional single algorithms cannot simultaneously take into account scheduling optimization and path safety, and creatively achieving efficient collaboration.

[0019] Preferably, as an improvement, the first stage of task allocation solution based on the improved meme algorithm includes: Initialization, generating the initial population This includes individuals generated based on heuristic rules from a mapping knowledge base and randomly generated individuals; Fitness calculation: Calculate the fitness of each individual i in the population. ,in The objective function value, The sum of penalty values ​​for violating hard constraints, where λ is the penalty coefficient; Genetic operations are performed sequentially, including population selection based on roulette wheel selection and elite retention strategies, uniform crossover based on task blocks, and mutation probability-based selection. Randomly replace the mutation operation performed by the robot; Local search involves performing customized local searches on elite individuals, including path generation, conflict detection, conflict resolution, and acceptance and iteration processes, optimizing the allocation scheme towards path feasibility and low conflict. Iterative updates are performed, repeatedly executing fitness calculations, genetic operations, and local searches until the maximum number of iterations is reached or fitness convergence is achieved. The optimal task allocation scheme is obtained by decoding the individual with the highest fitness. .

[0020] The beneficial effects of this improvement are: it improves the quality of the population by using heuristic and random hybrid initialization, eliminates infeasible solutions through fitness penalty mechanism, and strengthens the inheritance of high-quality genes and local feasibility optimization by combining task block crossover and elite local search, making the algorithm converge faster and obtain high-quality solutions more easily; compared with traditional genetic algorithms, it has higher search efficiency, higher feasibility, and stronger optimality, and has outstanding creativity and practicality.

[0021] Preferably, as an improvement, in the fitness calculation, when estimating the time for robot a to perform task t... Energy consumption At that time, a fast path simulation method based on a simplified topology map is adopted, and a conflict hotspot penalty mechanism is introduced: The estimated occupancy time of shared node v in all robot topology paths is calculated. If the same node v is occupied by multiple paths in a similar time period, additional risk penalties are applied to related tasks involving that node. .

[0022] The beneficial effects of this improvement are: introducing fast path simulation and conflict hotspot penalty in the fitness calculation stage enables early prediction and avoidance of conflicts, reducing the probability of path conflicts from the global search stage and reducing the pressure of later fine planning to resolve them; it not only speeds up the algorithm's convergence speed but also improves the security of the final solution, which is a forward-looking security optimization innovation design that traditional algorithms do not have.

[0023] Preferably, as an improvement, the second stage of conflict-oriented search (CBS) based multi-agent conflict-free path planning includes: The underlying path planning, for each robot 'a' with an assigned task, employs... The algorithm plans k shortest path candidates, and the adaptation includes search space customization, cost function customization, heuristic function customization, kinematic constraint verification and campus spatiotemporal constraint integration; To resolve high-level conflicts, a constraint tree (CT) is established, using the underlying path planning results of all robots as the initial root node. A branch exploration strategy is then employed to address detected spatiotemporal conflicts. Generate two child nodes, one for the robot. and Add a spatiotemporal constraint (a,v,t) to restrict it from occupying spatial unit v at time t, and replan the path for the constrained robot; Iterative solution: repeatedly execute conflict detection, constraint generation, and cost update processes until conflict-free feasible nodes are obtained, and output a set of conflict-free paths. And convert it into a collaborative spatiotemporal path plan .

[0024] The beneficial effect of this improvement is: through underlying customization Adapting to the motion characteristics of three types of heterogeneous robots, this method automatically resolves spatiotemporal conflicts through high-level constraint tree branch search, creatively achieving fine-grained path planning that is collision-free, kinematically feasible, and compliant with campus constraints for multiple robots. Compared with traditional path planning methods, this method can automatically handle complex spatiotemporal conflicts without human intervention, significantly improving collaborative safety and intelligence.

[0025] Preferably, as an improvement, in situation assessment and dynamic replanning, the replanning conditions include new emergencies, robot malfunctions, insufficient power, changes in task priority, or abnormal task execution. When the replanning condition is triggered, the system initiates a warm-start dynamic replanning mechanism: using the optimal task allocation scheme output from the previous round. With collaborative spatiotemporal path planning As warm-up information, the robot's real-time position, remaining battery power, and unfinished task sequence are used as new inputs to rerun the improved meme algorithm and set the initial population to be adjusted based on the original scheme, so as to quickly generate an adjusted collaborative scheme adapted to the current state.

[0026] The benefits of this improvement are: by reusing historical optimal solutions through hot start information, replanning does not require searching from scratch, significantly shortening emergency response time; in the face of dynamic changes such as new events, robot failures, and insufficient power, new collaborative solutions can be quickly generated, ensuring uninterrupted, continuous, and reliable inspections, and significantly improving the system's robustness, real-time performance, and emergency response capabilities.

