Multi-agent upheaval obstacle avoidance method fusing A satellite

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a multi-agent fleeing obstacle avoidance method fusing a satellite A. The method comprises the following steps: calculating the position and speed information of each agent by constructing a multi-agent motion model with speed constraint; constructing interaction items among the intelligent agents; meanwhile, an interaction item between the intelligent agent i and the nearest obstacle boundary point k is constructed; according to the motion condition of the agent i, constructing navigation feedback items of the target point and the virtual target point to the agent; sensing areas of the intelligent agents are divided into a safety area, an early warning area, a danger area and a collision area, the distance between the intelligent agents, the distance between the intelligent agent i and a target point and the distance between the intelligent agent i and a nearest obstacle boundary point k are calculated, and gain parameters in interaction items are dynamically adjusted through a fuzzy controller; and integrating all the interaction items to determine the control input of the intelligent agent. According to the invention, the communication requirement among multiple agents is reduced, the dependence on the shape of the obstacle is eliminated, and the real-time obstacle avoidance capability is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a multi-agent swarming obstacle avoidance method integrating an A-star. Background Art

[0002] A multi-agent system is a complex system composed of multiple interacting agents. As an important direction in the study of multi-agent systems, multi-agent swarming aims to enable multiple autonomous agents to achieve coordinated movements similar to natural phenomena such as bird flocks and fish schools through local rules and interactions. Multi-agent swarming technology has broad application prospects in the fields of robot formation control, drone formation, and vehicle autonomous driving. However, the existence of obstacles is inevitable in the movement of multi-agent swarms.

[0003] The artificial potential field method is a classic algorithm commonly used in robot obstacle avoidance and path planning. It converts the robot navigation problem into a process of moving in a potential field by simulating the physical potential field. This method was first proposed by Khatib in the 1980s and is mainly used for obstacle avoidance and path planning of a single robot in a dynamic environment. In the artificial potential field method, the target and obstacles in the environment are regarded as attraction fields and repulsion fields, respectively. Specifically: the target point exerts gravitational force on the robot, causing the robot to move in the direction of the target, and the obstacle exerts repulsive force on the robot, forcing the robot to stay away from the obstacle. In the paper "Flocking for multi-agent dynamic systems: algorithms and theory", Olfati Saber constructed a theoretical framework for multi-agent clustering through artificial potential fields, making important contributions to subsequent research in the field of multi-agents. Based on this framework, Daito Sakai proposed a clustering algorithm that does not distinguish between robots and obstacles in the paper "Flocking for Multirobots Without Distinguishing Robots and Obstacles". However, these methods can easily cause the agent to fall into a local minimum when dealing with complex obstacle combinations or concave areas, resulting in oscillation or stillness. In practical applications, the agent cannot fully grasp the global information of obstacles, and the shapes of obstacles are usually irregular and non-smooth. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-agent swarming obstacle avoidance method integrating A-star, aiming to completely get rid of the constraints of obstacle shape and state information, quickly escape after the agent falls into a local minimum, and improve the environmental adaptability and reliability of multi-agent swarming movement.

[0005] To achieve the above object, the present invention provides a multi-agent swarming obstacle avoidance method integrating A star, comprising the following steps:

[0006] Step 1: Construct a multi-agent motion model with speed constraints and calculate the position and speed information of each agent;

[0007] Step 2: Construct the interaction terms between agents based on the information of other agents within the perception radius of agent i;

[0008] Step 3: Based on the information of obstacles within the perception radius of agent i, construct the interaction term between agent i and its nearest obstacle boundary point k;

[0009] Step 4: According to the movement of agent i, construct a virtual target point and the navigation feedback item of the target point to agent i;

[0010] Step 5: Determine the collision distance, plan the sensing area of ​​agent i and divide it into a safe area, a warning area, a dangerous area, and a collision area, calculate the distances between agents, between agent i and the target point, and between agent i and its nearest obstacle boundary point k, and adjust the size of the gain parameter in the interaction term through the fuzzy controller;

[0011] Step 6: Determine the control input u of agent i by integrating the interaction terms between agents, the interaction terms between agent i and its nearest obstacle boundary point k, and the navigation feedback terms of the target point and the virtual target point to agent i i , and jump to step 1 to achieve swarming obstacle avoidance of multi-agent clusters.

