Distributed model predictive control unmanned cluster navigation method based on homotopy perception

Abstract

The invention discloses a distributed model predictive control unmanned cluster navigation method based on homotopy perception. The method comprises the following steps of: planning a global path and optimizing a local track; an optimal path planning algorithm based on homotopy perception is designed by utilizing a topological structure of an environment and space-time conflicts between intelligent agents, each intelligent agent in an unmanned cluster adopts a trajectory optimization method based on model predictive control, a dynamic feasible and collision-free trajectory is generated, the motion trajectory of the intelligent agent is optimized, and the optimal path planning algorithm is realized. The global coordination of the agents in the unmanned cluster and the dynamic adaptability of local motion are improved; and performing real-time dynamic adjustment on behaviors of the agents in the unmanned cluster globally and locally through an online re-planning strategy. By adopting the technical scheme of the invention, the cooperative trajectory planning capability of the unmanned cluster in an obstacle dense environment can be improved.

Description

Technical Field

[0001] The present invention relates to the technology of unmanned cluster intelligent cooperative navigation, and particularly to a distributed model predictive control unmanned cluster trajectory planning and navigation method based on homotopy perception. Background Art

[0002] Unmanned cluster cooperative navigation refers to planning trajectories for an unmanned cluster in a dynamic environment to ensure collision avoidance safety and efficiently reach the target position. The core of this problem lies in simultaneously meeting the two key requirements of safety and efficiency. Although existing research has made significant progress in ensuring safety, it still faces local congestion and deadlock problems in complex environments. Especially in areas with dense obstacles, when multiple agents in an unmanned cluster attempt to cross a narrow passage simultaneously, mutual blocking may occur, resulting in the obstruction or even failure of the navigation task.

[0003] The existing technology mainly uses a "global - local" two - layer framework for unmanned cluster cooperative navigation. At the global level, a single - agent path planning (SAPP) algorithm is usually used to generate a reference path to guide the agent to cross obstacles; at the local level, based on the global path and local safety constraints, a specific motion trajectory is optimized and generated. However, SAPP only considers the optimal path of a single agent, ignoring the cooperative relationship between multiple agents, making it difficult to achieve the global optimum of the entire system, and thus may lead to local congestion and affect the overall navigation efficiency. To improve the global cooperation ability, a multi - agent path finding (MAPF) algorithm is introduced. However, MAPF relies on strict timestamp execution and is difficult to adapt to the uncertainty of dynamic environments in real - world scenarios. In addition, although some existing research attempts to combine MAPF with trajectory optimization, this method does not fully consider the topological structure of the environment and may still cause multiple agents to tend to the same path area, thus exacerbating the local congestion problem.

[0004] To make more effective use of the topological information of the environment, the homotopy method is introduced into the field of trajectory planning. In homotopy theory, if two trajectories can be continuously deformed into each other without crossing obstacles, they belong to the same homotopy class; otherwise, they belong to different homotopy classes. The prior art has utilized the homotopy method to optimize single-agent trajectory planning to avoid getting trapped in local optima in unknown or dynamic environments. In the field of swarm navigation, there are also studies attempting to use homotopic path set planning to guide an indivisible unmanned swarm. However, in the scenario of unmanned swarm cooperative navigation, the existing methods mainly rely on local homotopy adjustment, that is, by adjusting the initial collision trajectories around obstacles to generate different homotopy paths and selecting the path with the lowest cost for obstacle avoidance. However, this method only optimizes the path within a local range and is difficult to effectively alleviate the congestion problem at the global level. Agents may still have path conflicts in critical areas. Therefore, how to reasonably utilize the environmental topological structure globally to achieve efficient multi-agent cooperative navigation remains a technical problem to be solved urgently. Summary of the Invention

[0005] To overcome the deficiencies of the above prior art, the present invention provides a distributed model predictive control unmanned swarm trajectory planning method based on homotopy perception to improve the cooperative trajectory planning ability of an unmanned swarm in an environment with dense obstacles.