[0027] A heterogeneous, unmanned, clustered, three-dimensional security patrol system for open spaces on campus includes: The central control and decision-making module is used to receive daily inspection tasks and emergency alarms, complete the formal definition of tasks; call the event-task-capability mapping knowledge base to decompose complex events into atomic task sets; construct a multi-objective optimization model to complete the unified solution of task allocation and path planning; issue scheduling instructions and monitor the status of the robot in the whole domain in real time; trigger dynamic replanning to ensure the robustness of the system in emergency scenarios; and generate structured inspection reports. The event-task-capability mapping knowledge base module has built-in three-dimensional mapping rules for typical campus security events, atomic tasks, and robot capabilities. It is used to automatically output the set of atomic tasks and capability requirement vectors according to the event type, provide initial robot combination suggestions for optimized scheduling, and support incremental updates and rule expansion of the knowledge base. The robot status perception and acquisition module is used to collect the real-time location, remaining battery power, and working status of drones, unmanned vehicles, and robot dogs, and aggregate them to form a global status set, providing input for task allocation, and supporting unified access and format standardization of multiple robots and multiple types of data; The multi-objective optimization and task allocation module is used to establish a multi-constraint optimization model with time, energy consumption, and risk as optimization objectives. Based on the topology map, it performs fast path simulation and conflict hotspot estimation, executes selection, crossover, mutation genetic operations and elite local search, and outputs the optimal task allocation scheme. It supports warm-start initialization to enable rapid replanning; The conflict-free collaborative path planning module is used to provide customized solutions for drones, unmanned vehicles, and robotic dogs. Path finding is achieved using CBS conflict-oriented search to resolve multi-machine path conflicts, generating conflict-free and accurate spatiotemporal trajectories, and outputting a collaborative spatiotemporal path plan. .

[0028] The beneficial effects of this improvement are: it constructs an integrated architecture of five core modules: decision-making, knowledge base, perception, optimization, and path planning, realizing an intelligent closed loop for the entire process from event perception, task parsing, optimization scheduling to path generation; the decoupled design and collaborative work of each module make the system highly scalable, highly compatible, and easy to maintain, and can be flexibly adapted to different campus sizes and robot models, possessing extremely strong engineering implementation value.

[0029] Preferably, as an improvement, it also includes: The heterogeneous unmanned execution cluster module includes a drone execution unit, an unmanned vehicle execution unit, and a robot dog execution unit. The drone execution unit is used to perform wide-area aerial inspection, panoramic monitoring, rapid positioning, and three-dimensional spatial movement. The unmanned vehicle execution unit is used to perform ground main road patrol, material transportation, voice broadcasting, and two-dimensional road movement. The robot dog execution unit is used to perform fine inspection of indoor / corridor / complex terrain, obstacle crossing, climbing stairs, and legged adaptive movement. The wireless communication and data transmission module is used by the central platform to send task and path instructions to the robots, and each robot transmits video, sensor data, status, position and power in real time to ensure stable and reliable communication between multiple robots in parallel. The multi-source data fusion and situation assessment module is used to fuse heterogeneous data such as video, images, location, status, and alarms, and to judge the progress of task execution and abnormal conditions in real time, triggering replanning condition judgment. The dynamic replanning module is used to automatically start when new events occur, robot malfunctions occur, or power is insufficient. It uses historical plans as hot start information to quickly generate new collaborative plans, ensuring that inspection tasks are not interrupted and emergency responses are real-time. The automatic inspection report generation module is used to record event logs, task processes, robot trajectories, and anomaly snapshots, and automatically generate structured security inspection reports containing handling suggestions. It supports archiving, querying, and exporting.

[0030] The beneficial effects of this improvement are: it achieves three-dimensional full-area coverage through three types of execution units: air, ground, and indoor. Combined with communication, data fusion, dynamic replanning, and report generation modules, it forms a complete intelligent security system of perception, decision-making, execution, feedback, and iteration. Compared with traditional monitoring or single robot inspection modes, this system achieves three-dimensional security with full coverage, high intelligence, strong collaboration, and traceability, fundamentally improving the level of campus security intelligence and possessing outstanding technological creativity and advanced application. Attached Figure Description

[0031] Figure 1 This is an overall flowchart of an embodiment of the present invention.

[0032] Figure 2 This is a structural diagram of the two-stage collaborative solution method according to an embodiment of the present invention. Detailed Implementation

[0033] The following detailed description illustrates the specific implementation method: Example The basics are as follows: Figure 1 As shown: A heterogeneous unmanned cluster-based three-dimensional security patrol method for open spaces on campus, including: S1. Task Reception and Formal Definition: The central control platform receives daily inspection instructions or abnormal event alarms and performs structured and formal definitions of the tasks.

[0034] S101, The task is defined as a tuple. ; Among them, Type indicates the task type, such as daily inspection or emergency response; Priority indicates the task priority. This is the set of atomic tasks decomposed from the task; Loc represents the geographic location information associated with the task.

[0035] S102, Atomic task capability modeling, for each atomic task Define the corresponding capability requirement vector ; ; in, Minimum maneuver range required to complete this task; Minimum load capacity required to complete this task; The necessary space for passage to complete this task; These vector dimensions represent the capability vectors required to complete the task, representing the necessary sensor configuration for the task.

[0036] S103, Correspondingly, the capability state of each robot is quantized into a multi-dimensional vector. These characteristics respectively represent its maneuverability, load capacity, spatial mobility, and sensor configuration.

[0037] The platform has a built-in structured knowledge base that maps "event type - atomic task - robot capability" as the basic rule engine for decision-making.

[0038] This knowledge base pre-associates typical security incidents with robot capabilities and collaborative strategies. When an incident occurs, the system first calls this knowledge base to quickly generate preliminary capability requirements and robot combination suggestions based on the incident type, providing a decision-making basis for subsequent dynamic scheduling of multi-objective optimization models.