[0012] Preferably, the expression of the intelligent agent motion model is:

[0013]

[0014] Among them, q i (t) represents the position vector of agent i at time t, p i (t) represents the velocity vector of agent i at time t, pmax represents the maximum velocity vector of agent i, u i (t) represents the control input of agent i at time t, represents the interaction term between agents at time t, represents the interaction term between agent i and its nearest obstacle boundary point k at time t, represents the navigation feedback item of the target point and the virtual target point to the agent i at time t, N represents the total number of agents, ||·|| represents the Euclidean norm, δ i Represents the weight coefficient. If the speed of agent i is greater than pmax, then δ i =1, otherwise δi =0.

[0015] Preferably, the intelligent agent i is equipped with a laser radar for identifying other intelligent agents or obstacles and determining the set of intelligent agents and obstacles within the perception radius of the intelligent agent i, where r is the perception radius of the laser radar.

[0016] Preferably, in step 2, the interaction term between agents is used to prevent collision between agents. Contains the position gradient term and the velocity consistency term, and the expression is

[0017]

[0018] in, is the set of all agents perceived by agent i within the perception radius, is the gain parameter of the control input of agent i, is the attraction / repulsion potential energy pair ψ between agent i and agent j α (||p j (t)-p i (t)|| σ ) at position p i The gradient at (t); defines the attraction / repulsion potential between agents where φ α (z)=φ(zd α )ρ h (z / r α ), r α =||r|| σ , d α =||d|| σ , a α , b α and h are positive system parameters, h∈(0,1), b α ≥a α >0, ε is the system parameter, ε>0, a ij (q) is the adjacency matrix of agent j, which is used to represent the connection relationship and weight between agent i and agent j, a ij (q) = ρ h (||q j -q i || σ / r α )∈[0,1],j≠i, impact function

[0019] Preferably, in step 3, the interaction term between agent i and its nearest obstacle boundary point k is used: Prevent collisions between agent i and obstacle boundary point k, interaction term Contains the position gradient term and the velocity consistency term, and the expression is

[0020]

[0021] In the formula, is the set of all obstacle boundary points perceived by agent i within the perception radius, and is the gain parameter of the control input of agent i, The obstacle boundary point k closest to agent i is called the virtual agent, and its position vector is given by express, satisfy The velocity vector is given by express; is the repulsive potential energy function between agent i and its nearest obstacle boundary point k At position p i The gradient at (t);

[0022] Define the repulsive potential energy function ψ between agent i and the nearest obstacle boundary point k β (z) is d β =||d′|| σ , a β 、b β and h are positive system parameters, h∈(0,1), b β ≥a β >0, ε is the system parameter, ε>0, is the rotation matrix, b i,k (q) is the adjacency matrix of the agent, which is used to represent the connection relationship and weight between the agent and the obstacle boundary point k. Impact function

[0023] Preferably, the execution process of step 4 includes the following steps:

[0024] Step 4.1: Combine the local minimum to determine the motion state of agent i, which are target tracking mode and obstacle bypassing mode. When agent i encounters a concave obstacle or a complex obstacle and falls into a local minimum, the obstacle bypassing mode is turned on; if agent i does not detect an obstacle or is not blocked by the obstacle after sensing the obstacle, it is in target tracking mode;

[0025] The judgment formula for agent i falling into a local minimum is:

[0026]

[0027] The above means that the displacement of agent i within a specific time range is less than a certain threshold λ, and the speed at time t is less than a certain threshold κ. When each of the above conditions is met at the same time, it means that agent i falls into a local minimum;

[0028] Step 4.2: If agent i falls into a local minimum and starts the obstacle bypassing mode, a virtual target point is needed to guide agent i to bypass. The location of the virtual target point is determined by the local A-star algorithm.