[0006] The present invention proposes a novel unmanned cluster distributed trajectory planning framework, enabling the cluster to achieve cooperative navigation in an environment with dense obstacles. The unmanned cluster system can be composed of various autonomous platforms, including but not limited to swarms of autonomous drones, swarms of autonomous navigation vehicles, and other clusters of mobile robots with swarm intelligence. This framework combines global path planning and local trajectory optimization to improve the global coordination of the unmanned cluster and the dynamic adaptability of local motion. At the global level, the present invention proposes an optimal path planning algorithm based on homotopy perception, which makes full use of the topological structure of the environment and considers possible spatio-temporal conflicts among agents in the unmanned cluster. To implement this technology, the present invention designs a complete passage detection mechanism for identifying passable areas in the environment and representing these complete passages with line segment tuples, reflecting the geometric characteristics of the passages in the most concise form possible. Based on these complete passages and the paths of the agents, we can calculate the passage time map to characterize the time distribution characteristics of the agents passing through these passages. Based on the above information, this algorithm can select the optimal path that minimizes spatio-temporal conflicts, thereby improving the overall navigation efficiency. At the local level, each agent adopts a trajectory optimization method based on model predictive control to generate a dynamically feasible and collision-free trajectory. This method can optimize the motion trajectory of the agent on the premise of satisfying the system dynamics constraints, enabling it to safely and efficiently perform navigation tasks in a complex environment. In addition, the present invention also introduces an online replanning strategy, enabling the agents to real-time sense environmental changes and dynamically adjust the behavior of the agents in the unmanned cluster globally and locally to enhance the system's adaptability to dynamic environments.

[0007] The technical solution provided by the present invention is as follows:

[0008] The distributed model predictive control unmanned cluster navigation method based on homotopy perception improves the global coordination of agents in the unmanned cluster and the dynamic adaptability of local motion through global path planning and local trajectory optimization; and then, through the online replanning strategy, dynamically adjusts the behavior of the agents in the unmanned cluster in real time globally and locally;

[0009] Global path planning designs a homotopy-aware optimal path planning algorithm by leveraging the topological structure of the environment and the spatio-temporal conflicts among agents, including: designing a complete passage detection mechanism to identify the topological structure of passages in the environment and representing complete passages as a tuple of line segments reflecting the geometric features of the passages; calculating the passage time map to obtain the time distribution characteristics of agents passing through the passages; implementing an optimal path that minimizes spatio-temporal conflicts to improve navigation efficiency; local trajectory optimization is that each agent in the unmanned cluster adopts a trajectory optimization method based on model predictive control to generate a dynamic feasible and collision-free trajectory, and optimize the motion trajectory of the agent on the premise of satisfying the system dynamics constraints; and then through an online replanning strategy, the agent can perceive environmental changes in real time and dynamically adjust the global path and local trajectory; thus realizing the navigation of an unmanned cluster based on homotopy-aware distributed model predictive control.

[0010] It includes the following steps:

[0011] 1) Preprocess the obstacle map: including Voronoi Graph generation and complete passage detection;

[0012] 2) Distributed online planning: including an optimal path planning algorithm based on homotopy awareness, communication interaction, local trajectory planning, and replanning.

[0013] 2-1) Optimal path planning algorithm based on homotopy awareness;

[0014] 2-2) Agents conduct communication and interaction: Define the passage time map of an agent to represent the timestamps when the agent passes through different passages. Each agent in the unmanned cluster exchanges the passage time map and its own planned trajectory with each other through communication;

[0015] The present invention defines a passage time map to represent the timestamps when an agent passes through different passages, that is, the passage time map of an agent reflects the time span when the agent passes through different passages. Among them, the timestamps of the passage time map can be obtained by dividing the cumulative path length starting from the starting point of the agent by the preset average speed of the agent. starting point of the agent preset average speed obtained.

[0016] 2-3) Local trajectory planning: Construct a trajectory optimization problem based on model predictive control and solve it to obtain the local trajectory;

[0017] 2-4) Online replanning strategy.

[0018] Furthermore, the method robustness can be improved through the online replanning strategy, and it can adapt to a dynamic interaction environment.

[0019] In an environment with dense obstacles, the agents in the unmanned cluster will avoid collisions by reducing speed or even stopping, so they generally cannot follow the predetermined average speed. . Therefore, the channel time map information calculated may have errors, and global path replanning needs to be dynamically adjusted in real time. At the local level, each agent in the unmanned cluster independently plans its trajectory, and can also ensure that local replanning is continuously carried out with the real-time changes of the environment by applying the trajectory optimization based on model predictive control in 2-3).

