Unmanned cluster navigation method based on distributed model predictive control of homotopy perception

Abstract

The application discloses a kind of distributed model prediction control unmanned cluster navigation methods based on homotopy perception, through global path planning and local trajectory optimization;Through the use of the topological structure of environment and the space-time conflict between agents, an optimal path planning algorithm based on homotopy perception is designed, each agent in the unmanned cluster uses a trajectory optimization method based on model prediction control, generates a dynamic feasible and collision-free trajectory, optimizes the motion trajectory of the agent, and improves the global coordination and local motion dynamic adaptability of the agent in the unmanned cluster;Then, through the online replanning strategy, the behavior of the agent in the unmanned cluster is dynamically adjusted in real time from the global and local. The technical scheme of the application can improve the cooperative trajectory planning capability of the unmanned cluster in the dense obstacle environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned swarm intelligent cooperative navigation, and particularly relates to a distributed model predictive control unmanned swarm trajectory planning navigation method based on homotopy perception. BACKGROUND

[0002] Unmanned swarm cooperative navigation refers to planning trajectories for unmanned swarms in dynamic environments to ensure safe and efficient arrival at target positions while avoiding collisions and obstacles. 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 safety assurance, it still faces local congestion and deadlock problems in complex environments, especially in obstacle-dense areas. When multiple agents in an unmanned swarm attempt to pass through narrow channels at the same time, mutual blocking may occur, leading to navigation tasks being hindered or even unable to be completed.

[0003] Existing technologies mainly use a "global-local" two-layer framework for unmanned swarm cooperative navigation. At the global level, a single-agent path planning (SAPP) algorithm is usually used to generate a reference path to guide agents through obstacles; at the local level, based on the global path and local safety constraints, specific motion trajectories are optimized and generated. However, SAPP only considers the optimal path of a single agent, ignoring the collaborative relationship between multiple agents, making it difficult to achieve global optimization of the entire system, which may lead to local congestion and affect overall navigation efficiency. To improve global collaboration, multi-agent path finding (MAPF) algorithms are introduced. However, MAPF relies on strict timestamps for execution, making it 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 topology of the environment, which may still lead multiple agents to converge on the same path area, exacerbating the problem of local congestion.

[0004] In order to make better 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 used the homotopy method to optimize single-agent trajectory planning to avoid the problem of local optimization in unknown or dynamic environments. In the field of swarm navigation, some studies have also attempted to use homotopic path set planning to guide non-separable unmanned swarms. However, in the scenario of cooperative navigation of unmanned swarms, the existing method mainly relies on local homotopy adjustment, that is, by adjusting the initial collision trajectory around the obstacle to generate different homotopic paths, and selecting the path with the lowest cost to avoid obstacles. However, this method only optimizes the path in a local range, and it is difficult to effectively alleviate the congestion problem from a global perspective, and the agents may still have path conflicts in key areas. Therefore, how to reasonably use the topological structure of the environment in a global range and achieve efficient multi-agent cooperative navigation is still a technical problem to be solved. SUMMARY

[0005] In order to overcome the deficiencies of the prior art described above, the present application provides a distributed model predictive control unmanned swarm trajectory planning method based on homotopy perception, to improve the cooperative trajectory planning capability of unmanned swarms in obstacle-dense environments.

[0006] This paper proposes a novel distributed trajectory planning framework for unmanned swarms, enabling collaborative navigation in obstacle-dense environments. Unmanned swarm systems can be composed of a variety of autonomous platforms, including but not limited to swarms of autonomous drones, swarms of autonomously navigated vehicles, and other mobile robot swarms with swarm intelligence. This framework combines global path planning with local trajectory optimization to improve the global coordination and dynamic adaptability of local motion within the swarm. At the global level, this paper proposes an optimal path planning algorithm based on homotopy perception, which leverages the topology of the environment and considers potential spatiotemporal conflicts between agents in the swarm. To implement this technology, the paper designs a complete passage detection mechanism to identify traversable areas in the environment. These complete passages are represented as line segment tuples, reflecting the geometric characteristics of the passages in the most concise form possible. Based on these complete passages and the agent paths, a passage time map is calculated to characterize the temporal distribution of the agents' traversal of these passages. Based on this information, the algorithm selects the optimal path that minimizes spatiotemporal conflicts, thereby improving overall navigation efficiency. At the local level, each agent employs a trajectory optimization method based on model predictive control to generate dynamically feasible and collision-free trajectories. This method optimizes the agent's motion trajectory while satisfying the system's dynamic constraints, enabling it to safely and efficiently perform navigation tasks in complex environments. Furthermore, the present invention introduces an online replanning strategy that enables the agent to perceive environmental changes in real time and dynamically adjust the behavior of the agents in the unmanned swarm both globally and locally, enhancing the system's adaptability to dynamic environments.