[0029] Step 4.3: Design the target point and virtual target point pair agent Navigation feedback items for

[0030]

[0031] Where M is the obstacle mark of agent i. and is the gain parameter of the control input of agent i, q γ is the position vector of the target point, p γ is the velocity vector of the target point, is the position vector of the virtual target point corresponding to agent i, is the velocity vector of the virtual target point,

[0032]

[0033] Preferably, the control input u of agent i i for

[0034]

[0035] The present invention provides a multi-agent swarming obstacle avoidance method integrating A star. First, a multi-agent motion model with speed constraints is constructed to calculate the position and speed information of each agent; then, interaction items between agents are constructed; at the same time, interaction items between agent i and its nearest obstacle boundary point k are constructed; finally, according to the motion of agent i, navigation feedback items of target points and virtual target points to the agent are constructed; the sensing area of ​​the agent is planned and divided into a safe area, a warning area, a dangerous area and a collision area, the distance between agents and between agent i and its nearest obstacle boundary point k is calculated, and the size of the gain parameter in the interaction item is adjusted by a fuzzy controller; all interaction items are combined to determine the control input item of the agent. Compared with the prior art, the present invention reduces the communication requirements between multiple agents, eliminates the dependence on obstacle shape and position information, and improves the local real-time obstacle avoidance capability of the agent. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a logic schematic diagram of a multi-agent swarming obstacle avoidance method integrating A-star of the present invention.

[0038] Figure 2 It is a schematic diagram of determining the virtual target point using the local A-star algorithm.

[0039] 2(a) The agent avoids obstacles clockwise;

[0040] 2(b) The agent avoids obstacles counterclockwise.

[0041] Figure 3 It is a schematic diagram of the partition within the perception radius of the intelligent agent of the present invention.

[0042] Figure 4 These are swarming movement trajectory diagrams of 20, 40, 60 and 80 intelligent agents in a multi-obstacle scenario in a specific embodiment of the present invention.

[0043] 4 (a) Evolution trajectory of 20 agents swarming and avoiding obstacles;

[0044] 4(b) Evolution trajectory of 40 agents swarming and avoiding obstacles;

[0045] 4(c) Evolution trajectory of 60 agents swarming and avoiding obstacles;

[0046] 4(d) Evolution trajectory of 80 agents swarming to avoid obstacles.

[0047] Figure 5 It is a schematic diagram of the motion states of different numbers of intelligent agents in a specific embodiment of the present invention.

[0048] 5(a) Schematic diagram of the distance between the center point and the target point of the multi-agent;

[0049] 5(b) The minimum distance result diagram when simulating swarming obstacle avoidance of 20 to 80 intelligent agents. DETAILED DESCRIPTION

[0050] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0051] The present invention provides a multi-agent swarming obstacle avoidance method integrating A-star, comprising the following steps:

[0052] Step 1: Construct a multi-agent motion model with speed constraints and calculate the position and speed information of each agent;

[0053] Step 2: Construct the interaction terms between agents based on the information of other agents within the perception radius of agent i;

[0054] Step 3: Based on the information of obstacles within the perception radius of agent i, construct the interaction term between agent i and its nearest obstacle boundary point k;

[0055] Step 4: According to the movement of agent i, construct a virtual target point and the navigation feedback item of the target point to agent i;

[0056] Step 5: Determine the collision distance, plan the sensing area of ​​agent i and divide it into a safe area, a warning area, a dangerous area, and a collision area, calculate the distances between agents, between agent i and the target point, and between agent i and its nearest obstacle boundary point k, and adjust the size of the gain parameter in the interaction term through the fuzzy controller;

[0057] Step 6: Determine the control input u of agent i by integrating the interaction terms between agents, the interaction terms between agent i and its nearest obstacle boundary point k, and the navigation feedback terms of the target point and the virtual target point to agent i i , and jump to step 1 to achieve swarming obstacle avoidance of multi-agent clusters.

[0058] The following is further described in conjunction with specific embodiments and execution steps:

[0059] See also Figures 1 to 5 , Figure 1 This is a logical block diagram of the multi-agent swarming obstacle avoidance method of the A-star fusion according to the present invention.