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] The present invention proposes a distributed trajectory planning method based on homotopy perception, which can effectively improve the navigation efficiency and safety of unmanned clusters in environments with dense obstacles. By introducing a global path planning algorithm based on homotopy perception, agents can make full use of the topological structure of the environment, avoid selecting channels with spatio-temporal conflicts as much as possible, thereby reducing local congestion and deadlock phenomena and improving overall coordination. At the same time, a trajectory optimization method based on model predictive control (MPC) is adopted to ensure the dynamic feasibility and collision-free of the trajectory, enabling agents to achieve safe and efficient navigation in complex environments. In addition, combined with an online replanning strategy, agents can sense environmental changes in real time and dynamically adjust paths and trajectories to enhance their adaptability to dynamic environments. The synergistic effect of the "global-local" double-layer optimization framework enables the system to improve the efficiency of path planning while ensuring safety in complex scenarios such as dense obstacles and multi-agent interactions. Description of the Drawings

[0022] Figure 1 It is a flowchart of the algorithm framework of the present invention.

[0023] Figure 2 It is a platform module constructed to implement the method of the present invention in specific implementation. Detailed Embodiment

[0024] The present invention will be further described below with reference to the drawings and through embodiments.

[0025] The flow of the distributed model predictive control unmanned cluster trajectory planning algorithm based on homotopy perception provided by the present invention is as Figure 1 shown, and the specific algorithm flow is as follows:

[0026] 1. Obtain the starting state, target state and obstacle map of the agent, define the priority of the agent, and initialize the current local trajectory of the agent as a sequence composed of the initial state.

[0027] 2. According to the obstacle map, perform complete channel detection and establish a Voronoi diagram.

[0028] 3. Set the replanning flag ReplanFlag to true to indicate whether replanning is required.

[0029] 4. Determine whether the agent has reached the end point. If the agent has reached the end point, the algorithm ends; if the agent has not reached the end point, loop through steps 4 - 9.

[0030] 5. If ReplanFlag is true, perform the optimal path planning algorithm based on homotopy perception to obtain the globally optimal path with the lowest cost, calculate the channel time graph corresponding to this path, and set ReplanFlag to false.

[0031] 6. Communicate with neighboring agents to exchange the channel time graphs corresponding to the global paths of each agent and the current trajectory data obtained by the model predictive control of each agent.

[0032] 7. Construct the objective function and corresponding constraints of the local trajectory optimization problem, solve the trajectory optimization problem, and obtain the current local trajectory of the agent.

[0033] 8. Publish the current local trajectory of the agent to the underlying controller to execute the trajectory.

[0034] 9. Perform replanning judgment and update the replanning flag ReplanFlag according to the replanning judgment result.

[0035] Specifically, when implemented, the distributed model predictive control unmanned cluster navigation method based on homotopy perception provided by the present invention includes the following steps:

[0036] 1) Preprocessing of the obstacle map: including Voronoi Graph generation and complete channel detection;

[0037] 1 - 1) Expand all obstacles in the environment to obtain the expanded obstacle set;

[0038] Assume that the environment only contains convex obstacles, and the set composed of all obstacles is denoted as . The radius of the agent itself can be used to expand to obtain the expanded obstacle set

[0039] 1 - 2) Perform spatial segmentation on the obstacle map composed of the expanded obstacle set to obtain multiple regions containing individual obstacles and generate a Voronoi Graph;

[0040] For the one composed of The obstacle map is used for space segmentation, dividing the space into several regions each containing a single obstacle. The Euclidean distance from any point in each region to the obstacle in that region is closer than to other obstacles. After discretizing the skeleton graph formed by the boundaries of all these regions into a graph structure, a Voronoi diagram can be obtained. For any starting and ending points in the space, they can be connected to the Voronoi diagram by finding the points on the Voronoi diagram closest to them, and then the graph search algorithm can be applied to find the path.

[0041] 1 - 3) For each ordered pair of obstacles in the dilated obstacle set, generate the corresponding passage. By performing a complete passage detection on the passage design conditions, obtain the complete passage formed by the ordered pair of obstacles.