[0007] The technical solutions provided by the present invention are as follows:

[0008] A distributed model predictive control unmanned swarm navigation method based on homology perception improves the global coordination of the intelligent agents in the swarm and the dynamic adaptability of local motion through global path planning and local trajectory optimization. Furthermore, an online replanning strategy is used to dynamically adjust the behavior of the intelligent agents in the swarm in real time, both globally and locally.

[0009] The global path planning is to design an optimal path planning algorithm based on homotopy perception by using the topological structure of the environment and the space-time conflict between agents, including: designing a complete channel detection mechanism for identifying the topological structure of the channel in the environment, and representing the complete channel as a line segment element reflecting the geometric characteristics of the channel; calculating the channel time graph to obtain the time distribution characteristics of the agent crossing the channel; realizing the optimal path of minimizing the space-time conflict and improving the navigation efficiency; the 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 optimizes the motion trajectory of the agent under the premise of meeting the system dynamics constraints; and then through the online replanning strategy, the agent can perceive the environmental changes in real time and dynamically adjust the global path and local trajectory; thus realizing the distributed model predictive control unmanned cluster navigation based on homotopy perception.

[0010] The method comprises the following steps:

[0011] 1) Preprocessing of the obstacle map: including Voronoi graph generation and complete channel detection;

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

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

[0014] 2-2) Communication interaction of the agent: defining the channel time graph of the agent to represent the time stamp of the agent passing through different channels, and each agent in the unmanned cluster exchanges the channel time graph and the planned trajectory of the agent through communication;

[0015] The application defines the channel time graph to represent the time stamp of the agent passing through different channels, that is, the channel time graph of an agent reflects the time span of the agent passing through different channels, wherein the time stamp of the channel time graph can be obtained by dividing the cumulative path length from the starting point of the agent by the preset average speed of the agent.

[0016] 2-3) Local trajectory planning: constructing a trajectory optimization problem based on model predictive control, and solving to obtain a local trajectory;

[0017] 2-4) Online replanning strategy.

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

[0019] ​​​​In a space with dense obstacles, the autonomous swarm will slow down or even stop to avoid collisions, so it is generally impossible to follow the predetermined average speed. Therefore, The calculated channel time graph information may contain errors, requiring real-time dynamic adjustments for global path replanning. At the local level, each agent in the unmanned swarm independently plans its trajectory. Trajectory optimization based on model predictive control (MPC) from step 2-3 can be applied to ensure that local replanning continues as the environment changes in real time.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] This paper proposes a distributed trajectory planning method based on homotopy perception, which can effectively improve the navigation efficiency and safety of unmanned swarms in obstacle-dense environments. By introducing a homotopy-aware global path planning algorithm, the intelligent agent can fully utilize the topological structure of the environment and avoid selecting channels with spatiotemporal conflicts as much as possible, thereby reducing local congestion and deadlock and improving overall coordination. Simultaneously, a trajectory optimization method based on model predictive control (MPC) is adopted to ensure the dynamic feasibility and collision-free nature of the trajectory, enabling the intelligent agent to achieve safe and efficient navigation in complex environments. Furthermore, incorporating an online replanning strategy, the intelligent agent can perceive environmental changes in real time and dynamically adjust its path and trajectory to enhance its adaptability to dynamic environments. The synergistic effect of the "global-local" two-layer optimization framework enables the system to ensure safety while improving path planning efficiency in complex scenarios such as dense obstacles and multi-agent interactions. BRIEF DESCRIPTION OF THE DRAWINGS

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

[0023] Figure 2 It is a platform module constructed to realize the method of the present invention during specific implementation. DETAILED DESCRIPTION

[0024] The present invention will be further described below through embodiments in conjunction with the accompanying drawings.

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

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

[0027] 2. Based on the obstacle map, perform complete channel detection and establish the 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, the algorithm loops through steps 4-9.