[0039] When an event occurs, the system uses this knowledge base to break down the event type into atomic tasks through rule-driven structured events and matches initial capability requirements with robot combination suggestions.

[0040] S2. The central control platform invokes the dynamic task allocation and collaborative planning algorithm. This algorithm performs the following sub-steps: S201. Real-time robot status information acquisition: Obtain the real-time status of all available robots, including UAVs, UGVs, and Quads. The system may have one or more UAVs, UGVs, or Quads. The system needs to acquire the real-time status information of all available robots and form a global status set S.

[0041] The global state set S is composed of the UAV state subset. Autonomous vehicle state subset and the robot dog state subset Together constitute, that is .

[0042] Each individual robot corresponds to a state. Each state ;in, This refers to the robot's location information; This represents the robot's current remaining battery power. This indicates the robot's current state, i.e., its current working status / task progress.

[0043] S202. Multi-objective optimization modeling: Based on the current task and robot state S, a constrained multi-objective optimization problem is constructed. The specific objective function is to minimize the global cost J.

[0044] in: A represents the set of available robots. ; This is a set of atomic tasks.

[0045] Let be a binary decision variable, representing whether atomic task t is assigned to the robot. . , , Robots The estimated execution time, energy consumption, and risk factor for task t.

[0046] and For normalization coefficients, represent the maximum estimated execution time for the robot to complete a single task in the preset campus scenario, and the maximum estimated energy consumption for the robot to complete a single task in the preset campus scenario, respectively. This is an adjustable weighting coefficient used to balance the importance of time, energy consumption, and risk. Its value is dynamically configured according to the task scenario, satisfying normalization constraints. .

[0047] S203. Constraint Settings: Optimization must satisfy the following core constraints. Task completion constraint: Each atomic task must be assigned one and only once. .

[0048] Capability matching constraint: Robot capability vector The task requirement vector must be satisfied. Key dimensions .

[0049] Energy constraint: The total energy consumption of all tasks assigned to a single robot must not exceed its remaining battery power. .

[0050] Spatiotemporal conflict-free constraints: The paths assigned to different robots do not overlap in space and time, by assigning a unique spatiotemporal window to each path segment. To ensure this. s represents the spatial position. The moment when robot a enters a specific spatial location s to begin performing related operations or traverses that path segment. The moment when robot a completes its operation at spatial location s or leaves that spatial location via that path segment.

[0051] S204. Problem Solving and Solution Generation: As attached Figure 2 As shown, a two-stage collaborative solution method combining "improved meme algorithm" and "conflict-oriented search (CBS)" is used to efficiently solve the objective function established in S202.

[0052] The first stage uses an improved meme algorithm to achieve global task allocation; the second stage uses... The algorithm performs fine-grained planning of conflict-free paths and ultimately outputs the optimal task allocation scheme. and collaborative spatiotemporal path planning .

[0053] Phase 1: Solving the task allocation problem based on the improved meme algorithm to obtain the optimal binary decision variable matrix. ; Initialization: Set algorithm parameters, including population size. Crossover probability Probability of mutation Local search probability Maximum Algebra and the number of elites .

[0054] Generate the initial population Each individual is encoded as a task assignment scheme, and the initial population is... This includes heuristically generated individuals and randomly generated individuals.

[0055] Heuristically generated individuals are created using a knowledge base mapping "event type - atomic task - robot capability" based on heuristic rules. This involves mapping the current event type and the robot capability vector. Task requirement vector As input, an initial allocation scheme is quickly generated based on the pre-defined rules of "event type - atomic task - robot capability" in the knowledge base.

[0056] Randomly generated individuals are achieved by randomly selecting a robot a for each task t, and the robot's capability vector is... ≥Task Requirement Vector The scheme is obtained by random allocation.

[0057] The initial population is obtained by combining some initial solutions generated through heuristic rules from the knowledge base with a randomly generated allocation scheme. This is to improve the quality of the initial population.

[0058] For the initial population Perform iterative optimization to obtain the g-th generation population after iterative optimization. , The iterative methods include: First, regarding the current generation of the population. Each individual Calculate its fitness The fitness formula is:

[0059] in, The value is calculated based on the objective function in S202; This is the sum of penalty values ​​for each hard constraint in S203 that the corresponding scheme for this individual violates. For example, the penalty settings for the hard constraints are accumulated based on the number of unassigned tasks, the gap value of the capability dimension, the gap value of the power supply, and the conflict hotspots. This is the penalty coefficient.

[0060] In a typical campus security patrol scenario (6-10 atomic tasks). The typical value range is 2.0 to 5.0; The penalty values ​​are designed so that individual penalties are roughly on the same order of magnitude (1~20), accumulating based on unassigned tasks, capability gaps, power exceeding limits, and conflict hotspots. In practical engineering, to ensure that the fitness of infeasible solutions is significantly lower than that of feasible solutions, thus allowing for natural elimination during evolution, and to maintain robustness and allow for minor violations to participate in the evolution, some marginal solutions are retained for search. The penalty value can be adjusted based on task complexity and the degree of violation. The value can be adjusted between 0.5 and 2.0.

