Multi-agent swarm intelligence cooperative control method and system for special limited space

Abstract

This invention provides a multi-agent swarm intelligent collaborative control method and system for special confined spaces. The method includes: spatiotemporally aligning the local perception information of each agent within the special confined space with the state information of neighboring agents to obtain unified spatiotemporal perception information for each agent; based on the unified spatiotemporal perception information, each agent performs swarm collaborative decision-making based on local information interaction with neighboring agents to obtain a swarm decision result; triggering judgment based on the swarm decision result and preset event-triggered communication conditions, and updating the state of each agent based on the triggering judgment result to obtain a neighboring agent state information group that is updated as needed; and decoupling the collision risk between multiple agents at the conflict level based on the swarm decision result and the neighboring agent state information group to obtain conflict-free cooperative motion constraints. This invention is adapted to the complex operating environment requirements of special confined spaces.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a multi-agent swarm intelligent collaborative control method and system for special confined spaces. Background Technology

[0002] Special confined spaces are critical scenarios in industrial production, urban operations, and safety assurance, such as underground utility tunnels, mine tunnels, oil and gas chambers, and confined spaces in nuclear power plants. They generally have typical characteristics such as being enclosed and narrow, having complex spatial structures, low light and dust, dense obstacles, easy communication interruptions, lack of satellite positioning (GPS denial), and high environmental risks. Traditional manual operation methods are prone to safety accidents such as poisoning, suffocation, collapse, and explosion. Therefore, multi-agent collaborative operation has become an inevitable trend to replace manual labor and improve operational safety and efficiency.

[0003] Currently, the most common multi-agent cooperative control method is the distributed cooperative control method based on periodic communication. Each agent continuously broadcasts its own state information at a fixed frequency, and completes distributed decision-making and obstacle avoidance planning by sharing local information, achieving basic multi-agent cooperation in an open and well-communicationed environment.

[0004] However, existing methods suffer from poor compatibility between periodic communication mechanisms and traditional swarm intelligence algorithms when applied to special confined spaces, leading to collaborative failure. Specifically, the wireless communication channel capacity in special confined spaces is limited. Fixed-frequency periodic broadcasts generate a large amount of redundant data, quickly causing channel congestion and resulting in packet loss and delays in neighboring machine status information. Traditional swarm intelligence algorithms are highly sensitive to the completeness and real-time nature of input information. Missing information prevents them from accurately assessing collision risks and optimal paths, leading to multi-agent path oscillations and multi-machine deadlocks. Path oscillations further exacerbate communication demands and channel congestion, ultimately causing the entire collaborative system to fail and making it unsuitable for the operational requirements of special confined spaces. Summary of the Invention

[0005] This invention provides a multi-agent swarm intelligent collaborative control method and system for special confined spaces, which solves the technical problem of collaborative failure of distributed collaborative control methods based on periodic communication in special confined spaces in the background art, and improves the safety and continuity of multi-agent collaborative operation in special confined spaces, thereby adapting to the complex operating environment requirements of special confined spaces.

[0006] In a first aspect, the present invention provides a multi-agent swarm intelligent cooperative control method for special confined spaces, comprising: Based on the spatiotemporal synchronization and alignment of the local perception information of each intelligent agent with the state information of neighboring machines in a special limited space, the unified spatiotemporal perception information of each intelligent agent is obtained. Based on the unified spatiotemporal perception information, each intelligent agent makes collaborative group decisions based on local information interaction with neighboring machines, and obtains the group decision results; Based on the group decision-making results and preset event-triggered communication conditions, a trigger judgment is made, and the state of each agent is updated based on the trigger judgment results to obtain a neighboring machine state information group that is updated as needed. Based on the group decision-making results and the neighbor machine state information group, the collision risk between multiple agents is decoupled at the conflict level to obtain conflict-free cooperative motion constraints. Based on the cooperative motion constraints and the unified spatiotemporal perception information, task allocation and cooperative trajectory planning are performed to generate collision-free cooperative operation trajectories for each agent, and each agent executes cooperative operation motion based on the instructions transformed from the collision-free cooperative operation trajectories.

[0007] Secondly, the present invention also provides a multi-agent swarm intelligent cooperative control system for special confined spaces, applied to the multi-agent swarm intelligent cooperative control method for special confined spaces as described in the first aspect; the multi-agent swarm intelligent cooperative control system for special confined spaces includes: The multi-source perception and spatiotemporal alignment module is used to perform spatiotemporal synchronization and alignment based on the local perception information of each intelligent agent in a special limited space and the state information of neighboring machines, so as to obtain the unified spatiotemporal perception information of each intelligent agent. The group decision-making module is used to make collaborative group decisions based on the unified spatiotemporal perception information and the local information interaction with neighboring machines, so as to obtain the group decision-making result. The event-triggered update module is used to make a trigger judgment based on the group decision result and the preset event-triggered communication conditions, and update the state of each agent based on the trigger judgment result to obtain the neighbor state information group that is updated as needed. The conflict-level decoupling module is used to decouple the collision risk between multiple agents based on the group decision results and the neighbor machine state information group, so as to obtain conflict-free cooperative motion constraints. The task allocation and collaborative trajectory planning module is used to allocate tasks and plan collaborative trajectories based on the collaborative motion constraints and the unified spatiotemporal perception information, generate collision-free collaborative operation trajectories for each intelligent agent, and enable each intelligent agent to execute collaborative operation motion based on the instructions transformed from the collision-free collaborative operation trajectory.

[0008] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the multi-agent swarm intelligent cooperative control method for special confined spaces as described above.

[0009] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the multi-agent swarm intelligent cooperative control method for special confined spaces as described above.

[0010] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the multi-agent swarm intelligent cooperative control method for special confined spaces as described above.

[0011] The multi-agent swarm intelligent collaborative control method for special confined spaces provided in this invention solves the problems of disorganized and inaccurate information utilization in traditional methods by synchronizing and aligning the local perception information of each agent with the state information of neighboring machines in a spatiotemporal manner. Based on this unified spatiotemporal perception information, each agent completes swarm collaborative decision-making through local information interaction with neighboring machines, obtaining swarm decision results without relying on a central node for global scheduling. Trigger judgment is performed based on the swarm decision results and preset event-triggered communication conditions, and neighboring machine state information groups are updated as needed. This breaks the traditional periodic broadcast mode, reduces redundant data generation, and alleviates channel congestion at its source. The machine state information group decouples the collision risks among multiple agents at the conflict level, obtaining conflict-free cooperative motion constraints. This solves the problem of inaccurate collision risk assessment caused by information gaps in traditional algorithms, and avoids path oscillation and multi-machine deadlock. Based on cooperative motion constraints and unified spatiotemporal perception information, task allocation and cooperative trajectory planning are performed to generate collision-free cooperative operation trajectories for each agent and convert them into instructions for execution. This ensures smooth and orderly cooperative operation among multiple agents, ultimately solving the technical problem of cooperative failure of distributed cooperative control methods based on periodic communication in special confined spaces. This improves the safety and continuity of multi-agent cooperative operation in special confined spaces, thus adapting to the complex operating environment requirements of special confined spaces. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the multi-agent swarm intelligent collaborative control method for special confined spaces provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of a multi-agent swarm intelligent collaborative control system for special confined spaces provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] See Figure 1 , Figure 1 This is a flowchart illustrating the multi-agent swarm intelligent cooperative control method for special confined spaces provided by the present invention. In this embodiment, the executing entity of the multi-agent swarm intelligent cooperative control method for special confined spaces is the cooperative control system. Therefore, the multi-agent swarm intelligent cooperative control method for special confined spaces includes: Step 10: Based on the local perception information of each agent in the special limited space and the state information of the neighboring machine, perform spatiotemporal synchronization and alignment to obtain the unified spatiotemporal perception information of each agent.

[0015] Optionally, the collaborative control system connects to each intelligent agent deployed within a special confined space and acquires local perception information and neighboring agent status information in real time based on the lidar, infrared sensors, vision cameras, and inertial measurement units onboard each agent. This local perception information includes local obstacle information and the agent's own pose information. The obstacle information includes the obstacle's position (spatial coordinates), size (length, width, and height), and motion state (stationary or moving direction, speed). The pose information includes the agent's own position (spatial coordinates), orientation (0-360 degree facing angle), and speed (speed and direction in meters per second). The neighboring agent status information is acquired through short-range wireless or wired communication and includes the neighboring agent's position (spatial coordinates), operational status (awaiting operation, currently operating, paused operation, or downtime due to fault), and remaining energy consumption (kilowatt-hours, reflecting the duration of continuous operation).

[0016] Furthermore, due to the influence of sensing device errors, deployment location differences, and communication delays, local sensing information and neighboring machine status information may be out of sync in time and space. Direct use of these discrepancies will lead to decision-making biases, thus requiring spatiotemporal synchronization and alignment. Specifically, a unified benchmark is first established. For example, the time benchmark uses the system startup time as zero point, with millisecond-level timing; the spatial benchmark uses a preset fixed reference point as the origin, establishing a three-dimensional spatial coordinate system to unify position representation. Then, the collaborative control system sends calibration commands to each agent, unifies the timing module, and simultaneously timestamps and calibrates the local sensing information and neighboring machine status information, discarding data with a deviation exceeding 10 milliseconds and compensating for data within 10 milliseconds to ensure time synchronization. In terms of spatial synchronization, the collaborative control system calibrates the position of each agent using the three-dimensional coordinate system, correcting sensing errors (deviation ≤ 5 cm). Simultaneously, coordinate transformation and calibration are performed on the positions of neighboring machines and obstacles to ensure all position information is based on a unified spatial benchmark. Finally, after synchronization is completed, the calibrated local perception information (including obstacle information and pose information) is integrated with the neighboring machine status information (including position, operation status, and remaining energy consumption) to form unified spatiotemporal perception information for each intelligent agent.

[0017] In one embodiment, taking the underground utility tunnel scenario in a special confined space (closed and narrow, with low light, easy communication interruption, and no satellite positioning) as an example, five intelligent agents equipped with lidar, inertial measurement units, and short-range wireless communication modules are deployed for inspection.

[0018] Each agent collects local obstacle information via lidar, such as protrusions on the inner wall of the utility tunnel (location: (10, 0.5, 2) meters, size: 1×0.3×0.4 meters, stationary) and maintenance tools (location: (15, 0.8, 1.5) meters, size: 0.5×0.3×0.2 meters, stationary); and collects its own pose information via inertial measurement unit. Agent 1's pose: position (5, 0.6, 2) meters, attitude 180 degrees (facing depth), speed 0.3 m / s along the axis towards depth; Agent 2 (adjacent to Agent 1)'s pose: position (7, 0.6, 2) meters, attitude 180 degrees, speed 0.3 m / s along the axis towards depth.

[0019] Through short-range wireless communication, Agent 1 obtains the status of Agent 2: Location (7, 0.6, 2) meters, currently inspecting, remaining energy consumption 5 kWh; Agent 2 obtains the status of Agent 1: Location (5, 0.6, 2) meters, currently inspecting, remaining energy consumption 4.8 kWh.