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

[0031] 6. Communicate with neighboring agents to exchange the channel time graph corresponding to each agent’s global path and the current trajectory data obtained by each agent’s model predictive control.

[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 intelligent agent.

[0033] 8. Publish the agent’s current local trajectory to the underlying controller and execute the trajectory.

[0034] 9. Perform a replanning judgment and update the replanning flag ReplanFlag based on the replanning judgment result.

[0035] In specific implementation, the distributed model predictive control unmanned cluster navigation method based on homology perception provided by the present invention includes the following steps:

[0036] 1) Obstacle map preprocessing: including Voronoi graph generation and complete channel detection;

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

[0038] Assuming that the environment contains only convex obstacles, the set of all obstacles is recorded as . You can use the agent's own radius right Perform expansion to obtain the expanded obstacle set .

[0039] 1-2) Spatially segment the obstacle map consisting of the expanded obstacle set to obtain multiple regions containing individual obstacles and generate a Voronoi diagram;

[0040] To The constructed obstacle map is spatially segmented, dividing the space into several regions containing individual obstacles. Points in each region have a closer Euclidean distance to the obstacle in that region than to other obstacles. Discretizing the skeleton graph consisting of the boundaries of all these regions into a graph structure yields a Voronoi diagram. For any starting and ending points in space, they can be connected to the Voronoi diagram by finding the points on the Voronoi diagram closest to them, allowing the use of graph search algorithms to find a path.

[0041] 1-3) Generate corresponding channels for the ordered pairs of obstacles in the expanded obstacle set, and obtain the complete channels formed by the ordered pairs of obstacles by performing complete channel detection on the channel design conditions;

[0042] For the expanded obstacle set Each ordered pair of obstacles , they can generate a channel, denoted as Through the extended visibility check technology proposed by Jing Huang et al. (J. Huang, Y. Tang, and KW Samuel Au, “Homotopic path setplanning for robot manipulation and navigation,” Robotics: Science and Systems, 2024), the valid channels in all channels can be retained and the redundant channels can be filtered out. On this basis, the order of obstacles is recorded. The shortest line segment in . Along the perpendicular Direction and the opposite direction , translate the straight line corresponding to the shortest segment Until one of the following three conditions is not met: , that is, a straight line along or When the direction is translated and there is an obstacle There is an intersection; , that is, a straight line along or When the direction is translated and there is an obstacle There is an intersection; , , that is, a straight line along or When the direction is translated, if there is an obstacle and All intersect, and the length of the intersecting line segment is no greater than .in Represents a straight line passing through two specified points. is the step size during translation, Is the maximum width of the customized channel. The complete channel can be expressed as ,in and It is the farthest achievable line segment that satisfies the above three conditions.

[0043] 2) Distributed online planning: includes optimal path planning algorithms based on homology perception, communication interaction, local trajectory planning, and replanning.

[0044] After preprocessing the obstacle map in the previous step, the distributed online planning process loops based on the real-time dynamic interaction of the unmanned swarm until each agent reaches its destination. The following describes each of these key components.

[0045] 2-1) Design an optimal path planning algorithm based on homology perception to find a path with the least spatial and temporal conflict for each agent in the unmanned swarm, that is, a global path with the lowest cost;

[0046] Based on the classic A* algorithm, this paper proposes an extended design that can fully realize the spatiotemporal coordination of the global path of unmanned clusters. The purpose is to provide a Find a path with the least space-time conflict. We can use Represents this path, and its function expression is ,here represents the two-dimensional real space, That is Remove The cost function designed by the present invention is:

[0047]

[0048] in, As the cost function, is the independent variable; is the independent variable, as mentioned above The first two terms of the expression on the right are the classic A* cost function design. Indicates from arrive The cumulative path length, It means from The Euclidean distance to the end point. The third term is used to penalize the path The length of the narrowest passage. arrive The set of all complete channels passed is ,but represents the length of the narrowest channel that the path has passed so far, and With this design, the agents in the unmanned swarm will be guided to wider channels that are more conducive to local coordination when performing global path planning.