[0042] For the dilated obstacle set each ordered pair of obstacles in it can generate a passage, denoted as . Through the extended visibility check technique proposed by Jing Huang et al. (J. Huang, Y. Tang, and K. W. Samuel Au, “Homotopic path set planning for robot manipulation and navigation,” Robotics: Science and Systems, 2024), the valid passages in all passages can be retained and the redundant passages can be filtered out. On this basis, denote the shortest line segment in the ordered pair of obstacles as . Along the direction perpendicular to and its opposite direction , translate the straight line corresponding to the shortest line segment until one of the following three conditions is not satisfied: , that is, when the straight line translates along or direction, it has an intersection with the obstacle ; , that is, when the straight line translates along or direction, it has an intersection with the obstacle ; , , that is, when the straight line translates along or direction, if it has an intersection with the obstacles and intersect with each other, and at this time the length of the intersecting line segment is not greater than . Among them represents the straight line passing through two specified points, is the step size during translation, is the maximum width of the custom channel. Thus, the complete channel formed by the ordered pair of obstacles can be expressed as , where and are the line segments that can reach the farthest while satisfying the above three conditions.

[0043] 2) Distributed online planning: It includes a homotopy-aware optimal path planning algorithm, communication interaction, local trajectory planning, and replanning.

[0044] After the preprocessing of the obstacle map in the previous step, the distributed online planning part will loop according to the real-time dynamic interaction of the unmanned cluster until each agent reaches its own end point. The main parts will be introduced in turn below.

[0045] 2-1) Design a homotopy-aware optimal path planning algorithm to find a path with the least spatio-temporal conflict degree for each agent in the unmanned cluster, that is, the globally optimal path with the least cost;

[0046] Based on the classical A* algorithm, the present invention proposes an extended design that can fully realize the spatio-temporal coordination of the global path of the unmanned cluster, aiming to find a path with the least spatio-temporal conflict for each agent in the cluster. We can use to represent this path, and its functional expression form is , where represents the two-dimensional real number space, is after removing from the free space. The cost function designed by the present invention is:

[0047]

[0048] Among them, is the cost function, with as the independent variable; is the independent variable, corresponding to the independent variable of the aforementioned ; the first two terms of the right-side expression are the classical A* cost function design, represents from to the cumulative path length, while The length of the narrowest passage passed. Denote the set of all complete passages passed from to as , then represents the length of the narrowest passage passed by the path so far, and is the weight of this item. With the design of this item, the agents in the unmanned cluster will be guided to wider and more conducive passages for local coordination during global path planning.

[0049] The fourth item is used to achieve path coordination within the unmanned cluster. For agent , when it plans, it only needs to consider the channel time graph of agents with higher priority than it (for example, the smaller the label, the higher the priority). The channel time graph reflects the timestamps when an agent passes through different channels. If the preset average speed of an agent is , then the timestamps of its channel time graph can be obtained by dividing the cumulative path length starting from the starting point of agent by . For example, for the complete passage composed of the ordered pair of obstacles , as described above, the complete passage can be represented as . If agent passes through this passage, then the channel time graph will record the time when agent passes through from or , which is the timestamp when agent enters or leaves this passage. And this timestamp is obtained by dividing the cumulative path length from the starting point of agent to the intersection point of the global path and the line segment or by . To measure the spatio-temporal conflict degree between agent and the global path at the th element in the complete channel set and , a conflict index is defined. If agent does not pass through the complete passage , . Otherwise, denote the time span intervals when agent and pass through this complete passage as and respectively, represent agent The timestamps for entering and leaving the current complete channel, for the agent is the same for symbolic recording. If , then . If , then , where , is a parameter reflecting the tolerance for time conflicts. With the definition of the conflict metric, the cost function term used to achieve path coordination within the unmanned swarm can be expressed as:

[0050]

[0051] where, is used to represent the number of elements in the complete channel set .

[0052] According to the cost function designed above, applying this algorithm on the Voronoi diagram can obtain the path with the optimal cost. The specific process of the algorithm is the same as that of the classical A* algorithm. Since the Voronoi diagram gives the topological structure of the free space, all paths connecting different homotopy classes of the fixed start and end points can be generated in the Voronoi diagram. Due to inheriting the optimality property of the A* algorithm in the design, finding the path with the optimal cost by applying this algorithm can be regarded as combining two operations into one: exploring different homotopy class paths; finding the path with the minimum cost in (1). Different from the exploration of different homotopy class paths in the sense of pure space topology, this algorithm considers time conflicts in the design and integrates time characteristics into path coordination, thus achieving spatio-temporal coordination among agents in the unmanned swarm.