[0061] Estimating the time it takes for robot a to perform task t Energy consumption At that time, a fast path simulation method based on a simplified topology map was adopted, and a coarse estimate of spatiotemporal conflicts was pre-introduced into the fitness calculation: Simplify topology map construction; the system pre-builds a topology map of the campus environment. , where nodes This represents key locations, including intersections, building entrances / exits, and task points, etc. This represents a feasible path segment, with each edge labeled with its average travel time. Energy consumption per unit distance This topology map ignores robot size, dynamic obstacles, and fine kinematic constraints, retaining only connectivity and macroscopic cost information to support fast path estimation.

[0062] Fast path simulation uses Dijkstra's algorithm to find the shortest topological path from the current position to each task point, and accumulates the travel time and energy consumption of each path segment to obtain the result. and The estimated value.

[0063] To address the coarse estimation of spatiotemporal conflicts and preemptively eliminate conflict-prone schemes at the fitness level, the system introduces a conflict hotspot penalty mechanism: This mechanism involves statistically analyzing shared nodes across all robot topology paths. and its estimated occupancy period .

[0064] If the same node If multiple paths are occupied within a similar timeframe, it is identified as a "potential conflict hotspot." This information is then applied to tasks involving that node. Additional risk penalties are imposed in the middle. This causes the fitness function to tend to choose an allocation scheme with more dispersed paths and a lower probability of conflict.

[0065] For example, the standard basic risk coefficient When there is no conflict, the basic risk for a single task is set at 0.1 to 0.3 to characterize the inherent risks of the task itself, such as high-altitude operations and nighttime inspections. Conflict penalty benchmark The base penalty value for overlapping time at the same node for a single group of robot paths is set to 1.0~2.0, approximately 5~10 times the base risk, to ensure penalty priority. Penalty cap To avoid excessively large penalty values ​​that could cause abnormal algorithm convergence, the upper limit is set to 5.0.

[0066] For each potential conflict hotspot node v, count the number of robots whose occurrence times overlap at that node. ,but: .

[0067] After completing the fitness calculation, the population is... Genetic operations are performed sequentially, including population selection, crossover, and mutation, specifically as follows: Population selection is based on roulette wheel selection and elite retention strategies, and on individual fitness. The probability of being selected is allocated proportionally; the higher the fitness, the greater the probability of being selected for the next generation of reproduction, while retaining the optimal [fitness / quality]. A select few individuals directly enter the next generation, forming a temporary population. .

[0068] Cross operations, for Individuals in the probability Perform a uniform crossover operation based on task blocks to divide tasks with strong coupling relationships such as temporal dependencies and spatial associations into overall task blocks. Genetic operations are performed with task blocks as the smallest unit to ensure that the allocation relationship of coupled tasks is not broken and to maintain the logical integrity of the tasks. Each time, two parent individuals are selected from the population, i.e., different task allocation schemes. Each task block independently decides to inherit from either parent with a 50% probability, generating two child individuals. This process is repeated until the number of child individuals is equal to the population size. Then, the selection operation is performed again to retain the best task-robot combination segments.

[0069] Mutation operation, with mutation probability The robot executing a task t is randomly replaced. The new robot starts from the one that satisfies... Select from the set and complete the mutation.

[0070] After selection, crossover, and mutation genetic operations, an updated offspring population is obtained. Furthermore, a number of individuals with high fitness were selected from the offspring population as elite individuals for subsequent local searches.

[0071] After the genetic operations are completed, a local search is performed. To reduce computational overhead, a customized local search is performed only on elite individuals, and the number of elite individuals is limited. This is a configurable parameter, and can be set to 5%-10% of the population size; specifically, it applies to the offspring population. Top of medium fitness ranking Individuals, in proportion The process of executing the local search operator is as follows: Path generation involves decoding the current elite individuals to obtain a task allocation scheme, and translating the abstract allocation scheme into an executable action plan with spatiotemporal trajectory; this is achieved using a method customized for collaborative scenarios involving heterogeneous robots on campus. The algorithm acts as a fast path planner, estimating the spatiotemporal trajectory of each robot a's task execution sequence. .

[0072] Custom The algorithm's inputs are: robot type, current position, target task point sequence, and campus spatiotemporal constraint map; The output consists of k candidate paths that satisfy kinematic feasibility, spatiotemporal constraint compliance, and optimal cost.

[0073] Conflict detection, iterating through all robot pairs Detecting whether there is a binary spacetime conflict Robot and The trajectories occupy the same map grid or vertex v at the same time t, and the first detected collision is recorded. .

[0074] Conflict resolution: If a conflict is detected, it is resolved by adjusting the task execution order or by meeting the requirements. Replace the execution robot in the set, generate a new task allocation scheme, and re-evaluate the fitness of the new scheme.

[0075] Acceptance and iteration: if the new solution has better fitness, replace the original individual with the new solution; repeat the above conflict detection and resolution process L times for the current individual to optimize the allocation scheme towards path feasibility and low conflict.

[0076] Population updates will retain elite individuals and optimize the offspring population through local search. Merging to form a new generation of populations This completes one full iteration.

[0077] Repeat the iterative process of fitness calculation, genetic operations, local search, and population update described above until the preset maximum number of iterations is reached. The iteration terminates when the fitness converges. At this point, the individual with the highest fitness is decoded, resulting in the optimal task allocation scheme. .