[0049] The fourth item is used to achieve path coordination within the unmanned cluster. For example, when planning, it only needs to consider the channel time graphs of agents with higher priority than it (for example, agents with smaller definition numbers have higher priority). The channel time graph reflects the timestamps of an agent passing through different channels. If an agent The default average speed is , then the timestamp of its channel time graph can be obtained from the agent Starting point The cumulative path length at the start is divided by For example, for an ordered pair of obstacles The complete channel composed of , if the agent After passing through the channel, the channel time graph will record the agent from or The time that passes is the agent The timestamp of entering or leaving the channel. This timestamp is the time from the agent Starting point To the global path With line segment or The cumulative length of the path at the intersection point divided by To measure the complete channel set No. Elements Agent and The degree of spatiotemporal conflict of the global path, defining a conflict index If the agent Not from the complete channel Passed, Otherwise, remember the agent and The time spans through the complete channel are and , Represents the intelligent agent The timestamps of entering and leaving the current complete channel are for the agent The same is true for symbolic records. ,but .if ,but ,in , It is a parameter that reflects the tolerance for time conflicts. With the definition of conflict index, the cost function term for path coordination within the unmanned cluster is can be expressed as:

[0050]

[0051] in, Used to represent the complete channel set The number of elements in .

[0052] According to the cost function designed above, the algorithm is applied to the Voronoi diagram to obtain the path with the optimal cost. The specific process of the algorithm is consistent with the classic A* algorithm. Since the Voronoi diagram gives the topological structure of the free space, all paths of different homology classes connecting fixed starting points and end points can be generated in the Voronoi diagram. Since the design inherits the optimality of the A* algorithm, the application of this algorithm to find the path with the optimal cost can be regarded as combining two steps into one: exploring different homology class paths; finding the path with the minimum cost under (1). Different from the exploration of different homology class paths in the sense of pure spatial topology, this algorithm takes time conflicts into consideration in its design and integrates time characteristics into path coordination, thereby achieving spatiotemporal coordination between intelligent agents in the unmanned cluster.

[0053] 2-2) Communication Interaction: Each intelligent agent in the unmanned swarm exchanges channel time maps and their own planned trajectories through communication;

[0054] 2-3) Local trajectory planning;

[0055] The intelligent agent in the unmanned swarm system has the following dynamic model

[0056]

[0057] in, , represents the horizon length of model predictive control, For a range from 1 to The set of natural numbers composed of It is an intelligent agent At the moment The status of the plan, is the position vector at that moment, is the velocity vector at that moment; is the sampling time; Provides control input for planning; , In addition, the agent has the following dynamic constraints,

[0058]

[0059] in, and is a positive definite matrix, and are the maximum acceleration and maximum speed 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 anti-collision safety constraints between the robot and other intelligent agents can be established in the following form:

[0061]

[0062] in, and Determine a linear dividing 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-denseenvironment." IEEE Robotics and Automation Letters 8.4 (2023): 2197-2204), the safe corridor in an obstacle environment is as follows

[0064]

[0065] in, and Determine a linear dividing 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 moment, and therefore the path at the moment Point on The objective function of the trajectory optimization problem is is defined as follows

[0067]

[0068] in, It is the Step position vector Distance corresponding 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 the 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 autonomous swarm will slow down or even stop to avoid collisions, so it is generally impossible to follow the predetermined average speed. Therefore, The calculated channel time graph information may contain errors, requiring real-time dynamic adjustments for global path replanning. At the local level, each agent in the unmanned swarm independently plans its trajectory. Trajectory optimization based on model predictive control (MPC) from step 2-3 can be applied to ensure that local replanning continues as the environment changes in real time.

[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 obtained by solving the trajectory planning If Greater than a pre-set threshold , then the agent It will update its own channel time graph according to the new average speed and broadcast the updated channel time graph to other agents. Calculate the new channel time diagram based on formula (2) Evaluate the current path's spatiotemporal 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 route will only be updated with the new average speed. Less than threshold , agent It will check whether the agent with higher priority than it has updated the channel time map. If so, the agent The current path's time and space conflicts will be re-evaluated and a decision will be made whether to re-plan. If there are no updates, no re-planning will be performed. Depend on and Composition, speed here It can also be solved by optimization problem. In the optimization problem, the current real-time position is usually the step, the next sampling point is usually the result of the optimization problem step.