[0053] 2-2) Communication interaction: Each agent in the unmanned swarm exchanges the channel time graph and its own planned trajectory with each other through communication;

[0054] 2-3) Local trajectory planning;

[0055] Denote that the agents in the unmanned swarm system have the following dynamic model

[0056]

[0057] where, , represents the horizon length of model predictive control, is a set of natural numbers from 1 to ; is the state planned by agent at time , is the position vector at this time, is the velocity vector at this time; is the sampling time; is the planned control input; , . In addition, the agent has the following dynamic constraints

[0058]

[0059] where and are positive definite matrices, and are the maximum acceleration and maximum velocity respectively.

[0060] According to the literature (Chen, Yuda, Meng Guo, and Zhongkui Li. "Deadlock resolution and recursive feasibility in mpc-based multi-robot trajectory generation." IEEE Transactions on Automatic Control (2024)), the collision avoidance safety constraint conditions with other agents can be established in the following form

[0061]

[0062] where and determine a linear partition surface, represents the position vector;

[0063] According to the literature (Chen, Yuda, et al. "Multi-robot trajectory planning with feasibility guarantee and deadlock resolution: An obstacle-dense environment." IEEE Robotics and Automation Letters 8.4 (2023): 2197-2204), the safety corridor in the obstacle environment is in the following form

[0064]

[0065] where and determine a linear partition surface, represents the position vector.

[0066] For the path For example, using average speed Calculate the timestamp of each location. In the present invention, in order to ensure accurate tracking of the path, a list of traction points will be selected in each planning step. In each planning step, the trajectory optimization problem will calculate The state at the time, and therefore the path at the time Point on The objective function of the trajectory optimization problem is is defined as follows

[0067]

[0068] in, It is for Step position vector Distance to traction point The penalty weight imposed by distance; It is the control input The trajectory optimization problem is formulated as an optimal control problem, which seeks to minimize (8) under constraints (3)-(7). and , from which we can get the corresponding agent's future The local trajectory of the step.

[0069] 2-4) Online replanning strategy.

[0070] In a space with dense obstacles, the agents in the unmanned swarm will slow down or even stop to avoid collisions, so they generally cannot follow the predetermined average speed. Therefore, The calculated channel time graph information may contain errors, and global path replanning needs to be adjusted dynamically in real time. At the local level, each agent in the unmanned swarm plans its trajectory independently, and can also ensure that local replanning is carried out continuously as the environment changes in real time by applying trajectory optimization based on model predictive control in 2-3).

[0071] In order to achieve global path-level replanning, we first define a time error index , which is the tracking error at the next sampling point divided by the velocity of the next sampling point planned by the trajectory planning solution If Greater than a pre-set threshold , then the agent It will update its own channel time map according to the new average speed and broadcast the updated channel time map to other agents. Based on its own new channel time diagram, it is calculated according to formula (2) Evaluate the current path's spatial and temporal conflict. If the result is greater than a threshold , then the agent The global path will be replanned; otherwise, the agent The timestamp of the path will only be updated with the new average speed. Less than threshold , Agent It will check whether the agent with higher priority has updated the channel time map. If so, the agent The current path will be re-evaluated for time and space conflicts and a decision will be made whether to re-plan. If there is no update, no re-planning will be performed. Depend on and Composition, speed here It can also be solved by optimization problem. In optimization problem, the current real-time position is usually step, the next sampling point is usually the result of the optimization problem corresponding to step.

[0072] Figure 2 The experimental platform for implementing the present invention is shown, including three modules: algorithm end module, communication and computing module, and execution layer module. The algorithm end module includes the unmanned cluster collaborative navigation algorithm and provides an API interface for experimenters to modify the algorithm and parameters. The communication and computing module includes computing platforms such as airborne computers and ground station hosts, and mainly provides multiple distributed computing terminals and data communication media. The execution layer module includes the Optitrack positioning system and unmanned platforms such as crazyflies. The positioning systems such as Optitack mainly provide the individual position information required in the control algorithm, and unmanned platforms such as crazyflies are the intelligent entities of the unmanned cluster.