[0020] The collaborative control system establishes a unified spatiotemporal reference: time is based on startup as zero point, with millisecond-level timing; space is based on the entrance of the utility tunnel as the origin, establishing a three-dimensional coordinate system (X-axis along the axis towards the depth, Y-axis horizontal and perpendicular to the axis, Z-axis perpendicular to the horizontal plane). The collaborative control system sends time calibration commands to unify the timing of each intelligent agent. Local sensing information and neighboring machine status information are timestamped; for example, the maintenance tool information collected by agent 1 has an 8-millisecond deviation, which is compensated and synchronized. All position information is calibrated based on the preset coordinate system, correcting the position deviation of agent 1 (originally (5.003, 0.602, 2.001) meters, calibrated to (5, 0.6, 2) meters).

[0021] Ultimately, Agent 1's unified spatiotemporal perception information is as follows: local obstacle information (inner wall protrusion: position (10, 0.5, 2) meters, size 1×0.3×0.4 meters, stationary; maintenance tool: position (15, 0.8, 1.5) meters, size 0.5×0.3×0.2 meters, stationary), its own pose information (position (5, 0.6, 2) meters, attitude 180 degrees, speed 0.3 m / s along the positive X-axis), and neighboring machine status information (Agent 2: position (7, 0.6, 2) meters, currently inspecting, remaining energy consumption 5 kWh). The other four agents similarly complete spatiotemporal synchronization alignment.

[0022] Step 20: Based on unified spatiotemporal perception information, each intelligent agent conducts group collaborative decision-making based on local information interaction with neighboring machines to obtain the group decision-making result.

[0023] Optionally, the collaborative control system controls each agent to interact with neighboring machines based on unified spatiotemporal perception information, exchanging only core content to avoid channel congestion. After receiving information from neighboring machines, each agent combines its own information to make collaborative group decisions, resulting in a group decision outcome, as detailed in steps 201 to 204. The group decision outcome includes the task execution order and speed adjustment range of each agent. The task execution order clarifies the sequence of tasks performed by each agent, avoiding conflicts and duplication; and the speed adjustment range includes acceleration, deceleration, or maintenance to adapt to the needs of group collaboration.

[0024] Step 30: Based on the group decision-making results and the preset event triggering communication conditions, a trigger judgment is made, and the state of each agent is updated based on the trigger judgment results to obtain the neighbor state information group that is updated as needed.

[0025] Optionally, to adapt to the channel capacity limitations of special confined spaces and avoid fixed-frequency broadcast congestion, the cooperative control system presets event-triggered communication conditions, which include the following conditions: 1. Changes in self-state exceeding thresholds: position change > 0.5 meters, attitude change > 10 degrees, speed change > 0.1 meters / second; or changes in work status (such as changing from working to paused / faulted, or from waiting to working to working). 2. Neighbor status is missing or lagging: Neighbor status has not been updated for more than 300 milliseconds or cannot be obtained at all; 3. Group decision-making requires updates: Decision results must be explicitly required to update the status and synchronize (e.g., adjustments to task order or significant speed adjustments); 4. Triggered by environmental changes: the appearance of new obstacles, changes in the motion state of existing obstacles, and sudden changes in communication signal strength (change > 30%).

[0026] The collaborative control system triggers each agent's status based on the group decision-making results and the communication conditions triggered by the event. Specifically, it collects its own status (pose, operational status, remaining energy consumption) in real time, compares it with the previous transmitted status, and determines whether it exceeds a threshold; it monitors the update time of neighboring agents' status to determine if it is missing or lagging; it combines the group decision-making results to determine if there is an update requirement; and it determines whether the environment has changed based on local perception information. When any condition is met, the agent updates its own status (latest pose, operational status, remaining energy consumption) and broadcasts it to neighboring agents. Agents that do not meet the conditions do not broadcast but only update the received neighboring agent status information. The information received by each agent from neighboring agents is integrated and replaced, and lagging and invalid information is removed. Finally, a group of neighboring agent status information that is updated on demand is formed.

[0027] In one embodiment, continuing the underground utility tunnel inspection scenario, the preset trigger conditions are: a position change threshold of 0.5 meters, an attitude of 10 degrees, and a speed of 0.1 meters per second; a neighboring machine status update time limit of 300 milliseconds; and environmental changes triggering the process as new obstacles, changes in the movement state of obstacles, or a sudden change in signal strength >30%. Assume the resulting group decision is as follows: Agent 1's task order is "left then right," with its speed increasing by 0.1 meters per second (to 0.4 meters per second); Agent 2's task order is "right then left," with its speed remaining at 0.3 meters per second; and Agents 3-5's tasks and speeds remain unchanged.

[0028] Agent 1's speed change reaches a threshold and the decision requires an update, thus meeting the conditions and broadcasting an update. Agents 2-5, with no significant changes in state, normal neighbor information, and no environmental changes, do not meet the conditions and do not broadcast. Agent 1's status is updated to: position (5, 0.6, 2) meters, attitude 180 degrees, speed 0.4 m / s along the positive X-axis, currently inspecting, remaining energy 4.8 kWh, and sent to neighboring agents (Agents 2 and 3). Agents 2-5 do not broadcast. Agent 2 receives Agent 1's update information, replaces its original information, and forms a neighbor status information group (including Agent 1's latest status: position (5, 0.6, 2) meters, speed 0.4 m / s, currently inspecting, remaining energy 4.8 kWh). Agent 3 updates similarly. The neighbor status information groups for agents 4 and 5 remain unchanged. Finally, all five agents receive the updated neighbor status information groups as needed.

[0029] Step 40: Based on the group decision-making results and the state information of neighboring machines, the collision risk between multiple agents is decoupled at the conflict level to obtain conflict-free cooperative motion constraints.

[0030] Optionally, the cooperative control system takes the group decision-making results and neighboring machine status information as input, identifies collision risks (such as the possibility of physical collision caused by overlapping positions or intersecting trajectories), and adopts conflict-level decoupling to ultimately form conflict-free cooperative motion constraints. These conflict-free cooperative motion constraints include a set of motion restriction instructions, spatial restricted boundaries, time passage windows, heading angle limits, and safety distance parameters, as detailed in steps 401 to 404. Specifically, the motion restriction instruction set refers to the specific instructions that restrict the agent's movement speed and acceleration; the spatial restricted boundaries refer to the areas where the agent is prohibited from entering within a special confined space; the time passage window refers to the allowed time period for the agent to pass through a specific area; the heading angle limit refers to the allowed range of heading angles for the agent; and the safety distance parameter refers to the minimum distance that the agent must maintain between itself and neighboring machines or obstacles.

[0031] Step 50: Based on cooperative motion constraints and unified spatiotemporal perception information, task allocation and cooperative trajectory planning are performed to generate collision-free cooperative operation trajectories for each agent, and each agent executes cooperative operation motion based on the instructions transformed from the collision-free cooperative operation trajectory.

[0032] Optionally, the cooperative control system performs task allocation and cooperative trajectory planning based on conflict-free cooperative motion constraints and unified spatiotemporal perception information. In the task allocation pool, it integrates the task execution order of each agent in the group decision results, balances the overall task allocation, avoids uneven load, and plans a unique collision-free operation trajectory for each agent to ensure that the trajectories do not intersect or overlap, and complete the task efficiently, as shown in steps 501 to 505.

[0033] Optionally, the collaborative control system generates a collision-free collaborative operation trajectory containing information such as position, attitude, and speed through trajectory planning, and converts it into digital control commands to specify the direction of movement, speed, and turning angle. After conversion, the commands are sent to each intelligent body for execution, and the motion status and operation progress are monitored in real time to ensure the stability and safety of operation in high-risk and complex environments.

[0034] The embodiments of the present invention solve the technical problem of collaborative failure of distributed cooperative control methods based on periodic communication in special confined spaces in the background art, and realize the improvement of the safety and continuity of multi-agent cooperative operation in special confined spaces, thereby adapting to the requirements of complex operating environment in special confined spaces.

[0035] Optionally, the processes of steps 201 to 204 include: Step 201: Based on the location and size of the obstacle in the unified spatiotemporal perception information and the position and posture of the agent, perform spatial relative coordinate mapping to obtain a real-time avoidance gap distribution map of each agent relative to the local obstacle. Based on the real-time avoidance gap distribution map and a preset safety threshold, identify the narrow passage bottleneck with limited width to obtain a list of narrow section nodes.

[0036] Optionally, the collaborative control system extracts the obstacle location and size from the unified spatiotemporal perception information, as well as its own position and orientation from the local obstacle information. Using the agent's own position as the origin and its own orientation as the positive direction, the relative coordinate system direction is determined, and the obstacle position in the local obstacle information is converted into relative coordinates in this relative coordinate system. Combined with the obstacle size, the avoidable space distance between the agent and each local obstacle in each direction is calculated, forming a real-time avoidance gap distribution map. This real-time avoidance gap distribution map is a two-dimensional distribution map centered on the agent, which can intuitively present the safe passage distance between the agent and each local obstacle in each surrounding direction. The distribution map labels the relative position, obstacle size, and corresponding avoidance gap distance for each obstacle. The avoidance gap distance refers to the shortest straight-line distance from the edge of the agent to the edge of the obstacle, reflecting the passable space margin.

[0037] Optionally, the collaborative control system retrieves a preset safety threshold, which is the minimum clearance distance required for an agent to avoid collisions during passage. This threshold is determined based on the maximum external dimensions of the agent, typically set to 1.2 times its maximum width, to ensure that collisions are avoided even with minor positional deviations during passage. Then, based on a real-time clearance distribution map, the clearance distances around each agent are checked one by one, and areas with clearances less than the preset safety threshold are identified as narrow passage bottlenecks with limited width. A narrow passage bottleneck refers to an area within a special confined space where passage is restricted due to obstacles or the space itself being narrow, allowing only a small number of agents to pass sequentially. All identified narrow passage bottlenecks are located, and the specific spatial coordinates (using the unified three-dimensional spatial coordinate system established in step 10) and extension range (e.g., length, width, height) of each bottleneck are recorded and uniquely identified, forming a list of narrow segment nodes. This list is structured data; each entry corresponds to a narrow passage bottleneck and includes the bottleneck's unique identifier, spatial coordinates, extension range, and corresponding obstacle information.

[0038] Step 202: Perform physical containment verification based on the minimum passage cross-sectional size of each bottleneck node in the narrow section node list and the maximum external dimensions of each intelligent agent body to obtain a pass status identifier, and perform one-way passage qualification determination based on the pass status identifier to generate a one-way passage intelligent agent directory.

[0039] Optionally, the collaborative control system extracts the extension range of each bottleneck node from the narrow section node list and calculates the minimum passage cross-section size. This minimum passage cross-section size refers to the length and width (or diameter) of the smallest cross-section within the narrow passage bottleneck that allows an object to pass. It is calculated as follows: based on the extension range of the bottleneck node, each cross-section within the bottleneck is intercepted, and the passage width and height of each cross-section are measured. The cross-section with the smallest width and height is selected as the minimum passage cross-section, and its width and height are recorded. Simultaneously, the collaborative control system obtains the maximum external dimensions of each agent, i.e., the maximum width and maximum height corresponding to the outermost contour of the agent. This dimension is an inherent parameter of the agent and is pre-stored in the system. For each bottleneck node, its minimum passage cross-section size is physically checked against the maximum external dimensions of each agent to determine whether the agent can pass through the bottleneck node smoothly. The check rule is: if the minimum passage cross-section width of the bottleneck node is greater than or equal to the maximum width of the agent, and the minimum passage cross-section height is greater than or equal to the maximum height of the agent, then it is determined that it can pass; if either condition is not met, it is determined that it cannot pass. After verification, a pass status flag is generated for each agent corresponding to each bottleneck node, which is divided into two types: "passable" and "unpassable". "Passable" means that the agent size is adapted to the minimum passage cross section of the bottleneck and can pass smoothly; "unpassable" means that the agent size cannot be adapted to the minimum passage cross section of the bottleneck and cannot pass.