[0072] Figure 2 The experimental platform for implementing the present invention is shown, which includes three modules: the algorithm-side module, the communication and computing module, and the execution layer module. The algorithm-side 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 power platforms such as airborne computers and ground station hosts, mainly providing multiple distributed computing terminals and data communication media. The execution layer module includes the Optitrack positioning system and unmanned platforms such as crazyflies. Among them, 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 aforementioned distributed model predictive control trajectory planning method for unmanned swarms based on homology perception. By designing a homology-aware global path planning algorithm, the intelligent agent can fully utilize the topological structure of the environment and avoid selecting channels with spatiotemporal conflicts as much as possible, thereby reducing local congestion and deadlock 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, enabling the intelligent agent to achieve safe and efficient navigation in complex environments. By designing an online replanning strategy, the intelligent agent can perceive environmental changes in real time and dynamically adjust its trajectory to enhance its adaptability to dynamic environments. Therefore, the present invention can improve the collaborative trajectory planning capabilities of unmanned swarms in obstacle-dense environments, enabling unmanned swarms to achieve collaborative navigation in obstacle-dense environments.

[0074] It should be noted that the purpose of disclosing the embodiments is to facilitate a further understanding of the present invention. However, those skilled in the art will appreciate that various substitutions and modifications are possible without departing from the scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the contents disclosed in the embodiments, and the scope of protection claimed by the present invention shall be determined by the scope defined in the claims.

Claims

1. A distributed model predictive control unmanned swarm navigation method based on homotopy perception, characterized in that: Through global path planning and local trajectory optimization, the global coordination of the intelligent agents in the unmanned swarm and the dynamic adaptability of local motion are improved. Then, through online replanning strategies, the behavior of the intelligent agents in the unmanned swarm is dynamically adjusted in real time from both global and local perspectives. Global path planning is to design an optimal path planning algorithm based on homotopy perception by utilizing the topological structure of the environment and the spatiotemporal conflicts between intelligent agents. This includes: designing a complete channel detection mechanism to identify the topological structure of the channel in the environment and representing the complete channel as a line segment tuple reflecting the channel's geometric characteristics; Calculate the channel time map and obtain the time distribution characteristics of the intelligent agent crossing the channel; achieve the optimal path that minimizes spatiotemporal conflicts and improves navigation efficiency; Local trajectory optimization is a method in which each agent in an unmanned swarm uses a trajectory optimization method based on model predictive control to generate dynamically feasible and collision-free trajectories, optimizing the motion trajectory of the agent while satisfying the system dynamics constraints. Then, through the online replanning strategy, the intelligent agent can perceive the changes in the environment in real time and dynamically adjust the global path and local trajectory; This realizes distributed model predictive control of unmanned swarm navigation based on homology perception.

2. The distributed model predictive control unmanned swarm navigation method based on homotopy perception as claimed in claim 1, characterized in that: The method comprises the following steps: 1) Obtain the starting state, target state, and obstacle map of the intelligent agents in the unmanned swarm, define the priority of the intelligent agents, and initialize the current local trajectory of the intelligent agents as a sequence of initial states; 2) Preprocessing the obstacle map, including complete channel detection and building a Voronoi diagram; For any starting point and end point in space, find the path through the Voronoi diagram; Through complete channel detection, the topological structure of the channels in the environment is identified, and geometric features are extracted to calculate the channel time graph. The channel time graph represents the timestamps when the agent passes through different channels. The timestamp of the agent's channel time graph is obtained by dividing the cumulative path length from the agent's starting point to the entrance or exit of the channel by the agent's average speed. 3) Set a replanning flag to indicate whether replanning is required; 4) Determine whether the agent has reached the destination. If the agent has reached the destination, the algorithm ends. If the agent has not reached the destination, set the replanning flag to true and loop through steps 4) to 7). 5) If the replanning flag is true, an optimal path planning algorithm based on homology perception is performed on the Voronoi diagram to find a path with minimal spatiotemporal conflict, that is, a path with the lowest cost, for each agent in the unmanned swarm, thereby obtaining a global path. And set the replanning flag to false; The optimal path planning algorithms based on homology perception include: Design cost function, expressed as: ,in, is the cost function; is the independent variable; Indicates from arrive The cumulative path length, Indicates from Euclidean distance to the end point; Indicates the path used for penalty The length of the narrowest passage passed; Represents the item used to achieve path coordination within the unmanned cluster; Define the spatiotemporal conflict degree index, which represents the spatiotemporal conflict degree of the global paths of the two agents in the complete channel, denoted as ; further Expressed as: ,in, Represents the complete channel set The number of elements in , No. Elements are recorded as ; According to the cost function, the optimal path planning based on homology perception is carried out on the Voronoi diagram to obtain the path with the optimal cost; When planning the optimal path, consider the agents with higher priority The channel time graph is used to obtain the timestamps of the agent entering and leaving the complete channel; 6) Communicate with neighboring agents, exchanging channel time graphs corresponding to the agent's global path and 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: The objective function of local trajectory optimization is defined as: ,in, It is the Step position vector Distance corresponding to traction point The penalty weight imposed by distance; It is the control input the weight of the penalty imposed; Define the dynamic model of the intelligent agent, dynamic constraints, anti-collision safety constraints between intelligent agents, and safety corridors in obstacle environments; Solve the minimization objective function under the corresponding constraints and obtain the agent's future The local trajectory of the step.