[0073] The present invention provides the above-mentioned distributed model predictive control unmanned cluster trajectory planning method based on homology perception. By designing a global path planning algorithm based on homology perception, the intelligent agent can make full use of the topological structure of the environment and avoid selecting channels with time and space conflicts as much as possible, thereby reducing local congestion and deadlock phenomena and improving overall coordination. At the same time, a trajectory optimization method based on model predictive control (MPC) is adopted to ensure the dynamic feasibility and collision-free nature of the trajectory, so that the intelligent agent can achieve safe and efficient navigation in a complex environment. By designing an online replanning strategy, the intelligent agent can perceive environmental changes in real time and dynamically adjust the trajectory to enhance its adaptability to dynamic environments. Therefore, the use of the present invention can improve the collaborative trajectory planning capabilities of unmanned clusters in obstacle-dense environments, and enable unmanned clusters to achieve collaborative navigation in obstacle-dense environments.

[0074] It should be noted that the purpose of disclosing the embodiments is to help further understand the present invention. However, those skilled in the art can understand that various substitutions and modifications are possible without departing from the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the scope of protection claimed by the present invention shall be subject to the scope defined by the claims.

Claims

1. A distributed model predictive control unmanned cluster navigation method based on homotopy perception, characterized in that Through global path planning and local trajectory optimization, the global coordination of agents in an unmanned cluster and the dynamic adaptability of local motion are improved; then, through an online replanning strategy, the behaviors of agents in the unmanned cluster are dynamically adjusted in real time globally and locally; Global path planning designs a homotopy-aware optimal path planning algorithm by utilizing the topological structure of the environment and the spatio-temporal conflicts occurring among agents, including: designing a complete channel detection mechanism for identifying the topological structure of channels in the environment and representing the complete channels as a tuple of line segments reflecting the geometric features of the channels; Calculating the channel time graph to obtain the time distribution characteristics of agents crossing the channels; realizing the optimal path with minimized spatio-temporal conflicts to improve the navigation efficiency; Local trajectory optimization is that each agent in the unmanned cluster adopts a trajectory optimization method based on model predictive control to generate a dynamically feasible and collision-free trajectory, and optimize the motion trajectory of the agent on the premise of satisfying the system dynamics constraints; Then, through the online replanning strategy, the agents can perceive environmental changes in real time and dynamically adjust the global path and local trajectory; Thus, the navigation of the unmanned cluster based on homotopy-aware distributed model predictive control is realized.

2. The homotopy perception-based distributed model predictive control unmanned cluster navigation method according to claim 1, characterized in that The method includes the following steps: 1) Obtain the starting state, target state, and obstacle map of agents in the unmanned cluster, define the priorities of the agents, and initialize the current local trajectory of the agents as a sequence composed of the initial state; 2) Preprocess the obstacle map, including complete channel detection and construction of the Voronoi diagram; For any starting point and ending point in space, find a path through the Voronoi diagram; Through complete channel detection, identify the topological structure of channels in the environment, extract geometric features, and calculate the channel time graph; the channel time graph represents the timestamps when agents pass through different channels; the timestamps of the channel time graph of an agent are obtained by dividing the cumulative path length from the starting point of the agent to the entrance or exit of the channel by the average speed of the agent; 3) Set a replanning flag to indicate whether replanning is required; 4) Judge whether the agent has reached the end point. If the agent has reached the end point, the algorithm ends; if the agent has not reached the end point, set the replanning flag to true, and loop through steps 4) - 7); 5) If the replanning flag is true, perform a homotopy-aware optimal path planning algorithm on the Voronoi diagram to find a path with the minimum spatio-temporal conflict, that is, the path with the optimal cost, for each agent in the unmanned cluster, thereby obtaining the global path; And set the replanning flag to false; The homotopy-aware optimal path planning algorithm includes: Design a cost function, expressed as: , where is the cost function; is the independent variable; represents the cumulative path length from to ; represents the Euclidean distance from to the end point; represents the term used to penalize the length of the narrowest channel passed by the path ; represents the term used to achieve path coordination within the unmanned cluster; defines the spatio-temporal conflict degree index, which represents the spatio-temporal conflict degree of the global paths of two agents in the complete channel, denoted as ; further, is expressed as: , where represents the number of elements in the complete channel set , the th element of is denoted as Based on the cost function, perform homotopy-aware optimal path planning on the Voronoi diagram to obtain the path with the optimal cost; when performing optimal path planning for the agent consider the channel time graph of the agent with a higher priority to obtain the timestamps when the agent enters and leaves the complete channel through the channel time graph; 6) Communicate with neighboring agents, and the two exchange the channel time graph corresponding to the global path of the agent and the current trajectory data obtained by the agent's model predictive control; 7) Construct the objective function and corresponding constraints for local trajectory optimization and solve them to obtain the current local trajectory of the agent; including: Define the objective function for local trajectory optimization as: , where is the penalty weight imposed on the distance between the position vector at the step and the corresponding traction point ; is the penalty weight imposed on the control input ; Define the dynamic model of the agent, dynamic constraints, anti-collision safety constraints between agents, and safety corridors in the obstacle environment; Solve the minimization of the objective function under the corresponding constraints to obtain the local trajectory of the agent in the future at the current moment. steps.