[0040] Optionally, the cooperative control system determines one-way passage eligibility based on status identifiers: if the passage status identifier of the bottleneck node corresponding to an agent is "passable," then the agent possesses one-way passage eligibility for that bottleneck node; if it is "not passable," then the agent is not eligible. Information on all agents with one-way passage eligibility is integrated, and the identity identifier of each agent and the bottleneck nodes with passage eligibility are recorded to form a one-way passage agent directory. In the structured list of this directory, each entry includes the agent's identity identifier, the identifier of the bottleneck node with passage eligibility, and the maximum external dimensions of the agent itself.

[0041] Step 203: Based on the identity identifier of each intelligent agent in the one-way passage intelligent agent catalog and the neighboring machine's operation status and remaining energy consumption in the neighboring machine status information, the passage urgency rule is matched to obtain the priority passage right level of each intelligent agent in local interaction, and the space occupancy right is allocated based on the priority passage right level to obtain the channel occupancy right allocation record table.

[0042] Optionally, the collaborative control system extracts the identity identifiers of all intelligent agents from the unidirectional access intelligent agent directory. Simultaneously, it retrieves the neighboring machine status information from the unified spatiotemporal perception information obtained in step 10 to acquire the neighboring machine's operational status (awaiting work, currently working, paused work, fault shutdown) and remaining energy consumption for each intelligent agent. It also retrieves preset access urgency rules, which are used to determine the urgency of intelligent agent access and establish priority access levels. These rules are based on the intelligent agent's own operational status and remaining energy consumption as the core judgment criteria, combined with the operational status and remaining energy consumption of neighboring machines for comprehensive judgment, ensuring the rationality of priority access allocation and adapting to the operational needs of special confined spaces. The specific rules are as follows: ; Based on the rules in the table, the collaborative control system matches the urgency rules of each agent in the one-way passage agent directory and determines the priority passage level according to the operation status and remaining energy consumption. The higher the level, the higher the passage priority, and the more likely the agent is to obtain the right to pass through the narrow passage bottleneck. Then, based on the priority passage level, spatial occupancy rights are allocated, that is, each agent with one-way passage qualification is assigned a passage time and passage position within the narrow passage bottleneck. The allocation principle is: agents with higher priority passage levels are allocated passage time first, with more ample time and more convenient positions (closer to the bottleneck entrance); agents with the same level are allocated according to the remaining energy consumption from high to low, with higher energy consumption resulting in earlier time slots. The identity, priority passage level, allocated passage bottleneck node identity, passage time, and passage position of each agent are recorded to form a channel occupancy right allocation record table. This table can determine the passage rights and arrangements of each agent within the narrow passage bottleneck.

[0043] Step 204: Based on the channel occupancy allocation record table and the task dependencies in the preset global task target sequence, adjust the rate and coordinate the timing to obtain the group decision result.

[0044] Optionally, the collaborative control system adjusts the rate and coordinates the timing based on the channel occupancy allocation record table and the task dependencies in the preset global task target sequence to obtain the group decision result, as shown in steps 2041 to 2045.

[0045] The embodiments of this invention establish passage order in narrow spaces entirely through local interaction and distributed computing, eliminating the need for central node scheduling. This avoids problems such as multiple agents competing for passage, path oscillation, and deadlock in narrow sections from the source, ensuring orderly, efficient, and collision-free passage of multiple agents in special confined spaces with limited width and dense obstacles. It also provides a stable and reliable decision-making basis for subsequent event-triggered communication, conflict decoupling, and trajectory planning.

[0046] Optionally, the process of steps 2041 to 2045 includes: Step 2041: Based on the priority occupying agent identifier in the channel occupancy right allocation record table and the task dependency relationship with the preset global task target sequence, the independent task execution initial sequence of each agent is parsed to obtain the task logical dependency topology graph based on the independent task execution initial sequence.

[0047] Optionally, the collaborative control system calls the channel occupancy right allocation record table obtained in step 203 to extract the priority occupancy agent identifier, which is the unique identity identifier of the agent that has the qualification for space occupancy right allocation (priority passage level is one to four). At the same time, it calls the preset global task target sequence, which contains the content, completion time limit and task dependency relationship of each specific task. The task dependency relationship includes direct dependency and indirect dependency. Direct dependency means that there is a direct order between two tasks, and the next task can only be started after the previous task is completed. Indirect dependency means that two tasks are associated through other tasks, and the intermediate associated task must be completed before the subsequent task can be started.

[0048] The parsing process is based on priority occupant agent identifiers and task dependencies. The parsing process is as follows: First, determine the preset task corresponding to each agent with occupant qualification; second, combine the task dependencies to sort out the order of each agent's own tasks and the dependency relationships between the agent's tasks and other agents' tasks, and eliminate logical conflicts; finally, sort all tasks for each agent according to the order of task dependencies to obtain the initial sequence of independent task execution, which represents the initial order list of each agent's own task execution, and the list clearly marks the task identifier, task content, expected execution time, and the pre-tasks that the task depends on.

[0049] Optionally, the collaborative control system generates a task logical dependency topology graph based on the independent task execution initial sequence of each agent. This topology graph graphically presents the independent task execution initial sequence of all agents and the dependencies between tasks, where nodes represent each specific task, directed line segments represent the dependencies between tasks (from the preceding task node to the subsequent task node), and the agent identifier and estimated execution time for each task are also labeled.

[0050] Step 2042: Based on the task triggering order in the task logic dependency topology graph and the velocity in its own pose information and the neighboring machine position in the neighboring machine state information, the relative motion trend is deduced to obtain the rate interference state identifier between multiple agents, and the rate suppression interval is defined based on the rate interference state identifier to generate an initial velocity adjustment amplitude mapping table.

[0051] Optionally, the collaborative control system extracts the task triggering order from the task logic dependency topology graph, which refers to the order in which tasks are started and completed in the task logic dependency topology graph. Simultaneously, it calls upon its own pose information and neighboring machine status information from the unified spatiotemporal perception information obtained in step 10 to extract the speed (movement rate and direction) and the position (specific spatial coordinates) of each priority occupying agent. Based on the above information, a relative motion trend is extrapolated. The extrapolation process includes: calculating the expected position at different future time points (divided in seconds) based on the speed and current position of each agent; calculating the expected position of neighboring machines at different future time points by combining the current position and speed of the neighboring machines; comparing the expected positions of different agents at the same time point, and considering the task triggering order, determining whether there are situations where agents interfere with or obstruct each other when executing tasks or passing through narrow bottlenecks due to speed mismatch. The extrapolation duration is set to the preset maximum task execution duration. For each pair of agents with a neighboring relationship, a rate interference state identifier is generated, which is divided into interference and no interference. Interference means that the motion trajectories of the two agents overlap or intersect in the future and their rates do not match, resulting in mutual obstruction and interference. No interference means that the motion trajectories of the two agents do not overlap or intersect in the future, or their rates match, and there will be no mutual interference.

[0052] Optionally, the cooperative control system defines a speed suppression interval based on the rate interference state identifier. For agent pairs identified as "interference exists," the range of speed reduction required by the interfering agent is calculated based on their relative motion trends to ensure that the adjusted speed does not overlap or intersect with the opponent's trajectory and does not affect the order of task triggering. For agents identified as "no interference," no speed suppression interval is defined, and their own speed remains unchanged. This generates an initial speed adjustment range mapping table, which records the agent identifier, initial speed, speed suppression interval (empty if there is no interference), and direction of speed adjustment (increase or decrease) for each priority occupying agent.

[0053] Step 2043: Based on the rate suppression interval of each agent in the initial speed adjustment range mapping table and the remaining energy consumption in the neighboring machine state information, perform energy consumption safety boundary verification to obtain the allowable lower limit of speed that meets the minimum energy consumption operation requirements, and perform interval pruning of the initial speed adjustment range based on the allowable lower limit of speed to obtain a speed adjustment range list.

[0054] Optionally, the cooperative control system extracts the rate suppression interval of each agent from the initial speed adjustment range mapping table (for agents without rate interference, the rate suppression interval is considered to be the interval where their initial speed is located), and simultaneously calls the remaining energy consumption of each agent from the neighboring machine status information obtained in step 10. An energy consumption safety boundary check is performed to determine whether the energy consumption is within a safe range after the agent adjusts its speed within the rate suppression interval, ensuring that all preset tasks can be completed. The check process is as follows: 1. Preset minimum energy consumption requirements: Calculated based on the number of tasks for each agent, the estimated execution time, and the energy consumption rate (kWh / min) at different speeds. The energy consumption rate is positively correlated with speed; the faster the speed, the higher the energy consumption. However, too low a speed will lead to longer task execution time, increasing the total energy consumption. 2. Calculate the total energy consumption corresponding to different speeds based on the rate suppression interval: Total energy consumption = Energy consumption rate × Estimated execution time; 3. Compare the total energy consumption at each speed with the current remaining energy consumption, and select the speed range where the total energy consumption is less than or equal to the remaining energy consumption. This range is the feasible speed range. 4. Extract the minimum speed value from the feasible speed range as the lower limit of the allowable speed. If the speed is lower than this, the task execution time will be too long, the total energy consumption will exceed the remaining energy consumption, and the task will not be completed.

[0055] Optionally, the cooperative control system prunes the rate suppression interval based on the allowable speed lower limit, eliminating speed ranges below the allowable speed lower limit and retaining speed ranges above or equal to the allowable speed lower limit that are within the original rate suppression interval. For agents without rate interference, if their initial speed is below the allowable speed lower limit, their speed adjustment interval is set to the range from the allowable speed lower limit to their initial speed (with an increase in speed); if their initial speed is above or equal to the allowable speed lower limit, it remains unchanged. Then, a specific speed adjustment range is determined for each priority occupying agent. The selection principle for the optimal speed is: within the pruned speed adjustment interval, choose a speed that avoids rate interference, minimizes energy consumption, and does not affect the task execution sequence. Based on this, a speed adjustment range list is generated, recording each agent's agent identifier, initial speed, pruned speed adjustment interval, optimal speed, specific speed adjustment range, and adjustment direction.

[0056] Step 2044: Based on the speed adjustment range in the speed adjustment range list and the task triggering order in the task logic dependency topology graph, perform time-series alignment mapping to obtain the task execution time-series correction parameters for each agent, and perform time-series correction on the independent task execution initial sequence of each agent based on the task execution time-series correction parameters to obtain the task execution time-series correction sequence.