3. The distributed model predictive control unmanned swarm navigation method based on homotopy perception as claimed in claim 2, characterized in that: Step 2) Preprocessing the obstacle map includes the following steps: a) For all obstacles in the environment, expand the radius of the agent itself to obtain the expanded obstacle set; b) Spatially segmenting the obstacle map consisting of the expanded obstacle set to obtain multiple regions containing individual obstacles and generating a Voronoi diagram; applying a graph search algorithm based on the Voronoi diagram to find a path; c) Generate corresponding channels for the ordered pairs of obstacles in the expanded obstacle set, and obtain the complete channels formed by the ordered pairs of obstacles by performing complete channel detection on the channel design conditions; Along the ordered lines perpendicular to the obstacle The shortest line segment Direction and the opposite direction , translate the line until one of the following three conditions fails: straight line along or When the direction is translated and there is an obstacle There is an intersection; straight line along or When the direction is translated and there is an obstacle There is an intersection; straight line along or When the direction is translated, if there is an obstacle and All intersect, and the length of the intersecting line segment is no greater than ; It is the customized maximum width of the channel; Therefore, the obstacles are orderly The complete channel is represented as ,in and It is the farthest achievable line segment that satisfies the above three conditions.

4. The distributed model predictive control unmanned swarm navigation method based on homology perception as claimed in claim 3, characterized in that: Design an online replanning strategy, including real-time dynamic adjustment of the global path; each agent in the unmanned swarm independently plans its trajectory.

5. The distributed model predictive control unmanned swarm navigation method based on homology perception as claimed in claim 4, characterized in that: The global path is dynamically adjusted in real time through global path replanning, including: First, define the time error index, which is obtained by dividing the tracking error of the next sampling point by the speed of the next sampling point obtained by the trajectory planning solution; If the time error index is greater than the set threshold , agent Update its own channel time graph according to the new average speed and broadcast the updated channel time graph to other agents; Agent The time-space conflict degree of the current path will be calculated based on its own new channel time graph; if the time-space conflict degree is greater than the set threshold , then the agent Replan 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 higher priority than this agent has updated the channel time map; if there is an update, the agent re-evaluates the degree of spatiotemporal conflict of the current path and determines whether to re-plan; if there is no update, no re-planning is performed.

6. The distributed model predictive control unmanned swarm navigation method based on homology perception as claimed in claim 5, characterized in that: Defining the spatial-temporal conflict degree index , indicating that in the complete channel set No. Elements Agent and the degree of spatiotemporal conflict of the global path; If the agent Not from the complete channel Passed, ; Otherwise, remember the agent and The time spans through the complete channel are and ; if ,but ; if ,but ,in , It is a parameter that reflects the tolerance for time conflicts.

7. A platform system for implementing the distributed model predictive control unmanned cluster navigation method based on homology perception as described in claim 1, characterized in that: It includes an algorithm-side module, a communication and computing module, and an execution layer module; among them, the algorithm-side module contains the unmanned cluster collaborative navigation algorithm and provides an API interface for modifying the algorithm and parameters; the communication and computing module contains the onboard computer and the ground station host computing power platform, providing multiple distributed computing terminals and data communication media; the execution layer module includes the positioning system and the unmanned platform.

8. The platform system according to claim 7, wherein: The execution layer module includes a positioning system, specifically the Optitrack positioning system, which is used to provide the individual position information required to control the unmanned swarm collaborative navigation algorithm.

9. The platform system according to claim 7, wherein: The unmanned platform specifically adopts the crazyflies unmanned platform, which is used as the intelligent body of the unmanned swarm.

Images