[0060] The execution process of step 1 includes the following steps:

[0061] Step 1.1: Construct a multi-agent motion model with speed constraints as

[0062]

[0063] Among them, qi (t) represents the position vector of agent i at time t, p i (t) represents the velocity vector of agent i at time t, pmax represents the maximum velocity vector of agent i, u i (t) represents the control input of agent i at time t, represents the interaction term between agents at time t, represents the interaction term between agent i and its nearest obstacle boundary point k at time t, represents the navigation feedback item of the target point and the virtual target point to the agent i at time t, N represents the total number of agents, δ i represents the weight coefficient, ||·|| represents the Euclidean norm;

[0064] Step 1.2: Design the weight coefficient δ according to the multi-agent motion model with speed constraints i for:

[0065]

[0066] Step 1.3: Determine the weight coefficient δ i , calculate the position vector q of agent i according to the multi-agent motion model with speed constraints i (t) and the velocity vector p i (t).

[0067] Step 2 includes the following steps:

[0068] Step 2.1: Interaction terms between agents Adjust the distance between agents to avoid collisions. The position gradient term in is designed as

[0069]

[0070] In the formula, is the set of all agents perceived by agent i within the perception radius, is the gain parameter of the control input of agent i, is the attraction / repulsion potential energy function pair ψ between agent i and agent j α (||p j (t)-p i (t)|| σ ) at position p i The gradient at (t). Define the attraction / repulsion potential ψ between agents α (z) is where φ α (z) = φ(zd α )ρh (z / r α ), r α =||r|| σ , d α =||d|| σ , a α 、b α and h are positive system parameters, h∈(0,1), b α ≥a α >0, ε is the system parameter, ε>0,

[0071] Step 2.2: Add the interaction terms between agents The speed consistency term in is designed as

[0072]

[0073] In the formula, is the gain parameter of the control input of agent i, a ij (q) is the adjacency matrix of agent j, which is used to represent the connection relationship and weight between agent i and agent j. ij (q) = ρ h (||q j -q i || σ / r α )∈[0,1],j≠i, impulse function ρ h (z) and ρ in step 2.1 h (z) the same;

[0074] Step 2.3: Interaction terms between agents Designed for

[0075]

[0076] The execution process of step 3 includes the following steps:

[0077] Step 3.1: The interaction term between agent i and its nearest obstacle boundary point k within the sensing radius r Adjust the distance between agent i and obstacle boundary point k to avoid collision. The position gradient term in is designed as

[0078]

[0079] In the formula, is the set of all obstacle boundary points perceived by agent i within the perception radius, is the gain parameter of the control input of agent i, The obstacle boundary point closest to agent i is called the virtual agent, and its position vector is given by express, satisfy The velocity vector is given by express. is the repulsive potential energy function between agent i and its nearest obstacle boundary point k At position p i The gradient at (t). Define the repulsive potential function ψ between agent i and the nearest obstacle boundary point k β (z) is Impulse function ρ h (z) and ρ in step 2.1 h (z) same, d β =||d′|| σ , a β 、b β and h are positive system parameters, h∈(0,1), b β ≥a β >0, ε is a system parameter, ε>0;

[0080] Step 3.2: Add the interaction term The speed consistency term in is designed as

[0081]

[0082] In the formula, is the gain parameter of the control input of agent i, b i,k (q) is the adjacency matrix of the agent, which is used to represent the connection relationship and weight between the agent i and the obstacle boundary point k. Impulse function ρ h (z) and ρ in step 2.1 h (z) the same, is the velocity vector of the virtual agent, where is the rotation matrix, The above ensures It ensures that the speed direction of the virtual agent always helps agent i avoid obstacles;

[0083] Step 3.3: The interaction term between agent i and its nearest obstacle boundary point k The final design is

[0084]

[0085] Step 4 includes the following steps:

[0086] Step 4.1: Determine the motion state of agent i and divide the motion mode of agent i into target tracking mode and obstacle bypassing mode. The conditions for determining whether agent i falls into a local minimum are as follows:

[0087]

[0088] The above means that the displacement of agent i within a specific time range is less than a certain threshold λ, and the speed at time t is less than a certain threshold κ. When each of the above conditions is met at the same time, it means that agent i falls into a local minimum and the obstacle bypassing mode is activated. Set the obstacle bypassing mode flag to M. When agent i falls into a local minimum, the obstacle bypassing mode is activated. At this time, M=0. When agent i does not detect an obstacle or is not blocked by an obstacle after sensing an obstacle, it changes to the target tracking mode. At this time, M=1;