[0057] Optionally, the collaborative control system extracts the speed adjustment range and optimal speed of each agent from the speed adjustment range list. Simultaneously, it extracts the task triggering order and estimated execution time of each task from the task logical dependency topology graph. The extracted content is then time-aligned and mapped, i.e., the speed adjustment range of the agent is associated with the task triggering order. The impact of speed adjustment on the execution time of each task is analyzed, and parameters for correcting the task execution timing are calculated to ensure that the task execution timing of all agents matches the task dependencies. The specific mapping process is as follows: The actual execution time of each task after speed adjustment is calculated based on the speed adjustment range and optimal speed, i.e., actual execution time = work area length ÷ optimal speed. If the task does not require movement (e.g., data upload), the actual execution time remains consistent with the estimated execution time. Combining the task triggering order, the actual execution time is compared with the estimated execution time to calculate the execution time deviation (i.e., actual - estimated). Based on the execution time deviation, task execution timing correction parameters are calculated, including task start delay time and task execution interval time, to ensure that the triggering order of each task after correction conforms to the task dependencies and that task execution between agents does not conflict.

[0058] Optionally, the collaborative control system performs timing correction on the initial sequence of independent task execution based on timing correction parameters. Specifically, it retains the task order in the initial sequence, using the task triggering order in the task logical dependency topology as a benchmark. The system adjusts the start and finish times of each task according to the timing correction parameters; for tasks with extended execution times, the start time of subsequent dependent tasks is appropriately delayed; for tasks with shortened execution times, the start interval of subsequent dependent tasks is appropriately shortened. This ultimately forms the timing-corrected sequence of task execution for each agent. In other words, this timing-corrected sequence is an agent task execution sequence that has undergone timing adjustment and meets the requirements of task dependencies and speed adjustment. The list clearly marks the task identifier, task content, actual execution time, start time, finish time, dependent prerequisite tasks, and corresponding agent identifier for each task.

[0059] Step 2045: Based on the task execution timing correction sequence and the spatial occupancy allocation information in the channel occupancy allocation record table, conflict elimination is performed to obtain a conflict-free group decision result that includes the independent task execution order and speed adjustment range of each agent.

[0060] Optionally, the collaborative control system extracts the task execution timing correction sequence of each agent and simultaneously calls the spatial occupancy allocation information (including bottleneck node identifier, passage time period, and passage location) from the channel occupancy allocation record table obtained in step 203 to determine whether there is a spatial occupancy conflict among the agents during task execution. This involves associating the task execution timing correction sequence of each agent with the corresponding spatial occupancy allocation information to clarify whether the time each agent spends passing through the narrow passage bottleneck matches the passage time period allocated by the spatial occupancy allocation; checking for situations where multiple agents pass through the narrow passage bottleneck at the same time and location, and if so, determining it as a spatial occupancy conflict; and adjusting the task execution timing correction sequence of the conflicting agents (mainly adjusting the start time of bottleneck-related tasks) to ensure that the passage times of the conflicting agents are staggered without affecting task dependencies and overall operational efficiency. By integrating the task execution timing correction sequence and speed adjustment range list of all agents, a conflict-free group decision result is formed. This result includes the identity of each priority occupying agent, the independent task execution order, the speed adjustment range (optimal speed, adjustment direction), and the space occupation arrangement (passage time period, passage position). At the same time, the task execution requirements of agents without space occupation rights are clarified (pause work or wait for instructions).

[0061] The embodiments of the present invention do not rely on the global scheduling of the central node. They can generate conflict-free and highly adaptable group decision-making results through local information interaction and hierarchical optimization. This not only adapts to the environmental characteristics of special limited spaces and narrow communication, but also ensures the timing rationality and energy consumption safety of multi-agent collaborative operations.

[0062] Optionally, the processes of steps 401 to 404 include: Step 401: Based on the execution order of each agent's independent tasks in the group decision-making results and the position coordinates of the neighboring machines in the neighboring machine state information group, the static spatial occupancy area is extracted to obtain the static operation interference area. Based on the static operation interference area and the obstacle movement state in the local perception information, dynamic trajectory extrapolation is performed to divide the conflict risk level map.

[0063] Optionally, the collaborative control system extracts the independent task execution order of each agent based on the group decision-making results, that is, the order in which each agent needs to perform various specific tasks in the current work scenario. Simultaneously, it extracts the neighboring machine position coordinates from the neighboring machine status information group, that is, the specific spatial coordinates of the neighboring machines in a unified three-dimensional spatial coordinate system within the special confined space. Based on this extracted information, static spatial occupancy areas are extracted. Specifically, according to the independent task execution order of each agent, the specific work position for each agent to perform each task is determined; combined with the agent's body dimensions, a static spatial occupancy area is formed by expanding outward from the work position, with the expansion range consistent with the agent's body dimensions; the static spatial occupancy areas of all agents are superimposed and compared, and areas with overlap are selected, which are the static work interference areas. If multiple agents work simultaneously within this area, spatial occupancy conflicts will occur, leading to collision risks. Specific information includes the area's spatial coordinate range, overlapping agent identifiers, and corresponding work task identifiers.

[0064] Optionally, the collaborative control system extracts obstacle motion states based on local perception information from unified spatiotemporal perception information. This includes the motion of all obstacles within the agent's perception range, including whether the obstacle is in motion, and if so, its direction and speed. Stationary obstacles are marked as stationary, with no direction or speed specified. Dynamic trajectory extrapolation is performed based on the static operation interference area and obstacle motion states. Combining the spatial range of the static operation interference area and the obstacle motion states, the system predicts the motion trajectories of each agent and obstacles within the next 5 seconds (covering the prediction range of high, medium, and low priority collision risks). The system then categorizes these trajectories hierarchically based on trajectory intersections and the severity of collision risks, presenting the results in a graphical representation.

[0065] The specific generation process includes: Based on the independent task execution order and current speed of each agent, the movement trajectory of each agent within the next 5 seconds is predicted, and the expected position at each time point is determined; based on the movement state of obstacles, the movement trajectory of moving obstacles within the next 5 seconds is predicted, while the trajectory of stationary obstacles remains fixed; the movement trajectories of each agent are compared with the movement trajectories of other agents and obstacles to determine whether there are trajectory intersections or position overlaps, and the collision risk level (low-level conflict, medium-level conflict, high-level conflict) is determined by combining the intersection time and distance; all collision risk levels, corresponding agent identifiers, obstacle information, trajectory intersection positions, and risk occurrence times are integrated and presented in a graphical form to generate a conflict risk level map. In this map, different colors distinguish conflict levels, lines represent the movement trajectories of agents and obstacles, nodes represent trajectory intersection positions, and each risk is labeled with the corresponding agent, obstacle, risk level, and expected occurrence time.

[0066] Step 402: Based on the boundary coordinates of high-risk areas in the conflict risk hierarchy map and the current attitude angle and body shape of each agent, perform attitude envelope geometric expansion to obtain the dynamic safety envelope boundary of each agent, and delineate geometric restricted areas based on the dynamic safety envelope boundary to generate a geometric restricted area constraint list.

[0067] Optionally, the collaborative control system extracts the boundary coordinates of high-risk areas from the conflict risk hierarchy map. These are the specific coordinates of the spatial boundaries of high-level conflict areas in the conflict risk hierarchy map within a unified three-dimensional spatial coordinate system. These areas represent the most severe collision risks and require priority for safety control. Simultaneously, the system acquires the current attitude angle and body dimensions of each agent. The current attitude angle is the current orientation angle of each agent (based on the horizontal direction, ranging from 0 to 360 degrees), extracted from its own pose information. The body dimensions are the specific length, width, and height dimensions corresponding to the outermost contour of the agent, pre-stored in the collaborative control system. Centered on the agent's current position and combined with the current attitude angle, the agent's body dimensions are reasonably expanded in all directions (e.g., the expansion distance is set to 0.5 times the agent's maximum body dimensions) to form a dynamic boundary that completely covers the space the agent may occupy during movement. This is the dynamic safety envelope boundary of each agent. During the expansion process, the expansion range is adjusted based on the boundary coordinates of high-risk areas to ensure that the expanded boundary does not exceed the control range of the high-risk areas.

[0068] Optionally, the cooperative control system delineates geometrically restricted areas based on the dynamic safety envelope boundary. The delineation process is as follows: all high-risk areas in the conflict risk hierarchy map are designated as geometrically restricted areas; areas outside the dynamic safety envelope boundary of each agent that may collide with other agents or obstacles are designated as geometrically restricted areas; combined with the obstacle position in the local perception information, a certain range around the obstacle (e.g., the range is 0.5 times the maximum size of the obstacle) is designated as a geometrically restricted area. After the geometrically restricted areas are delineated, the spatial range (boundary coordinates), the reason for restriction (high-risk area, area around the obstacle, risk area outside the agent's safety envelope boundary), and the restriction validity period (from the time of delineation to the time of risk elimination) of each geometrically restricted area are recorded to form a geometrically restricted area constraint list.

[0069] Step 403: Based on the spatial range of each restricted area in the geometric restricted area constraint list and the speed adjustment range and current speed of each agent in the group decision results, the uniform motion time occupation is mapped to obtain the expected time occupation interval of each agent. Based on the expected time occupation interval, the motion interference factors in the time dimension are removed to generate the temporal isolation constraint window sequence.

[0070] Optionally, the cooperative control system extracts the spatial range of each restricted area based on the geometric restricted area constraint list, that is, the three-dimensional spatial range defined by the boundary coordinates of each geometric restricted area. At the same time, it retrieves the group decision results and extracts the speed adjustment range and current speed of each agent. The speed adjustment range is the specific amount that each agent needs to adjust based on its current speed to adapt to the group's cooperative operation, including increasing speed, decreasing speed, or keeping speed unchanged; the current speed is the actual movement rate of the agent after speed adjustment, which is calculated from the agent's initial speed (extracted from its own pose information in step 10) and speed adjustment range. Based on the extracted data, a uniform motion time occupancy mapping is performed. The specific mapping process is as follows: According to the spatial range of the geometrically restricted area, the passable area for each agent (the area outside the geometrically restricted area within the special confined space) is determined, and the passable path and path length of each agent from its current position to the target work position are clarified; According to the current speed of each agent, the total time for each agent to pass through the passable path is calculated by time = path length ÷ current speed, and the time nodes for the agent to pass through key nodes on the passable path (such as near the boundary of the restricted area, path turning points) are also divided; Combining the key time nodes, the expected time occupancy interval of each agent within the passable area is determined, and presented in the form of start time - end time.

[0071] Optionally, the collaborative control system removes motion interference factors in the time dimension based on the expected time occupancy interval. These motion interference factors refer to factors where two or more agents occupy the same passable area simultaneously, leading to temporal motion conflicts and thus collision risks. The specific removal process is as follows: The expected time occupancy intervals of all agents are compared, and intervals with temporal overlap are selected. The severity of the temporal overlap is analyzed: if the overlap time exceeds 100 milliseconds, it is determined that there is temporal motion interference; if the overlap time is less than or equal to 100 milliseconds, it is considered that there is no temporal motion interference (which can be resolved by slight speed adjustments). For agents with temporal motion interference, their expected time occupancy intervals are adjusted by delaying the start time of one or more agents, fine-tuning their speed, etc., to ensure that the expected time occupancy intervals of each agent do not overlap. After the temporal motion interference factors are removed, the expected time occupancy interval (after adjustment without overlap), passable path, current speed, and key time nodes of each agent are recorded, forming a temporal isolation constraint window sequence.