3. The homotopy perception-based distributed model predictive control unmanned cluster navigation method according to claim 2, wherein Step 2) The preprocessing of the obstacle map specifically includes the following steps: a) Expand all obstacles in the environment using the radius of the agent itself to obtain the expanded obstacle set; b) Perform spatial segmentation on the obstacle map composed of the expanded obstacle set to obtain multiple regions containing individual obstacles and generate a Voronoi diagram; search for a path according to the Voronoi diagram using a graph search algorithm; c) Generate corresponding channels for the ordered pair of obstacles in the expanded obstacle set, and obtain the complete channels formed by the ordered pair of obstacles by performing a complete channel detection on the channel design conditions; Along the direction perpendicular to the shortest line segment in the ordered pair of obstacles and its opposite direction until one of the following three conditions is not satisfied, translate the straight line : Straight line Along Or When translating in the direction, there are intersections with obstacles There are intersections Straight line Along Or When translating in the direction, there are intersections with obstacles There are intersections; Straight line When translating along or direction, if it intersects with obstacles and both, the length of the intersecting line segment at this time is not greater than ; is the maximum width of the custom channel; Thus, the ordered pair of obstacles constitute a complete passageway represented as , where and are the farthest-reachable line segments that satisfy the above three conditions.

4. The homotopy perception-based distributed model predictive control unmanned cluster navigation method according to claim 3, characterized in that, Design an online replanning strategy, including dynamically adjusting the global path in real time; each agent in the unmanned cluster independently plans its trajectory.

5. The homotopy perception-based distributed model predictive control unmanned cluster navigation method according to claim 4, wherein Dynamically adjust the global path in real time through the replanning of the global path, including: First, define a time error metric, which is obtained by dividing the tracking error at the next sampling point by the speed planned at the next sampling point obtained from the trajectory planning solution; If the time error index is greater than the set threshold , the agent updates its own channel time graph according to the new average speed and broadcasts the updated channel time graph to other agents; Agent will calculate the spatio-temporal conflict degree of the current path based on its own new channel time map; if the spatio-temporal conflict degree is greater than the set threshold , then the agent will re-plan the global path; otherwise, the agent updates the timestamp of the path with the new average speed; If the time error index is less than the set threshold , check whether the agent with a higher priority than this agent has updated the channel time map; if there is an update, this agent re-evaluates the spatio-temporal conflict degree of the current path and determines whether to re-plan; if there is no update, no re-planning is performed.

6. The homotopy perception-based distributed model predictive control unmanned cluster navigation method according to claim 5, wherein Define the spatio-temporal conflict degree index , indicating the spatio-temporal conflict degree between the agent at the th element of the complete channel set and the global path; If the agent does not pass through the complete channel ; Otherwise, denote the agent and The time-span intervals for passing through this complete channel are respectively and ; If , then ; If , then , where , is a parameter reflecting the allowable degree for time conflicts.

7. A platform system for implementing the homotopy-aware distributed model predictive control unmanned cluster navigation method described in claim 1, characterized in that, It includes an algorithmic module, a communication and computing module, and an execution layer module; among them, the algorithmic module contains an unmanned cluster cooperative navigation algorithm and provides an API interface for modifying algorithms and parameters; the communication and computing module contains an on-board computer and a ground station host computing platform, providing multiple distributed computing terminals and data communication media; the execution layer module contains a positioning system and an unmanned platform.

8. The platform system according to claim 7, characterized in that The execution layer module contains a positioning system, specifically the Optitrack positioning system, which is used to provide the individual position information required for controlling the unmanned cluster cooperative navigation algorithm.

9. The platform system according to claim 7, characterized in that, The unmanned platform specifically uses the crazyflies unmanned platform as the agent of the unmanned cluster.

Images