[0078] Phase Two: Based on Multi-agent conflict-free path planning The optimal task allocation scheme was obtained through the first-stage improvement of the meme algorithm. Next, the fine-grained path planning stage begins. This stage employs the Conflict-Oriented Search (CBS) algorithm to plan precise, conflict-free spatiotemporal paths that satisfy kinematic constraints for each robot with assigned tasks. The specific execution steps are as follows: The algorithm takes as input the optimal task allocation scheme output from the first stage. As input for this stage, based on Determine all the tasks that each robot needs to perform and the corresponding task locations.

[0079] Solving using the CBS algorithm: Underlying path planning, for each robot with an assigned task. With its current initial position Based on task allocation scheme The corresponding sequence in the task bag As the endpoint, an adaptive kinematic constraint is adopted for the robot itself. The algorithm plans and generates k shortest path candidates for each robot, ensuring that the paths are kinematically feasible and compliant within the spatiotemporal constraints of the campus.

[0080] Adapting to the robot's own kinematic constraints The algorithm is based on the standard A algorithm framework and is customized for the motion characteristics of three types of heterogeneous robots: drones, unmanned vehicles, and robot dogs. The specific customization method is as follows: Search space customization: Build exclusive navigable maps for different robots. Drones use 3D grid maps (marking no-fly zones and height-restricted zones), unmanned vehicles use 2D road topology maps (marking restricted zones and traffic lights), and robot dogs use 3D unstructured terrain maps (marking stairs and obstacles). Only the areas that robots can traverse are retained as search nodes. Cost function g(n) customization: Design a dedicated cost function for the robot's movement mode, incorporating multiple dimensions of cost such as path length / time, energy consumption, risk / stability, etc., to balance task efficiency, energy consumption and safety; Customized heuristic function h(n): Adopt an acceptable heuristic function that is adapted to the robot's motion mode (such as the three-dimensional Euclidean distance of the UAV, the shortest road distance of the unmanned vehicle, and the terrain-weighted distance of the robot dog) to ensure the optimality and search efficiency of the algorithm. Kinematic constraint verification: During the path search process, the feasibility of motion between adjacent nodes is verified in real time, including maximum turning angle, minimum turning radius, obstacle crossing height, slope limit, etc., and paths that do not conform to the robot's kinematic characteristics are eliminated. Campus spatial and temporal constraints integration: Constraints such as no-fly zones, restricted areas, and peak pedestrian hours on campus are integrated into the search process to ensure paths comply with campus security management rules. Through the above customized design, The algorithm can adapt to the motion characteristics of three types of heterogeneous robots, and plan k kinematically feasible, spatiotemporally compliant, and cost-optimal path candidates for each robot, providing a foundation for subsequent high-level conflict resolution in CBS.

[0081] High-level conflict resolution, establishing a constraint tree ( Using the underlying path planning results of all robots as the initial path set, construct the initial root node of the constraint tree. The root node contains the initial paths of all robots. The initial global solution cost is calculated using the CBS high-level cost function. ).

[0082] Construct the OPEN set, which is a priority queue sorted by node cost in ascending order. It is used to store all constraint tree nodes to be explored. The smaller the node cost, the higher the search priority.

[0083] Conflict detection involves retrieving the node N with the lowest current cost from the OPEN set and extracting the path set corresponding to that node. Iterate through all robot pairs, detect the first spatiotemporal conflict in the path, and denot it as... , indicating robot and The trajectories occupy the same spatial unit v at the same time t.

[0084] For constraint generation and node expansion, since the algorithm cannot predetermine which robot should avoid the detected spatiotemporal conflict to achieve the global optimum, a branch exploration strategy is adopted: Generate the first child node For robots Add spacetime constraints ,limit At time t, space unit v must not be occupied; Generate a second child node For robots Add spacetime constraints ,limit At time t, space unit v must not be occupied; replan the path for the constrained robot in each of the two child nodes and update the node path information.

[0085] Cost updates and node enqueueing are performed using the CBS high-level cost function to calculate the total cost of the new node. The cost function is as follows:

[0086] in, It's a robot. Path length, It's a robot. The time to complete the last task, This is a time weighting coefficient used to balance path length and execution time.

[0087] Add the two child nodes that have completed path update and cost calculation to the OPEN set, and continue to execute the conflict detection, constraint generation, and cost update process in a loop until a conflict-free feasible node appears in the OPEN set.

[0088] Spatiotemporal window encapsulation: When a node retrieved from the OPEN set has no spatiotemporal conflicts, the path set corresponding to that node is the optimal conflict-free path. The algorithm terminates and outputs the set of conflict-free paths. .

[0089] The system will have conflict-free paths Perform a standardized transformation, assigning an exclusive spatiotemporal window to each path segment of each robot, represented as a triple. ,in, For path segment numbering, The start time for entering this path segment. The final collaborative spatiotemporal path plan, defined by the end time of leaving the path segment, is then generated and can be directly assigned to the robot for execution. .

[0090] The optimal task allocation scheme output in the first stage Coordinated spatiotemporal path planning with the output of the second phase The combined solution is then issued and executed as the final collaborative plan for this scheduling process.