[0072] Step 404: Based on the temporal isolation constraint window sequence, the remaining energy consumption of neighboring machines, the list of geometrically restricted areas, and the operating status of neighboring machines, the constraint solution and output are performed to obtain conflict-free cooperative motion constraints.

[0073] Optionally, the cooperative control system solves and outputs constraints based on the temporal isolation constraint window sequence, the remaining energy consumption of neighboring machines, the list of geometrically restricted areas, and the operating status of neighboring machines, to obtain conflict-free cooperative motion constraints, as described in steps 4041 to 4045.

[0074] This invention simultaneously resolves collision risks among multiple agents in special confined spaces from two dimensions: spatial geometry and temporal sequence. It accurately distinguishes risk levels and implements hierarchical constraints to avoid path oscillations and multi-machine deadlock, thereby improving the real-time performance and reliability of conflict resolution.

[0075] Optionally, the processes of steps 4041 to 4045 include: Step 4041: Based on the time occupancy intervals in the time-series isolation constraint window sequence and the remaining energy consumption of neighboring machines, a low-power passage grace rule is matched to obtain the time window margin expansion result. Based on the time window margin expansion result, a passage order constraint rule table that takes into account the energy consumption stability of the mandatory passage order rule is constructed.

[0076] Optionally, the collaborative control system extracts the time occupancy intervals from the time-series isolation constraint window sequence, i.e., the time range from entry to exit of each agent within the passable area. Simultaneously, it extracts the remaining energy consumption of neighboring agents from the neighboring agent status information group, i.e., the total amount of energy currently available for operation for each agent. Based on the above information, a low-power passage grace rule is matched. This low-power passage grace rule is preset in the decision optimization system and grants agents with different remaining energy consumption levels a certain margin for expanding their time occupancy intervals. Specifically, the rule is divided into three levels: remaining energy consumption greater than or equal to 5 kWh is the high-power level, with no time window margin expansion; remaining energy consumption greater than or equal to 3 kWh but less than 5 kWh is the medium-power level, with a time window margin expansion of 1000 milliseconds; and remaining energy consumption less than 3 kWh is the low-power level, with a time window margin expansion of 2000 milliseconds. The margin expansion direction is to advance the start time and delay the end time, ensuring that low-power agents have sufficient time to pass at a constant speed, reducing the energy consumption rate.

[0077] The rule matching process is as follows: the remaining energy consumption of each agent's neighboring machines is compared with the three levels to determine the energy consumption level corresponding to each agent; according to the energy consumption level, a corresponding time window margin is extended, and the extended time occupancy interval is calculated: extended start time = original start time - margin extension duration / 2, extended end time = original end time + margin extension duration / 2; the original time occupancy interval, remaining energy consumption, energy consumption level, margin extension duration, and extended time occupancy interval of each agent are recorded to obtain the time window margin expansion result.

[0078] Optionally, the cooperative control system constructs a mandatory passage order rule that takes into account energy consumption stability based on the time window margin expansion result. This mandatory passage order rule refers to determining the passage order according to the remaining energy consumption of each agent and the overlap of the expanded time occupancy interval, ensuring that agents with low power have priority passage and avoiding new time movement interference after the time window expansion. Specifically, the construction process is as follows: sort all agents by their remaining energy consumption: the lower the remaining energy consumption, the higher the passage priority (low power > medium power > high power); agents of the same energy consumption level are sorted by the start time of the expanded time occupancy interval, with the earlier the start time, the higher the priority; the mandatory passage order of each agent is determined according to the sorting result, and the passage priority, expected start time, and expected end time are marked; information such as agent identification, remaining energy consumption, energy consumption level, expanded time occupancy interval, mandatory passage order, and passage priority are integrated to form a passage order constraint rule table. This rule table contains the mandatory passage order and related constraint information for all agents, ensuring that agents with low power have priority passage, energy consumption is stable, and time-based motion interference is avoided.

[0079] Step 4042: Based on the forced passage order in the passage order constraint rule table, the kinematic steering capability boundary is compared with the maximum allowable turning angular velocity and current velocity of each agent to obtain a set of feasible heading angles. Invalid heading instructions are then eliminated based on the set of feasible heading angles to obtain a list of feasible heading constraint domains.

[0080] Optionally, the cooperative control system extracts the mandatory passage order from the passage order constraint rule table, i.e., the order in which each agent passes, and obtains the maximum permissible turning angular velocity and current speed of each agent. The maximum permissible turning angular velocity refers to the maximum rate of change of turning angle that an agent can withstand during movement; the current speed refers to the actual movement rate of the agent after speed adjustment. Based on the above information, a kinematic turning capability boundary comparison is performed. This comparison process simultaneously avoids heading angles pointing towards low / medium risk intersection areas in the conflict risk hierarchy map, indirectly mitigating collision risks caused by heading deviations at low / medium risk levels.

[0081] The comparison process includes: determining the movement time and path range of each agent according to the mandatory passage order, identifying key nodes that need to turn (path turning points, turning points to avoid restricted areas); calculating the maximum and minimum turning angles that can be achieved at each key node based on the current speed and the maximum permissible turning angular velocity: maximum turning angle = maximum permissible turning angular velocity × turning time, minimum turning angle = 0 degrees; the turning time is determined based on the current speed and the distance between key nodes; integrating the turning angle ranges of all key nodes for each agent, removing overlapping and contradictory angles, and eliminating angles pointing to low / medium risk intersection areas to obtain a set of feasible heading angles; performing the calculations sequentially for all agents to obtain the set of feasible heading angles corresponding to each agent.

[0082] Optionally, the cooperative control system eliminates invalid heading commands based on the set of feasible heading angles. Invalid heading commands refer to those that exceed the range of the feasible heading angle set, are impossible to implement, or whose implementation would affect the mandatory passage order and lead to collision risks. The remaining heading commands are retained and integrated to form a list of feasible heading constraint domains for each agent. This list contains all valid heading commands and related constraint information for each agent, labeled with the agent identifier, set of feasible heading angles, valid heading commands, corresponding key nodes, and turning time.

[0083] Step 4043: Based on the feasible heading angles in the feasible domain list of heading constraints and the restricted areas in the restricted area list of geometric restricted areas, determine the non-intersection of spatial trajectories to obtain safe passage trajectory segments. Then, based on the safe passage trajectory segments, eliminate redundant motion paths with trajectory intersection risks to obtain a set of non-intersecting safe trajectory segments.

[0084] Optionally, the cooperative control system extracts feasible heading angles from the feasible domain list based on heading constraints, and extracts the spatial range of each restricted area from the geometric restricted area constraint list. It then generates motion trajectory segments based on the feasible heading angles, compares each trajectory segment with the spatial ranges of all geometric restricted areas, and filters out trajectory segments with no overlap, thus obtaining safe passage trajectory segments. This determination process also simultaneously avoids overlapping areas corresponding to low / medium risk levels.

[0085] Optionally, the cooperative control system eliminates redundant motion paths with trajectory intersection risks based on safe passage trajectory segments. These redundant paths refer to paths among multiple safe passage trajectory segments that intersect and may lead to collisions, or paths that are too long or inefficient. The specific elimination process is as follows: The safe passage trajectory segments of all agents are superimposed and compared to filter out combinations with trajectory intersection risks. For combinations with intersection risks, combined with the mandatory passage order, the trajectory of the priority agent is retained, while the trajectory of the agent with the next priority is eliminated or adjusted. Simultaneously, redundant trajectories with a path length exceeding 1.5 times the shortest path are eliminated. Finally, all remaining safe passage trajectory segments are integrated to form a set of non-intersecting safe trajectory segments. This set specifically includes all agent trajectory segments with no trajectory intersections, no intersections with prohibited areas, and high efficiency and safety, labeled with agent identification, feasible heading angle, trajectory coordinates, trajectory length, and estimated passage time.

[0086] Step 4044: Based on the time occupancy intervals in the temporal isolation constraint window sequence and the safe passage trajectory segments in the non-intersecting safe trajectory segment set, perform spatiotemporal coordinate synchronization and alignment mapping to obtain the allowed motion state set of each agent. Based on the allowed motion state set, eliminate the state coupling conflict when multiple agents move together to obtain the decoupled single agent motion constraint set.

[0087] Optionally, the cooperative control system extracts each safe passage trajectory segment from the set of non-intersecting safe trajectory segments, and simultaneously retrieves the temporal isolation constraint window sequence obtained in step 403 to extract the time occupancy interval of each agent. Based on the extracted information, a spatiotemporal coordinate synchronization and alignment mapping is performed to associate the safe passage trajectory segments (spatial dimension) with the time occupancy interval (temporal dimension), achieving a one-to-one correspondence between spatial coordinates and time nodes. The specific mapping process is as follows: based on the trajectory length and the duration of the expanded time occupancy interval, the agent's uniform motion speed (consistent with the current speed) is calculated; the expanded time occupancy interval is divided into time nodes at the millisecond level, and the expected spatial coordinates of each time node on the trajectory are calculated based on the uniform motion speed; the time nodes, corresponding spatial coordinates, motion speed, and heading angle are integrated to form a set of permissible motion states.

[0088] Optionally, the cooperative control system eliminates state coupling conflicts during multi-agent joint movement based on the set of permissible motion states. This state coupling conflict refers to the spatiotemporal overlap and mutual interference of the permissible motion states of different agents during joint movement. The specific elimination process is as follows: The permissible motion state sets of all agents are compared, and combinations with spatiotemporal overlap are selected; the permissible motion state sets of the passing agents are adjusted by delaying time nodes or fine-tuning spatial coordinates, in conjunction with a mandatory passage order and a non-intersecting safety trajectory; all conflicting combinations are adjusted sequentially, and the adjusted permissible motion state sets of each agent are integrated to obtain the decoupled single-agent motion constraint set. This single-agent motion constraint set is a set of motion states for each agent that are independent, without coupling conflicts, and conform to spatiotemporal constraints and safety requirements.

[0089] Step 4045: Based on the permissible motion state parameters in the single agent motion constraint set and the adjacent machine operation status, integrate the high-risk operation safety distance expansion rules to obtain a unified format motion restriction instruction set. Then, based on the motion restriction instruction set, combine the spatial no-entry boundary, time passage window, heading angle limit and safety distance parameters to encapsulate and obtain conflict-free cooperative motion constraints.

[0090] Optionally, the cooperative control system extracts the permissible motion state parameters (such as time nodes, spatial coordinates, motion speed, and heading angle) from the single agent motion constraint set, and simultaneously calls the neighboring machine status information group obtained in step 30 to extract the neighboring machine's operational status (such as currently operating, suspended, or out of service due to malfunction). Based on this information, a high-risk operation safety distance expansion rule is integrated. This high-risk operation safety distance expansion rule refers to increasing the safety distance for agents currently operating to prevent other agents from approaching them. For example, if a neighboring machine is operating: the safety distance threshold is expanded from 0.8 meters to 1.2 meters; if a neighboring machine is suspended or out of service due to malfunction: the safety distance remains unchanged at 0.8 meters. The integration process is as follows: the permissible motion state parameters of each agent are compared with the neighboring machine's operational status to determine whether the neighboring machine is currently operating; the safety distance threshold is adjusted according to the rule, and if the neighboring machine is operating, the spatial coordinate constraints are expanded and updated; the adjusted parameters of all agents, the safety distance threshold, and the neighboring machine's operational status are integrated and organized in a unified format to obtain a motion restriction instruction set. This motion restriction instruction set contains motion constraint instructions in a unified format for all agents. Each entry corresponds to one agent and explicitly specifies the allowed motion state parameters, safe distance thresholds, neighboring machine operation status, and other related constraints.