[0089] Step 4.2: Determine the location of the virtual target point. If agent i falls into a local minimum and starts the obstacle bypassing mode, a virtual target point is needed to guide agent i to bypass. The location of the virtual target point is determined by the local A-star algorithm. The A-star algorithm is a classic path planning algorithm. During the search process of the A-star algorithm, the cost value of each node is determined by the cost function f(n), which is defined as follows:

[0090] f(n)=g(n)+h(n)

[0091] Wherein, n represents a grid node in the path; f(n) represents the total cost estimate from the starting point through node n to the target node; g(n) represents the actual path cost from the starting point to node n; h(n) represents the estimated cost from node n to the target node. The estimation function h(n) of the A-star algorithm can be calculated by different distance measurement methods, such as Manhattan distance, Chebyshev distance or Euclidean distance. In the present invention, Euclidean distance is selected as the estimated cost to calculate h(n). Since the A-star algorithm selects the node with the lowest current cost for expansion at each step, and guides the search direction by the estimation function h(n), the resulting path is not only the lowest cost, but also the shortest. However, in a complex environment, the A-star algorithm will cause a sudden increase in the amount of calculation as the agent i moves in real time to plan the global route, and the agent i will not recognize the global information. Therefore, the present invention proposes a novel local A-star algorithm for determining the virtual target point required for the agent to bypass obstacles.

[0092] The local A-star algorithm searches for paths within the sensing range of agent i, rasterizes the environment within the sensing radius of agent i, selects the node with the smallest f value as the current node, and determines whether it is within the sensing range. If the current node is within the sensing range of agent i and the next node will exceed the boundary of the sensing range of agent i, the current node is used as a virtual target point and the path search is stopped. Figure 2 The figure shows the process of the agent determining the virtual target point using the A-star algorithm, where the purple dots represent the candidate nodes generated by the local A-star algorithm during path search, and the red dots represent the virtual target point. This ensures that the agent i can quickly find the virtual target point after falling into a local minimum and guide it without the need for global path planning.

[0093] It is particularly noteworthy that when agent i uses the A-star algorithm to determine the virtual target point during the obstacle avoidance process, agent i adjusts the node search direction of the A-star algorithm according to the obstacle avoidance direction. Specifically, when agent i avoids obstacles clockwise, the node search direction of the A-star algorithm is limited to "up, upper left, left" and "lower left"; when agent i avoids obstacles counterclockwise, the node search direction is limited to "upper right, right, lower right" and "lower". This restriction on the search direction can ensure that agent i can find the virtual target point more effectively during the obstacle avoidance process and avoid the left and right oscillation of the virtual target point;

[0094] Step 4.3: Design navigation feedback items of target point and virtual target point to agent i for

[0095]

[0096] Where M is the obstacle mark of agent i. and is the gain parameter of the control input of agent i, q γ is the position vector of the target point, p γ is the velocity vector of the target point, is the position vector of the virtual target point corresponding to agent i, is the velocity vector of the virtual target point,

[0097]

[0098] Step 5 specifically includes the following steps:

[0099] Step 5.1: Determine the collision distance, plan the sensing area of ​​agent i and divide it into a safe area, a warning area, a dangerous area and a collision area. Figure 3 As shown;

[0100] Step 5.2: Set the fuzzy controller according to the sensing area of ​​agent i, calculate the small distances between agents, between agent i and the target point, and between agent i and its nearest obstacle boundary point k, and adjust the gain parameter in the interaction term through the fuzzy controller size.

[0101] The execution process of step 6 includes the following steps:

[0102] Step 6.1: Design the control input term u by integrating the interaction terms between agents i constructed in steps 2 to 5, the interaction terms between agent i and its nearest obstacle boundary point k, and the navigation feedback terms of the target point and the virtual target point to agent i. i for

[0103]

[0104] Step 6.2, jump to step 1 to implement multi-agent swarming obstacle avoidance.