[0091] If a replanning condition is triggered during task execution, such as a new unexpected event, robot malfunction, insufficient power, or task change, the system will activate a hot-start dynamic replanning mechanism. The output of the previous round and As the most frequently started information; The robot's real-time position, remaining battery power, and unfinished task sequence are used as the latest global state as new inputs; The first-stage improved meme algorithm was rerun, and the initial population of the genetic algorithm was set to be adjusted based on the original scheme. By making full use of historical scheduling information, the iteration search time was greatly shortened, and a new adjusted collaborative scheme adapted to the current state was quickly generated to ensure the continuous and stable execution of the inspection task.

[0092] S3, The central control platform will use the optimal task allocation scheme generated by optimizing S2. With collaborative spatiotemporal path planning The task instructions and path constraints are simultaneously transmitted to the corresponding drones, unmanned vehicles and robot dogs in the solution through wireless communication links, ensuring that each heterogeneous robot receives a unique and matching task instruction and path constraint.

[0093] After receiving the collaborative instructions from the central control platform, the S4 drones, unmanned vehicles, and robotic dogs strictly follow the collaborative spatiotemporal path plan. The prescribed spatiotemporal window and path trajectory trigger the cooperative motion operation, and the specific motion execution method is as follows: S401. Drone Motion: Based on its corresponding path point sequence, the drone is driven by the onboard flight control system to operate the rotor motor, generating stable lift and horizontal thrust, and realizing point-to-point translational motion in three-dimensional space from the current real-time coordinates to the target mission coordinates, completing the aerial wide-area inspection displacement.

[0094] S402, Unmanned Vehicle Motion: The unmanned vehicle follows the planned path, with the wheel hub motors driven by the onboard controller providing the propulsion, and at the same time controlling the steering servo to adjust the heading. Under the constraints of campus roads, it completes two-dimensional rolling motion and achieves efficient cruising on the main ground roads.

[0095] S403, Robot Dog Motion: For paths with complex terrain including stairs, obstacles, and uneven ground, the main controller coordinates the servo motors of each leg joint to work together, dynamically adjusting gait parameters and foot trajectory to achieve adaptive leg movement in unstructured environments. It can stably cross obstacles, go up and down stairs, and complete precise inspection displacement in indoor and complex terrain areas.

[0096] S5. During the cooperative movement and after arriving at the corresponding task point, the heterogeneous robots follow the optimal task allocation scheme. Assigned by the robot, the robot performs specific atomic tasks, including but not limited to video data collection, campus facility status inspection, emergency voice broadcasting, and emergency material delivery. At the same time, the robot uploads and transmits task execution status, real-time audio and video, sensor data, its own location and power information to the central control platform in real time, providing data support for the platform's overall analysis.

[0097] S6: The central control platform uniformly receives, parses, integrates, and comprehensively analyzes the multi-source heterogeneous data transmitted from each robot, and monitors the overall inspection situation and task execution progress in real time. If new, higher-level alarms, sudden abnormal events, changes in task priorities, or abnormal robot states are triggered during task execution, the system immediately returns to S2 and initiates a hot-start dynamic replanning mechanism. Based on the latest global status information, including the robot's real-time position, remaining battery power, unfinished tasks, and new events, the system reconstructs and solves the multi-constraint optimization model, quickly generating an adjusted collaborative solution adapted to the current scenario, ensuring the robustness of the inspection system and the real-time nature of emergency response.

[0098] S7. After all inspection or emergency response tasks are completed, the central control platform will summarize, organize and store the data of the entire inspection process and automatically generate a standardized and structured security inspection report. The report includes the full life cycle log of the event and task, the movement trajectory of each robot, snapshots of abnormal data, the handling process of abnormal events and optimization suggestions, providing complete and traceable records and decision support for campus security management.

[0099] A heterogeneous, unmanned, clustered, three-dimensional security patrol system for open spaces on campus includes: The central control and decision-making module receives daily inspection tasks and emergency alarms, and completes the formal definition of tasks; it calls the event-task-capability mapping knowledge base to decompose complex events into sets of atomic tasks; it constructs a multi-objective optimization model to complete the unified solution of task allocation and path planning; it issues scheduling instructions to monitor the status of the robot across the entire domain in real time; it triggers dynamic replanning to ensure the robustness of the system in emergency scenarios; and it generates structured inspection reports.

[0100] The event-task-capability mapping knowledge base module has built-in three-dimensional mapping rules for typical campus security events, atomic tasks, and robot capabilities; it automatically outputs atomic task sets and capability requirement vectors based on event types; it provides initial robot combination suggestions for optimized scheduling; and it supports incremental updates and rule expansion of the knowledge base.

[0101] The robot status perception and acquisition module collects the real-time location, remaining battery power, and working status of drones, unmanned vehicles, and robot dogs; it aggregates these data to form a global status set, providing input for task allocation; and it supports unified access and standardized formatting of multiple robots and multiple types of data.

[0102] The multi-objective optimization and task allocation module establishes a multi-constraint optimization model with time, energy consumption, and risk as optimization objectives; performs fast path simulation and conflict hotspot estimation based on a topology map; executes selection, crossover, mutation genetic operations, and elite local search; and outputs the optimal task allocation scheme. It supports warm-start initialization, enabling rapid replanning.

[0103] The conflict-free collaborative path planning module provides customized solutions for drones, unmanned vehicles, and robotic dogs. Path solving; CBS conflict-oriented search is used to resolve multi-machine path conflicts; conflict-free accurate spatiotemporal trajectories are generated; and a cooperative spatiotemporal path plan is output. .