[0091] Optionally, the cooperative control system encapsulates motion restriction instruction sets based on spatial no-entry boundaries, time passage windows, heading angle limits, and safety distance parameters. First, the motion restriction instruction set is integrated with the four types of parameters to clarify the spatial motion boundaries, time passage ranges, heading restrictions, and safety distance requirements for each agent. Redundant information is removed to ensure that all constraint parameters are unified, standardized, and conflict-free. Finally, the data is organized according to a preset encapsulation format to form a unified file containing conflict-free motion constraints for all agents, resulting in conflict-free cooperative motion constraints.

[0092] This invention further refines the constraint-solving process, overcoming the limitations of single-dimensional constraints. It achieves multi-dimensional collaborative management of energy consumption, steering, space, time, and operational status. It focuses on managing high-risk levels while indirectly mitigating collision risks at low / medium risk levels through core actions in each step. There is no need to set additional control rules for low / medium risk levels. This ensures that the generated collaborative motion constraints are both adapted to the performance of the intelligent agent and meet the complex environmental requirements of special confined spaces. It effectively avoids path oscillations and multi-machine deadlock, further improving the accuracy and reliability of conflict resolution.

[0093] Optionally, the processes of steps 501 to 505 include: Step 501: Based on the coordinates of the spatial restricted boundary in the cooperative motion constraint and the local obstacle position in the unified spatiotemporal perception information, perform geometric exclusion region union processing to obtain a composite static restricted region set.

[0094] Optionally, the cooperative control system extracts the coordinates of the spatial restricted boundaries, i.e., the three-dimensional coordinates of the boundary of the geometrically restricted area, based on cooperative motion constraints. Simultaneously, it extracts the local obstacle locations from the unified spatiotemporal perception information, i.e., the three-dimensional coordinates of temporary or fixed obstacles within the special confined space. Temporary obstacles include work tools, temporary construction areas, etc.; fixed obstacles include pipe gallery supports, tunnel walls, and cabin equipment, etc. Based on the above information, a geometric exclusion region union process is performed: the basic restricted area defined by the spatial restricted boundary coordinates and the obstacle areas corresponding to the local obstacle locations are treated as two independent geometric exclusion regions. Through spatial geometric operations, all inaccessible spatial ranges are merged, and overlapping areas are removed to form a complete and unified set of static restricted areas, thus obtaining the composite static restricted area set.

[0095] Step 502: Based on the safety distance parameter in the composite static restricted area set combined with the cooperative motion constraint, the boundary is expanded outward to generate a geometric passage corridor with safety redundancy for each intelligent agent.

[0096] Optionally, the cooperative control system, based on the composite static restricted area set and the safety distance parameters in the cooperative motion constraints obtained in step 4045, uses the boundary of each restricted area in the composite static restricted area set as a reference to uniformly expand the distance corresponding to the safety distance parameters outward in the passable direction of the restricted area, thereby expanding the actual avoidance range of the restricted area. Then, according to the passage path range of each agent in the passage order constraint rule table, combined with the expanded restricted area boundary, a dedicated passable space is allocated for each agent. The outer boundary is defined according to the boundary of the passage corridor of adjacent agents or the inner wall boundary of the limited space. Finally, the dedicated passable space is regularized into a long strip-shaped passable area with fixed boundaries and sufficient safety redundancy, i.e., a geometric passage corridor dedicated to each agent. This geometric passage corridor is a dedicated passable area separately defined for each agent, in the shape of a long strip, and marked with three-dimensional boundary coordinates, corresponding agent identifier, and safety redundancy distance.

[0097] Step 503: Extract the curvature radius of each turning point based on the central axis of the geometric passageway, and compare it with the kinematic compatibility of the heading angle limit in the cooperative motion constraint to obtain a set of feasible path nodes that meet the maximum turning capability limit.

[0098] Optionally, the collaborative control system determines the central axis based on the unique geometric corridor of each agent. This central axis is a continuous three-dimensional curve located in the middle of the corridor along its length. The determination process is as follows: Multiple sampling points are divided along the length of the geometric corridor at 1-meter intervals. Each sampling point is the center point of the corresponding cross-section of the corridor. All sampling points are connected sequentially to form a continuous three-dimensional curve, which is the central axis of the geometric corridor. Next, inflection points are extracted, and it is determined whether the angle between the line connecting two adjacent sampling points and the line connecting the current point and the next sampling point is greater than 10 degrees. If it is greater, the current sampling point is determined as the inflection point. Finally, with the inflection point as the center, two sampling points before and after it (a total of five consecutive points) are selected, and a circular arc curve is fitted to calculate the radius, which is the radius of curvature of the inflection point.

[0099] The collaborative control system extracts the heading angle limit obtained in step 4042, and the range of feasible heading angles for each agent, i.e., the maximum turning angle range. It calculates the turning angle (inversely proportional to the radius of curvature) based on the radius of curvature of each turning point and the agent's current speed. The turning angle is compared with the heading angle limit: points within the range are feasible turning points, and those outside the range are infeasible turning points. All feasible turning points and non-turning point sampling points are integrated to form a feasible path node set. This feasible path node set includes all feasible turning points and non-turning point sampling points on the central axis of the geometric corridor, and is labeled with three-dimensional coordinates, node type, radius of curvature, and turning angle.

[0100] Step 504: Based on the feasible path node set, sharp corner path segments with curvature radii smaller than the minimum allowable value are eliminated to obtain a geometrically smooth and compliant path skeleton.

[0101] Optionally, the cooperative control system retrieves a preset minimum permissible radius of curvature, i.e., the minimum radius of curvature corresponding to a corner that the agent can smoothly pass through. This radius can be determined based on the agent's inherent turning performance and preset in the cooperative control system, such as being set to 3 meters, meaning the minimum permissible value is 3 meters. Then, based on the feasible path node set, sharp corner path segments with radii of curvature smaller than the minimum permissible value are selected. These are path segments between two adjacent feasible turning points in the feasible path node set, with at least one turning point having a radius of curvature smaller than the minimum permissible radius of curvature. The sharp corner path segments are removed from the node set, and the corresponding nodes are deleted. The remaining nodes are connected in their original order to form a geometrically smooth and compliant path skeleton, meaning that the radius of curvature of all turning points in the geometrically smooth and compliant path skeleton is greater than or equal to the minimum permissible radius of curvature.

[0102] Step 505: Based on the geometrically smooth compliant path skeleton and the task order in the group decision results, perform task spatiotemporal mapping to obtain the collision-free collaborative operation trajectory of each agent.

[0103] Optionally, the collaborative control system performs a spatiotemporal mapping of tasks based on the geometrically smooth compliant path skeleton and the task order in the group decision results to obtain the collision-free collaborative operation trajectory of each agent, as specifically in steps 5051 to 5055.

[0104] This invention optimizes trajectory planning from four dimensions: path smoothness, spatial specialization, turning adaptability, and task timing. This enables collision-free, smooth, and orderly collaborative operation of agents within a special confined space, eliminating risks such as sudden path changes, sharp turns, and spatial collisions. Optionally, the processes of steps 5051 to 5055 include: Step 5051: Based on the spatial coordinates of the path nodes in the geometrically smooth compliant path skeleton and the execution order of the independent tasks of each agent in the group decision results, spatial target index binding is performed to obtain the target waypoint sequence of each agent arranged in the order of task.

[0105] Optionally, the collaborative control system retrieves the geometrically smoothed compliant path skeleton and the spatial coordinates of its path nodes, i.e., the three-dimensional coordinates of each node on the path. Simultaneously, it extracts the individual task execution order of each agent from the group decision-making results, i.e., the order in which tasks are assigned to each agent, with each task corresponding to a spatial target location. The spatial target location corresponding to each task in the individual task execution order is matched and associated with the spatial coordinates of the path nodes in the geometrically smoothed compliant path skeleton, assigning a corresponding path node as a target waypoint for each task. These waypoints are then arranged in task order to form a target waypoint sequence. This target waypoint sequence is labeled with the three-dimensional coordinates of each waypoint, the corresponding task identifier, the task name, and the task execution order.

[0106] Step 5052: Based on the target waypoint sequence, the continuous path skeleton is truncated into local navigation segments corresponding to each sub-task, and a local navigation segment mapping table is obtained.

[0107] Optionally, the cooperative control system truncates the continuous path skeleton into local navigation segments corresponding to each subtask based on the target waypoint sequence. The local navigation segment refers to the path segment between two adjacent target waypoints in the geometrically smooth and compliant path skeleton, and each local navigation segment corresponds to a subtask. The specific truncation is as follows: For each agent, traverse its target waypoint sequence and determine the path node positions of two adjacent target waypoints; from the geometrically smooth and compliant path skeleton, extract all path nodes between two adjacent target waypoints, and determine the path segment formed by connecting these path nodes as the local navigation segment of the subtask between the two target waypoints; the path segment before the first target waypoint is used as the initial navigation segment (corresponding to the preceding path of the first task), and the path segment after the last target waypoint is used as the ending navigation segment (corresponding to the following path of the last task); assign a unique segment identifier to each local navigation segment, and label the subtask identifier, task name, starting target waypoint coordinates, ending target waypoint coordinates, and coordinates of all path nodes contained in the segment; organize all local navigation segments of each agent according to the task sequence to form a structured table containing detailed information of each local navigation segment, which is the local navigation segment mapping table.

[0108] Step 5053: Calculate the transit time at nominal speed based on the spatial length of each road segment in the local navigation road segment mapping table and the time travel window in the cooperative motion constraint, to obtain the target arrival time node for each road segment.

[0109] Optionally, the cooperative control system extracts the spatial length of each road segment from the local navigation road segment mapping table, i.e., the actual spatial distance of each local navigation road segment, which is calculated as the straight-line distance between the starting and ending target waypoints of the road segment. Simultaneously, it extracts the time travel window from the cooperative motion constraints, i.e., the expanded time occupancy interval of each agent in the time window margin expansion result obtained in step 4041. It also obtains a preset nominal speed, i.e., the standard speed of the agent when traveling at a constant speed under normal operating conditions, which can be preset according to the agent's performance parameters and operational requirements, such as setting it to 0.3 m / s.

[0110] Based on the spatial length and nominal speed of each road segment, the time required for the agent to traverse that road segment is calculated. Then, combined with the time passage window, the start time within the time passage window is extracted as the agent's initial passage time. The passage time of each road segment is accumulated sequentially according to the order of the local navigation road segments to determine the start passage time and target arrival time node of each road segment. It is checked whether the target arrival time node of each road segment is within the time passage window range. If it is outside the range, the nominal speed is finely adjusted (not exceeding the agent's maximum speed), and the passage time is recalculated until the target arrival time node is within the time passage window range. Finally, the target arrival time node of each road segment is determined.

[0111] Step 5054: Based on the target arrival time node, perform a timestamp marking operation on the local navigation road segment to obtain a set of spatiotemporally synchronized and coordinated trajectory nodes.