[0105] Furthermore, the present invention is verified by specific experiments. In this embodiment, considering N=20, 40, 60, 80 agents moving in a two-dimensional Euclidean space in a complex obstacle environment including diamond, ellipse and C-shaped obstacles, the initial position vectors and velocity vectors of N=20, 40, 60, 80 agents are randomly generated in the vectors [5, 30]×[5, 25] and [0, 2]×[0, 2], respectively. The position vector and velocity vector of the target point are (125, 130) respectively. T and (0,0) T , the maximum speed of agent i pmax = 6m / s, agent i runs in the proposed algorithm for 300s, and the experimental results are as follows Figure 5 shown.

[0106] Figure 4 The black object in the middle is an obstacle, the green triangle is the target point, and the purple triangle is the agent. The agent's initial position vector is randomly selected from [5,30]×[5,25], and the initial velocity vector is randomly selected from [0,2]×[0,2]. It runs for 300 seconds in an environment containing complex obstacles such as diamonds, ellipses, and C-shaped obstacles. Figure 5 It shows that the agents can bypass complex obstacles and reach the target point, and reconverge after reaching the target point to achieve multi-agent consistency. Figure 5 It can be seen that the minimum distance between agents and between agents and obstacle boundary points is always greater than the safety distance d c=1m. These results show that the proposed algorithm not only overcomes the limitation of the original algorithm on the obstacle shape, but also dynamically adjusts various potential energy gain parameters through fuzzy adaptive parameters, thereby improving the adaptability of the intelligent agent in complex environments.

[0107] In summary, compared with the prior art, the advantages and technical effects of the present invention are as follows:

[0108] (1) In view of the fact that the original model is restricted by the shape of obstacles when large-scale or ultra-large-scale intelligent agent clusters avoid obstacles, the present invention rewrites the interaction term between intelligent agent i and obstacles, gets rid of the shape restriction of obstacles, increases the size of the potential energy function, and prevents collisions between intelligent agents and between intelligent agents and obstacle boundary points.

[0109] (2) The original A-star algorithm may occupy a large amount of memory in large-scale graphs, and the agents cannot recognize global information. The present invention redesigns and rewrites the original A-star algorithm so that the agents in the cluster can quickly generate virtual target points at the boundary of the agent's sensing range after falling into a local minimum.

[0110] (3) In order to solve the problem that the original model is prone to falling into a local minimum and unable to reach the target point when encountering complex obstacles, the present invention redesigns and rewrites the navigation feedback item, and divides the target point into two types: real target points and virtual target points, so that the intelligent agents in the cluster can quickly escape after falling into a local minimum when avoiding obstacles, thereby improving the obstacle avoidance efficiency.

[0111] (4) In order to address the problem that the original model causes the distance between intelligent agents to decrease and collisions to occur easily when avoiding obstacles in high-density or high-speed movement, the present invention further effectively prevents the occurrence of collisions by introducing a fuzzy controller to dynamically adjust the gain parameters in the interaction term.

[0112] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A multi-agent swarming obstacle avoidance method integrating A-star, characterized in that: The following steps are involved: Step 1: Construct a multi-agent motion model with speed constraints and calculate the position and speed information of each agent; Step 2: Construct the interaction terms between agents based on the information of other agents within the perception radius of agent i; Step 3: Based on the information of obstacles within the perception radius of agent i, construct the interaction term between agent i and its nearest obstacle boundary point k; Step 4: According to the movement of agent i, construct a virtual target point and the navigation feedback item of the target point to agent i; Step 5: Determine the collision distance, plan the sensing area of ​​agent i and divide it into a safe area, a warning area, a dangerous area, and a collision area, calculate the distances between agents, between agent i and the target point, and between agent i and its nearest obstacle boundary point k, and adjust the size of the gain parameter in the interaction term through the fuzzy controller; Step 6: Determine the control input u of agent i by integrating the interaction terms between agents, the interaction terms between agent i and its nearest obstacle boundary point k, and the navigation feedback terms of the target point and the virtual target point to agent i i , and jump to step 1 to achieve swarming obstacle avoidance of multi-agent clusters.