[0104] The heterogeneous unmanned execution cluster module comprises three types of sub-units: The drone execution unit enables wide-area aerial inspection, panoramic monitoring, rapid positioning, and three-dimensional spatial movement.

[0105] The unmanned vehicle execution unit is used for cruising on main ground roads, transporting materials, providing voice broadcasts, and moving on two-dimensional roads.

[0106] The robot dog's execution unit enables precise inspection, obstacle crossing, stair climbing, and adaptive legged movement in indoor / corridor / complex terrain environments.

[0107] The wireless communication and data transmission module enables the central platform to issue task and path instructions to the robots; each robot transmits video, sensor data, status, position, and battery level back in real time, ensuring stable and reliable communication between multiple robots concurrently.

[0108] The multi-source data fusion and situation assessment module integrates heterogeneous data such as video, images, location, status, and alarms; it can judge the progress of task execution and abnormal conditions in real time; and it can trigger replanning conditions.

[0109] The dynamic replanning module automatically starts when new events occur, robot malfunctions occur, or power is insufficient; it uses historical plans as warm-up information to quickly generate new collaborative plans; ensuring uninterrupted inspection tasks and real-time emergency response.

[0110] The automatic inspection report generation module records event logs, task flows, robot trajectories, and anomaly snapshots; it automatically generates structured security inspection reports with handling suggestions; and it supports archiving, querying, and exporting.

[0111] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus, characterized in that: include: The central control platform receives inspection instructions or abnormal event alarms, and based on the built-in event type-atomic task-robot capability mapping knowledge base, it breaks down complex events into sets of atomic tasks. And define a capability requirement vector for each atomic task t. ; The real-time states of drones, unmanned vehicles, and robot dogs are collected to form a global state set S. Based on the current task and robot state, a constrained multi-objective optimization problem is constructed. A two-stage collaborative solution method combining an improved meme algorithm and conflict-oriented search is adopted to output the optimal task allocation scheme. and collaborative spatiotemporal path planning ; The central control platform will optimize the task allocation scheme. and collaborative spatiotemporal path planning The data is simultaneously transmitted to the corresponding drones, unmanned vehicles, and robot dogs via wireless communication links. After receiving the instructions, each heterogeneous robot follows the collaborative spatiotemporal path plan. The specified time and space window and path trajectory initiate coordinated motion operations; Each heterogeneous robot is assigned a task according to the optimal task allocation scheme. The system assigns atomic tasks and uploads the task execution status, real-time audio and video, sensor data, location and power information to the central control platform in real time. The central control platform integrates and analyzes the returned data. When the replanning condition is triggered, it uses historical solutions as hot start information, reconstructs and solves the multi-constraint optimization model based on the current global state, generates the adjusted collaborative solution, and sends it out for execution. After the task is completed, the central control platform summarizes all the data from the entire process.

2. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 1, characterized in that, In dynamic task allocation and collaborative planning, the objective function of the multi-objective optimization problem is to minimize the global cost J: Where A is the set of available robots; For a set of atomic tasks, Let be a binary decision variable, representing whether atomic task t is assigned to the robot. ; , , Robots The estimated execution time, energy consumption, and risk factor for task t; and These are the normalization coefficients; These are adjustable weighting coefficients.

3. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 2, characterized in that, In dynamic task allocation and collaborative planning, the multi-objective optimization problem must satisfy the following constraints: Task completion constraint: Each atomic task must be assigned one and only once; Capability matching constraint, if Then the robot's capability vector The task requirement vector must be satisfied. ; Energy constraint: The total energy consumption of all tasks assigned to a single robot must not exceed its remaining power. Spatiotemporal conflict-free constraints ensure that the paths assigned to different robots do not overlap in space and time, which is guaranteed by assigning a unique spatiotemporal window to each path segment.

4. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 3, characterized in that: Two-stage collaborative solution methods include: In the first stage, the task allocation is solved based on the improved meme algorithm to obtain the optimal binary decision variable matrix. ; In the second stage, a multi-agent conflict-free path planning based on conflict-oriented search (CBS) is performed, using the optimal task allocation scheme output from the first stage. As input, it plans precise, conflict-free spatiotemporal paths for each robot that satisfy kinematic constraints, and outputs a cooperative spatiotemporal path plan. .

5. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 4, characterized in that, The first stage of task allocation based on the improved meme algorithm includes: Initialization, generating the initial population This includes individuals generated based on heuristic rules from a mapping knowledge base and randomly generated individuals; Fitness calculation: Calculate the fitness of each individual i in the population. ,in The objective function value, The sum of penalty values ​​for violating hard constraints, where λ is the penalty coefficient; Genetic operations are performed sequentially, including population selection based on roulette wheel selection and elite retention strategies, uniform crossover based on task blocks, and mutation probability-based selection. Randomly replace the mutation operation performed by the robot; Local search involves performing customized local searches on elite individuals, including path generation, conflict detection, conflict resolution, and acceptance and iteration processes, optimizing the allocation scheme towards path feasibility and low conflict. Iterative updates are performed, repeatedly executing fitness calculations, genetic operations, and local searches until the maximum number of iterations is reached or fitness convergence is achieved. The optimal task allocation scheme is obtained by decoding the individual with the highest fitness. .

6. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 5, characterized in that: In the fitness calculation, the time for robot a to perform task t is estimated. Energy consumption At that time, a fast path simulation method based on a simplified topology map is adopted, and a conflict hotspot penalty mechanism is introduced: The estimated occupancy time of shared node v in all robot topology paths is calculated. If the same node v is occupied by multiple paths in a similar time period, additional risk penalties are applied to related tasks involving that node. .

7. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 6, characterized in that, The second phase, based on conflict-oriented search (CBS) multi-agent conflict-free path planning, includes: The underlying path planning, for each robot 'a' with an assigned task, employs... The algorithm plans k shortest path candidates, and the adaptation includes search space customization, cost function customization, heuristic function customization, kinematic constraint verification, and integration of campus spatiotemporal constraints. To resolve high-level conflicts, a constraint tree (CT) is established, using the underlying path planning results of all robots as the initial root node. A branch exploration strategy is then employed to address detected spatiotemporal conflicts. Generate two child nodes, one for the robot. and Add a spatiotemporal constraint (a,v,t) to restrict it from occupying spatial unit v at time t, and replan the path for the constrained robot; Iterative solution: repeatedly execute conflict detection, constraint generation, and cost update processes until conflict-free feasible nodes are obtained, and output a set of conflict-free paths. And convert it into a collaborative spatiotemporal path plan .

8. The heterogeneous unmanned cluster three-dimensional security patrol method for open spaces on campus according to claim 7, characterized in that: In situation assessment and dynamic replanning, the replanning conditions include new emergencies, robot malfunctions, insufficient power, changes in task priority, or abnormal task execution. When the replanning condition is triggered, the system initiates a warm-start dynamic replanning mechanism: using the optimal task allocation scheme output from the previous round. With collaborative spatiotemporal path planning As warm-up information, the robot's real-time position, remaining battery power, and unfinished task sequence are used as new inputs to rerun the improved meme algorithm and set the initial population to be adjusted based on the original scheme, so as to quickly generate an adjusted collaborative scheme adapted to the current state.

9. A heterogeneous unmanned cluster three-dimensional security patrol system for open spaces on campus, used to implement the method described in any one of claims 1-8, characterized in that, include: The central control and decision-making module is used to receive daily inspection tasks and emergency alarms, and to complete the formal definition of tasks; The event-task-capability mapping knowledge base is invoked to break down composite events into sets of atomic tasks. A multi-objective optimization model is constructed to achieve a unified solution for task allocation and path planning; scheduling instructions are issued and the status of the robot across the entire domain is monitored in real time; dynamic replanning is triggered to ensure the system's robustness in emergency scenarios; Generate structured inspection reports; The event-task-capability mapping knowledge base module has built-in three-dimensional mapping rules for typical campus security events, atomic tasks, and robot capabilities. It is used to automatically output the set of atomic tasks and capability requirement vectors according to the event type, provide initial robot combination suggestions for optimized scheduling, and support incremental updates and rule expansion of the knowledge base. The robot status perception and acquisition module is used to collect the real-time location, remaining battery power, and working status of drones, unmanned vehicles, and robot dogs, and aggregate them to form a global status set, providing input for task allocation, and supporting unified access and format standardization of multiple robots and multiple types of data; The multi-objective optimization and task allocation module is used to establish a multi-constraint optimization model with time, energy consumption, and risk as optimization objectives. Based on the topology map, it performs fast path simulation and conflict hotspot estimation, executes selection, crossover, mutation genetic operations and elite local search, and outputs the optimal task allocation scheme. It supports warm-start initialization to enable rapid replanning; The conflict-free collaborative path planning module is used to provide customized solutions for drones, unmanned vehicles, and robotic dogs. Path finding is achieved using CBS conflict-oriented search to resolve multi-machine path conflicts, generating conflict-free and accurate spatiotemporal trajectories, and outputting a collaborative spatiotemporal path plan. .

10. The heterogeneous unmanned cluster three-dimensional security patrol system for open spaces on campus according to claim 9, characterized in that, Also includes: The heterogeneous unmanned execution cluster module includes a drone execution unit, an unmanned vehicle execution unit, and a robot dog execution unit. The drone execution unit is used to perform wide-area aerial inspection, panoramic monitoring, rapid positioning, and three-dimensional spatial movement. The unmanned vehicle execution unit is used to perform ground main road patrol, material transportation, voice broadcasting, and two-dimensional road movement. The robot dog execution unit is used to perform fine inspection of indoor / corridor / complex terrain, obstacle crossing, climbing stairs, and legged adaptive movement. The wireless communication and data transmission module is used by the central platform to send task and path instructions to the robots, and each robot transmits video, sensor data, status, position and power in real time to ensure stable and reliable communication between multiple robots in parallel. The multi-source data fusion and situation assessment module is used to fuse heterogeneous data such as video, images, location, status, and alarms, and to judge the progress of task execution and abnormal conditions in real time, triggering replanning condition judgment. The dynamic replanning module is used to automatically start when new events occur, robot malfunctions occur, or power is insufficient. It uses historical plans as hot start information to quickly generate new collaborative plans, ensuring that inspection tasks are not interrupted and emergency responses are real-time. The automatic inspection report generation module is used to record event logs, task processes, robot trajectories, and anomaly snapshots, and automatically generate structured security inspection reports containing handling suggestions. It supports archiving, querying, and exporting.

Images