[0112] Optionally, the collaborative control system retrieves the target arrival time nodes for each road segment and extracts the spatial coordinates of all path nodes contained in each local navigation road segment from the local navigation road segment mapping table. Based on the above information, a timestamp marking operation is performed on the local navigation road segments: the target arrival time node of each road segment and the corresponding time of each path node within the road segment are marked onto each path node, so that each path node contains spatial coordinates and a timestamp. All timestamped path nodes of all agents are integrated and arranged according to agent identification and path sequence to form a spatiotemporally synchronized collaborative trajectory node set.

[0113] Step 5055: Based on the spatial coordinates, timestamps, and heading angles of each node in the spatiotemporal synchronized collaborative trajectory node set, and the motion restriction instruction set in the collaborative motion constraints, data structure fields are aligned and encapsulated to obtain the collision-free collaborative operation trajectory.

[0114] Optionally, the cooperative control system extracts the spatial coordinates, timestamps, and heading angles of each node based on the spatiotemporally synchronized cooperative trajectory node set. Simultaneously, it extracts the motion constraint instruction set from the cooperative motion constraints, i.e., a set containing motion constraint instructions in a unified format for all agents. Each instruction item corresponds to one agent, specifying its allowed motion state parameters, safe distance thresholds, and constraints related to adjacent aircraft operation states. This information is then encapsulated by aligning data structure fields. Specifically, the spatial coordinates, timestamps, and heading angles of each node in the spatiotemporally synchronized cooperative trajectory node set are organized and aligned according to the data structure field format of the motion constraint instruction set, supplementing relevant constraint information to form a unified format collision-free cooperative operation trajectory.

[0115] The embodiments of the present invention further refine the implementation process of task spatiotemporal mapping, making up for the defects of traditional trajectory planning such as weak association between tasks and paths, spatiotemporal asynchrony, and inconsistent data formats. It ensures that the generated collision-free collaborative operation trajectory not only conforms to the turning performance and operation task requirements of the intelligent agent, but also meets the cooperative motion constraints, realizing spatiotemporal coordination and safe collision-free operation trajectory of each intelligent agent.

[0116] Optionally, after performing the collaborative work motion, it also includes: Step 506: Based on the collision-free collaborative operation trajectory, monitor the communication connection status of each intelligent agent in the special confined space in real time.

[0117] Optionally, the collaborative control system uses the collision-free collaborative operation trajectory as the operating benchmark. During the collaborative operation of each intelligent agent, it monitors the communication connection status in real time. That is, at fixed time intervals, it continuously collects the signal strength, data transmission and reception success rate, and communication delay values ​​between each intelligent agent and its neighboring machine to determine whether each intelligent agent is in a normal connection, signal attenuation, temporary packet loss, or complete disconnection state, and forms a real-time communication connection status result.

[0118] Step 507: When the communication connection status is abnormal, a neighborhood connection graph is generated based on the unified spatiotemporal awareness information and the signal quality of neighboring machines, and the neighborhood connection graph is dynamically reconstructed to obtain an updated neighborhood connection graph.

[0119] Optionally, when the cooperative control system determines that an anomaly has occurred in the communication connection status, it extracts and records the information of the anomalous agent, such as its identifier, anomaly type, and anomaly occurrence timestamp. Simultaneously, it extracts unified spatiotemporal perception information, from which it extracts the current actual location of each agent, neighbor identifiers, and neighbor status. A neighborhood connection graph is generated using online agents as nodes and the communication connections between agents as edges. The edge weights are determined by the neighbor signal quality, ranging from 0 to 1. For example, a weight greater than 0.7 indicates good communication, a weight between 0.3 and 0.7 indicates average communication, and a weight less than 0.3 indicates poor communication. The cooperative control system then dynamically reconstructs the neighborhood connection graph, removing disconnected agent nodes, retaining valid communication connection edges, and re-establishing the shortest communication links between online agents, resulting in an updated neighborhood connection graph with a stable topology and reliable communication.

[0120] Step 508: Based on the updated neighborhood connection graph and the original independent task execution order in the group decision results, the task nodes corresponding to the lost agents are redistributed to the online agents according to the nearest spatial distance rule, so as to obtain the updated independent task execution order of each agent.

[0121] Optionally, the collaborative control system extracts the identifiers, current locations, and adjacent node information of all online agents from the updated neighborhood connection graph. Simultaneously, it extracts the original independent task execution order from the group decision-making results, filtering out the task nodes corresponding to the lost agents. These task nodes represent the spatial target location, task identifier, task name, and execution requirements for each task in the original independent task execution order. The task nodes corresponding to the lost agents are then reassigned to online agents according to the nearest spatial distance rule: that is, the straight-line distance between the spatial target location of each lost agent's task node and the current locations of all online agents is calculated, and the task node is assigned to the online agent with the smallest straight-line distance. After the assignment is completed, the updated independent task execution order for each agent is obtained.

[0122] Step 509: Based on the adjacency relationships in the updated neighborhood connection graph, the safety distance parameter in the cooperative motion constraint is corrected by communication compensation to obtain the cooperative motion constraint after communication limitation correction.

[0123] Optionally, the cooperative control system extracts the adjacency relationships between online agents (including the weight of each adjacency edge and the identifier of the adjacent node) from the updated neighborhood connection graph. Simultaneously, it extracts the safety distance parameter from the cooperative motion constraints. Based on the adjacency relationships in the updated neighborhood connection graph, it performs communication compensation correction on the safety distance parameter: that is, it adjusts the size of the safety distance parameter according to the communication quality (adjacency edge weight) between online agents. The worse the communication quality, the larger the safety distance parameter; the better the communication quality, the safety distance parameter can maintain its initial value or be appropriately reduced. For example, in this embodiment, good communication: the compensation correction coefficient is 1.0 (no adjustment); average communication: the compensation correction coefficient is 1.2 (increased by 20%); poor communication: the compensation correction coefficient is 1.5 (increased by 50%). The corrected safety distance parameter is integrated into the cooperative motion constraints, and other relevant parameters in the cooperative motion constraints are updated synchronously to obtain the cooperative motion constraints after communication limitation correction.

[0124] Step 510: Based on the cooperative motion constraints corrected for communication limitations, the updated independent task execution order of each agent, and the unified spatiotemporal perception information, task allocation and cooperative trajectory planning are performed to generate a collision-free cooperative operation trajectory for the online agents.

[0125] Optionally, the cooperative control system takes the cooperative motion constraints corrected for communication limitations, the updated independent task execution order of each agent, and unified spatiotemporal perception information as input, and performs task allocation and cooperative trajectory planning again, that is, repeats the process from steps 501 to 505. This generates a collision-free cooperative operation trajectory for online agents that is only applicable to the current online agents, collision-free, conflict-free, and capable of continuous operation.

[0126] This invention enables autonomous recovery and continuous operation of multi-agent clusters in special confined spaces under scenarios of weak network, disconnection, and communication interruption by real-time monitoring of communication status, rapid reconstruction of neighborhood topology when communication is abnormal, reassignment of lost tasks to nearby locations, communication compensation correction for safe distances, and online replanning of agent trajectories. It completely solves the problems of topology collapse, task interruption, and operation failure caused by communication abnormalities in traditional collaborative systems, significantly improves the fault tolerance and environmental adaptability of distributed decentralized collaborative systems, and ensures the continuity and safety of the entire operation process.

[0127] Furthermore, the multi-agent swarm intelligent collaborative control system for special confined spaces provided by the present invention will be described below. The multi-agent swarm intelligent collaborative control system for special confined spaces described below can be referred to in correspondence with the multi-agent swarm intelligent collaborative control method for special confined spaces described above.

[0128] Optional, refer to Figure 2 , Figure 2This is a schematic diagram of the structure of the multi-agent swarm intelligent collaborative control system for special confined spaces provided by the present invention. The multi-agent swarm intelligent collaborative control system for special confined spaces includes: The multi-source perception and spatiotemporal alignment module 210 is used to perform spatiotemporal synchronization alignment based on the local perception information of each intelligent agent in a special limited space and the state information of neighboring machines to obtain the unified spatiotemporal perception information of each intelligent agent. The group decision-making module 220 is used to make collaborative group decisions based on unified spatiotemporal perception information and by interacting with local information from neighboring machines to obtain the group decision results. The event-triggered update module 230 is used to make a trigger judgment based on the group decision results and the preset event-triggered communication conditions, and update the state of each intelligent agent based on the trigger judgment results to obtain the neighbor machine state information group that is updated as needed. The conflict level decoupling module 240 is used to decouple the collision risk between multiple agents based on the group decision results and the neighbor machine state information group, so as to obtain conflict-free cooperative motion constraints. The task allocation and collaborative trajectory planning module 250 is used to allocate tasks and plan collaborative trajectories based on collaborative motion constraints and unified spatiotemporal perception information, generate collision-free collaborative operation trajectories for each intelligent agent, and enable each intelligent agent to execute collaborative operation motion based on the instructions transformed from the collision-free collaborative operation trajectory.

[0129] The embodiments of the present invention solve the technical problem of collaborative failure of distributed cooperative control methods based on periodic communication in special confined spaces in the background art, and realize the improvement of the safety and continuity of multi-agent cooperative operation in special confined spaces, thereby adapting to the requirements of complex operating environment in special confined spaces.

[0130] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an electronic device 300 provided in this embodiment of the invention includes a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the processes of steps 10 to 50.

[0131] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it implements the processes of steps 10 to 50.

[0132] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-agent swarm intelligent cooperative control method for special confined spaces provided by the above methods, which includes steps 10 to 50.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-agent swarm intelligent cooperative control method for special confined spaces, characterized in that, include: Based on the spatiotemporal synchronization and alignment of the local perception information of each intelligent agent with the state information of neighboring machines in a special limited space, the unified spatiotemporal perception information of each intelligent agent is obtained. Based on the unified spatiotemporal perception information, each intelligent agent makes collaborative group decisions based on local information interaction with neighboring machines, and obtains the group decision results; Based on the group decision-making results and preset event-triggered communication conditions, a trigger judgment is made, and the state of each agent is updated based on the trigger judgment results to obtain a neighboring machine state information group that is updated as needed. Based on the group decision-making results and the neighbor machine state information group, the collision risk between multiple agents is decoupled at the conflict level to obtain conflict-free cooperative motion constraints. Based on the cooperative motion constraints and the unified spatiotemporal perception information, task allocation and cooperative trajectory planning are performed to generate collision-free cooperative operation trajectories for each agent, and each agent executes cooperative operation motion based on the instructions transformed from the collision-free cooperative operation trajectories.