2. The multi-agent swarming obstacle avoidance method of integrating A-star as claimed in claim 1, characterized in that: The expression of the intelligent agent motion model is: Among them, q i (t) represents the position vector of agent i at time t, p i (t) represents the velocity vector of agent i at time t, pmax represents the maximum velocity vector of agent i, u i (t) represents the control input of agent i at time t, represents the interaction term between agents at time t, represents the interaction term between agent i and its nearest obstacle boundary point k at time t, represents the navigation feedback item of the target point and the virtual target point to the agent i at time t, N represents the total number of agents, · represents the Euclidean norm, δ i Represents the weight coefficient. If the speed of agent i is greater than pmax, then δ i =1, otherwise δ i =0.

3. The multi-agent swarming obstacle avoidance method of integrating A-star as claimed in claim 2, characterized in that: The intelligent agent i is equipped with a laser radar for identifying other intelligent agents or obstacles and determining the set of intelligent agents and obstacles within the perception radius of the intelligent agent i, where r is the perception radius of the laser radar.

4. The multi-agent swarming obstacle avoidance method of integrating A-star as claimed in claim 3, characterized in that: In step 2, the interaction terms between agents are used to prevent collisions between agents. Contains the position gradient term and the velocity consistency term, and the expression is in, is the set of all agents perceived by agent i within the perception radius, is the gain parameter of the control input of agent i, is the attraction / repulsion potential energy pair ψ between agent i and agent j α (||p j (t)-p i (t)|| σ ) at position p i The gradient at (t); defines the attraction / repulsion potential ψ between agents α (z) is where φ α (z)=φ(zd α )ρ h (z / r α ), r α =||r|| σ , d α =||d|| σ , a α , b α and h are positive system parameters, h∈(0,1), b α ≥a α >0, ε is the system parameter, ε>0, a ij (q) is the adjacency matrix of agent j, which is used to represent the connection relationship and weight between agent i and agent j, a ij (q) = ρ h (||q j -q i || σ / r α )∈[0,1],j≠i, impact function 5. The multi-agent swarming obstacle avoidance method of integrating A-star as claimed in claim 4, characterized in that: In step 3, the interaction term between agent i and its nearest obstacle boundary point k is used Prevent collisions between agent i and obstacle boundary point k, interaction term Contains the position gradient term and the velocity consistency term, and the expression is In the formula, is the set of all obstacle boundary points perceived by agent i within the perception radius, and is the gain parameter of the control input of agent i, The obstacle boundary point k closest to agent i is called the virtual agent, and its position vector is given by Express, satisfy The velocity vector is given by express; is the repulsive potential energy function between agent i and its nearest obstacle boundary point k At position p i The gradient at (t); Define the repulsive potential energy function ψ between agent i and the nearest obstacle boundary point k β (z) is d β =||d′|| σ , a β , b β and h are positive system parameters, h∈(0,1), b β ≥a β >0, ε is the system parameter, ε>0, is the rotation matrix, b i,k (q) is the adjacency matrix of the agent, which is used to represent the connection relationship and weight between the agent and the obstacle boundary point k. Impact function 6. The multi-agent swarming obstacle avoidance method of integrating A-star as claimed in claim 5, characterized in that: The execution process of step 4 includes the following steps: Step 4.1: Combine the local minimum to determine the motion state of agent i, which are target tracking mode and obstacle bypassing mode. When agent i encounters a concave obstacle or a complex obstacle and falls into a local minimum, the obstacle bypassing mode is turned on; if agent i does not detect an obstacle or is not blocked by the obstacle after sensing the obstacle, it is in target tracking mode; The judgment formula for agent i falling into a local minimum is: The above means that the displacement of agent i within a specific time range is less than a certain threshold λ, and the speed at time t is less than a certain threshold κ. When each of the above conditions is met at the same time, it means that agent i falls into a local minimum; Step 4.2: If agent i falls into a local minimum and starts the obstacle bypassing mode, a virtual target point is needed to guide agent i to bypass. The location of the virtual target point is determined by the local A-star algorithm. Step 4.3: Design the target point and virtual target point pair agent Navigation feedback items for Where M is the obstacle mark of agent i. and is the gain parameter of the control input of agent i, q γ is the position vector of the target point, p γ is the velocity vector of the target point, is the position vector of the virtual target point corresponding to agent i, is the velocity vector of the virtual target point, 7. The multi-agent swarming obstacle avoidance method of integrating A-star as claimed in claim 6, characterized in that: Control input u of agent i i for

Images