2. The multi-agent swarm intelligent cooperative control method for special confined spaces according to claim 1, characterized in that, The local perception information includes local obstacle information and its own pose information. The local obstacle information includes the obstacle's location, size, and motion state. The own pose information includes its position, attitude, and speed. The neighboring machine status information includes the neighboring machine's location, operating status, and remaining energy consumption. The process of obtaining the group decision-making results includes: Based on the location and size of the obstacle in the unified spatiotemporal perception information and the position and posture of the agent itself, a spatial relative coordinate mapping is performed to obtain a real-time avoidance gap distribution map of each agent relative to the local obstacle. Based on the real-time avoidance gap distribution map and a preset safety threshold, the narrow passage bottleneck with limited width is identified to obtain a list of narrow section nodes. Physical containment verification is performed based on the minimum passage cross-sectional size of each bottleneck node in the narrow section node list and the maximum external dimensions of each intelligent agent body to obtain a pass status identifier. One-way passage qualification is determined based on the pass status identifier to generate a one-way passage intelligent agent catalog. Based on the identity identifier of each intelligent agent in the one-way passage intelligent agent catalog and the neighboring machine's operating status and remaining energy consumption in the neighboring machine status information, the passage urgency rule is matched to obtain the priority passage right level of each intelligent agent in local interaction, and the space occupancy right is allocated based on the priority passage right level to obtain the channel occupancy right allocation record table. Based on the channel occupancy allocation record table and the task dependencies in the preset global task target sequence, rate adjustment and timing coordination are performed to obtain the group decision result.

3. The multi-agent swarm intelligent cooperative control method for special confined spaces according to claim 2, characterized in that, The process of adjusting the rate and coordinating the timing based on the channel occupancy allocation record table and the task dependencies in the preset task list to obtain the group decision result includes: Based on the priority occupying agent identifier in the channel occupancy right allocation record table and the task dependency relationship with the preset global task target sequence, the independent task execution initial sequence of each agent is obtained, and a task logic dependency topology graph is generated based on the independent task execution initial sequence. Based on the task triggering order in the task logic dependency topology graph, the relative motion trend is deduced from the velocity in the self-pose information and the neighboring machine position in the neighboring machine state information. The rate interference state identifier between multiple agents is obtained, and the rate suppression interval is defined based on the rate interference state identifier to generate an initial speed adjustment amplitude mapping table. Based on the rate suppression interval of each agent in the initial speed adjustment range mapping table and the remaining energy consumption in the neighboring machine state information, an energy consumption safety boundary check is performed to obtain the allowable lower limit of speed that meets the minimum energy consumption operation requirements. Based on the allowable lower limit of speed, the initial speed adjustment range is pruned to obtain a speed adjustment range list. Based on the speed adjustment range in the speed adjustment range list and the task triggering order in the task logic dependency topology graph, a timing alignment mapping is performed to obtain the task execution timing correction parameters for each agent. Based on the task execution timing correction parameters, the independent task execution initial sequence of each agent is time-corrected to obtain the task execution timing correction sequence. Based on the task execution timing correction sequence and the spatial occupancy allocation information in the channel occupancy allocation record table, conflict elimination is performed to obtain a conflict-free group decision result that includes the independent task execution order and speed adjustment range of each agent.

4. The multi-agent swarm intelligent cooperative control method for special confined spaces according to claim 1, characterized in that, The process of decoupling the collision risk among multiple agents based on the group decision-making results and the neighboring machine state information group to obtain conflict-free cooperative motion constraints includes: Based on the execution order of each agent's independent tasks in the group decision-making results and the position coordinates of the neighboring machines in the neighboring machine status information group, static spatial occupancy areas are extracted to obtain static operation interference areas. Based on the static operation interference areas and the obstacle movement status in the local perception information, dynamic trajectory extrapolation is performed to generate a conflict risk level map. Based on the boundary coordinates of high-risk areas in the conflict risk hierarchy map and the current attitude angle and body shape of each agent, the attitude envelope is geometrically expanded to obtain the dynamic safety envelope boundary of each agent. Based on the dynamic safety envelope boundary, the geometric restricted area is delineated to generate a geometric restricted area constraint list. Based on the spatial range of each restricted area in the geometric restricted area constraint list and the speed adjustment range and current speed of each agent in the group decision result, the uniform motion time occupation is mapped to obtain the expected time occupation interval of each agent. Based on the expected time occupation interval, motion interference factors in the time dimension are removed to generate a temporal isolation constraint window sequence. The constraints are solved and output based on the temporal isolation constraint window sequence, the remaining energy consumption of neighboring machines, the list of geometrically restricted areas, and the operating status of neighboring machines, to obtain the conflict-free cooperative motion constraints.

5. The multi-agent swarm intelligent cooperative control method for special confined spaces according to claim 4, characterized in that, The constraint solution and output based on the temporal isolation constraint window sequence, the remaining energy consumption of neighboring machines, the list of geometrically restricted areas, and the operating status of neighboring machines yield the conflict-free cooperative motion constraints, including: Based on the time occupancy intervals in the time-series isolation constraint window sequence and the remaining energy consumption of the neighboring machine, a low-power passage grace rule is matched to obtain the time window margin expansion result. Based on the time window margin expansion result, a passage order constraint rule table that takes into account the energy consumption stability is constructed. Based on the forced passage order in the passage order constraint rule table, the kinematic steering capability boundary is compared with the maximum allowable turning angular velocity and current velocity of each agent to obtain a set of feasible heading angles. Invalid heading instructions are then eliminated based on the set of feasible heading angles to obtain a list of feasible heading constraint domains. Based on the feasible heading angles in the feasible domain list of heading constraints and the prohibited areas in the prohibited areas list of geometric prohibited areas, spatial trajectory non-intersection is determined to obtain safe passage trajectory segments. Based on the safe passage trajectory segments, redundant motion paths with trajectory intersection risks are eliminated to obtain a set of non-intersecting safe trajectory segments. Based on the time occupancy intervals in the temporal isolation constraint window sequence and the safe passage trajectory segments in the non-intersecting safe trajectory segment set, spatiotemporal coordinate synchronization and alignment mapping is performed to obtain the set of allowed motion states for each agent. Based on the set of allowed motion states, the state coupling conflict during the joint motion of multiple agents is eliminated to obtain the decoupled set of motion constraints for a single agent. Based on the permissible motion state parameters in the single agent motion constraint set and the adjacent machine's operation status, the high-risk operation safety distance expansion rules are integrated to obtain a unified format motion restriction instruction set. Based on the motion restriction instruction set, the spatial no-entry boundary, time passage window, heading angle limit and safety distance parameters are combined and encapsulated to obtain the conflict-free cooperative motion constraint.

6. The multi-agent swarm intelligent cooperative control method for special confined spaces according to claim 1, characterized in that, The cooperative motion constraints include a set of motion restriction instructions, spatial no-entry boundaries, time passage windows, heading angle limits, and safety distance parameters. The step of assigning tasks and planning cooperative trajectories based on the cooperative motion constraints and the unified spatiotemporal perception information to generate collision-free cooperative operation trajectories for each agent includes: Based on the spatial restricted boundary coordinates in the cooperative motion constraints and the local obstacle positions in the unified spatiotemporal perception information, a geometric exclusion region union processing is performed to obtain a composite static restricted region set. Based on the composite static restricted area set and the safety distance parameter in the cooperative motion constraint, the boundary is expanded outward to generate a geometric passage corridor with safety redundancy for each intelligent agent. Based on the central axis of the geometric passageway, the curvature radius of each turning point is extracted and compared with the heading angle limit in the cooperative motion constraint to obtain a set of feasible path nodes that meet the maximum turning capability limit. Based on the set of feasible path nodes, sharp corner path segments with curvature radii less than the minimum allowable value are eliminated to obtain a geometrically smooth and compliant path skeleton. Based on the geometrically smooth compliant path skeleton and the task order in the group decision results, a task spatiotemporal mapping is performed to obtain the collision-free collaborative operation trajectory of each agent.

7. The multi-agent swarm intelligent cooperative control method for special confined spaces according to claim 6, characterized in that, The step of performing a task spatiotemporal mapping based on the geometrically smooth compliant path skeleton and the task order in the group decision results to obtain the collision-free collaborative operation trajectory of each agent includes: Based on the spatial coordinates of the path nodes in the geometrically smooth and compliant path skeleton and the execution order of the independent tasks of each agent in the group decision results, spatial target index binding is performed to obtain the target waypoint sequence of each agent arranged in the order of task sequence. Based on the target waypoint sequence, the continuous path skeleton is truncated into local navigation segments corresponding to each sub-task, and a local navigation segment mapping table is obtained. Based on the spatial length of each road segment in the local navigation road segment mapping table and the time passage window in the cooperative motion constraint, the passage time at nominal speed is calculated to obtain the target arrival time node of each road segment. Based on the target arrival time node, the local navigation route segment is timestamped to obtain a set of spatiotemporally synchronized and cooperative trajectory nodes. The collision-free collaborative operation trajectory is obtained by aligning and encapsulating the spatial coordinates, timestamps, and heading angles of each node in the spatiotemporal synchronized collaborative trajectory node set with the motion restriction instruction set in the collaborative motion constraints.

8. The multi-agent swarm intelligent cooperative control method for special confined spaces according to any one of claims 1-7, characterized in that, Also includes: Based on the collision-free collaborative operation trajectory, the communication connection status of each intelligent agent in the special confined space is monitored in real time. When the communication connection status is abnormal, a neighborhood connection graph is generated based on the unified spatiotemporal awareness information and the signal quality of neighboring machines, and the neighborhood connection graph is dynamically reconstructed to obtain an updated neighborhood connection graph. Based on the updated neighborhood connection graph and the original independent task execution order in the group decision results, the task nodes corresponding to the lost agents are reassigned to the online agents according to the nearest spatial distance rule, thus obtaining the updated independent task execution order of each agent. Based on the adjacent edge relationships in the updated neighborhood connection graph, the safety distance parameter in the cooperative motion constraint is corrected by communication compensation to obtain the cooperative motion constraint after communication limitation correction. Based on the cooperative motion constraints corrected for communication limitations, the updated independent task execution order of each agent, and the unified spatiotemporal perception information, task allocation and cooperative trajectory planning are performed to generate a collision-free cooperative operation trajectory for online agents.

9. A multi-agent swarm intelligent collaborative control system for special confined spaces, characterized in that: Applied to the multi-agent swarm intelligent cooperative control method for special confined spaces as described in any one of claims 1 to 7; The multi-agent swarm intelligent collaborative control system for special confined spaces includes: The multi-source perception and spatiotemporal alignment module is used to perform spatiotemporal synchronization and alignment based on the local perception information of each intelligent agent in a special limited space and the state information of neighboring machines, so as to obtain the unified spatiotemporal perception information of each intelligent agent. The group decision-making module is used to make collaborative group decisions based on the unified spatiotemporal perception information and the local information interaction with neighboring machines, so as to obtain the group decision-making result. The event-triggered update module is used to make a trigger judgment based on the group decision result and the preset event-triggered communication conditions, and update the state of each agent based on the trigger judgment result to obtain the neighbor state information group that is updated as needed. The conflict-level decoupling module is used to decouple the collision risk between multiple agents based on the group decision results and the neighbor machine state information group, so as to obtain conflict-free cooperative motion constraints. The task allocation and collaborative trajectory planning module is used to allocate tasks and plan collaborative trajectories based on the collaborative motion constraints and the unified spatiotemporal perception information, generate collision-free collaborative operation trajectories for each intelligent agent, and enable each intelligent agent to execute collaborative operation motion based on the instructions transformed from the collision-free collaborative operation trajectory.

10. A non-transitory computer-readable storage medium, characterized in that, The storage medium stores a computer software program, which, when executed by a processor, implements the multi-agent swarm intelligent cooperative control method for special confined spaces as described in any one of claims 1 to 7.

Images