Multi-agent cluster obstacle avoidance method and device, electronic equipment and storage medium

Abstract

The invention provides a multi-agent cluster obstacle avoidance method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining initial kinetic parameters of a plurality of agents contained in an agent cluster, the initial kinetic parameters comprising displacement, speed and initial control input; according to the initial kinetic parameters, a plurality of second intelligent agents in a preset range of a first intelligent agent are determined, and the first intelligent agent is any one of the intelligent agents; acquiring a motion parameter of the first intelligent agent and an obstacle position obtained by observation; determining a first control input of the first intelligent agent based on the motion parameter, the obstacle position and the initial kinetic parameter; and performing obstacle avoidance simulation verification on the first intelligent agent according to the first control input. According to the invention, the practicability of a multi-agent cluster obstacle avoidance algorithm can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of cluster control, and specifically to a multi-agent cluster obstacle avoidance method, device, electronic device and storage medium. Background Art

[0002] Unmanned system clusters are collaborative systems composed of multiple autonomous or semi-autonomous agents, covering various forms of unmanned equipment such as drones, surface unmanned ships, underwater unmanned vehicles, unmanned vehicles, and mobile robots. These agents exchange information and work together through the network to complete complex tasks together. This type of system is also called a "multi-agent cluster". Its notable feature is that each agent in the cluster has a certain independent decision-making ability, and can collaborate, communicate, and make decisions in a distributed manner in the group, thereby optimizing the overall goal and improving efficiency.

[0003] Obstacle avoidance is one of the core challenges in multi-agent swarm motion control, especially in complex environments. In 1986, Reynolds proposed three basic principles for multi-agent swarm motion: aggregation, collision avoidance, and speed matching. The aggregation principle requires each agent to get as close to neighboring agents as possible to maintain the cohesion of the group; the collision avoidance principle ensures that the agents maintain a sufficient distance to avoid collisions; and the speed matching principle emphasizes the consistency of speed between neighboring agents to maintain the coordination of the cluster. When the cluster encounters an obstacle during movement, in addition to following the obstacle avoidance strategy to avoid collisions with the obstacle, it must also continue to meet these three basic principles in the overall obstacle avoidance process.

[0004] Traditional flocking algorithms are widely used due to their robustness in cluster control, especially in ensuring the aggregation of cluster members, non-collision and speed consistency. However, faced with the need for obstacle avoidance in dynamic and complex environments, flocking algorithms have shown certain limitations. In order to achieve obstacle avoidance, it is usually necessary to combine the artificial potential field method, which relies on the pre-modeling of the global environment and guides the agent to avoid obstacles by calculating the virtual force field between the agent and the obstacle. The artificial potential field method is very sensitive to environmental changes. When the environment changes, the algorithm and model often need to be redesigned, resulting in poor adaptability and difficulty in coping with the challenges of dynamic or unknown environments. Therefore, a method is needed to improve the practicality of multi-agent cluster obstacle avoidance algorithms. Summary of the invention

[0005] The present application provides a multi-agent cluster obstacle avoidance method, device, electronic device and storage medium, which can improve the practicality of the multi-agent cluster obstacle avoidance algorithm.

[0006] In a first aspect of the present application, a multi-agent cluster obstacle avoidance method is provided, the method comprising:

[0007] Acquire initial dynamic parameters of a plurality of agents in an agent cluster, wherein the initial dynamic parameters include displacement, velocity, and initial control input;

[0008] Determine, according to the initial dynamic parameters, a plurality of second agents within a preset range of the first agent, wherein the first agent is any one of the plurality of agents;

[0009] Obtaining motion parameters of the first agent and an observed obstacle position;

[0010] Determining a first control input of the first agent based on the motion parameters, the obstacle position, and the initial dynamic parameters;

[0011] According to the first control input, an obstacle avoidance simulation verification is performed on the first agent.

[0012] Optionally, determining the first control input of the first agent based on the motion parameter, the obstacle position and the initial dynamics parameter specifically includes:

[0013] According to the first speed of the first agent, the first displacement of the first agent, the second speed of the second agent, and the second displacement of the second agent, a collision avoidance control input of the first agent is determined, and the collision avoidance control input is expressed as follows:

[0014]

[0015] Among them, f i g is the collision avoidance control input, q i is the first displacement of the first agent, p i is the first speed of the first agent, q j is the second displacement of the jth second agent, p j is the second speed of the jth second agent, c1 and c2 are the control parameters of the first agent, n ij is the displacement vector of the first agent’s displacement pointing to the jth second agent’s displacement, a ij is the interaction strength coefficient, Φ α is the distance function;

[0016] Obtaining the target point position and target point speed of the intelligent agent cluster;

[0017] According to the target point position and the target point speed, an initial control input is determined, and the initial control input is expressed as follows:

[0018] f in =-e1(q i -q target )-e2(p i -p target )

[0019] Among them, f i n is the initial control input, q i is the first displacement, p i is the first speed, q target is the target point position, p target is the target point speed, e1 and e2 are control parameters;

[0020] Based on the collision avoidance control input and the initial control input, the first control input is determined, and the first control input is expressed as follows:

[0021]

[0022] Among them, u i is the first control input, f i g To avoid collisions between multiple agents, f i n is the initial control input, f i β To control the obstacle avoidance input.

[0023] Optionally, the determining the first control input based on the obstacle avoidance control input specifically further includes:

[0024] Based on the first displacement and the first velocity, an observation space is determined, and the observation space is expressed as follows:

[0025]

[0026] Among them, s i is the observation space, are the first displacement and the first speed, is the observation of the second agent by the first agent, S i obstacle is the observation of the obstacle by the first intelligent agent;

[0027] Based on the observation space, the obstacle avoidance control input is determined, and the obstacle avoidance control input is expressed as follows:

[0028]

[0029] Among them, f iβ is the obstacle avoidance control input, Δp i is the speed change.

[0030] Optionally, determining the collision avoidance control input of the first agent according to the first speed of the first agent, the first displacement of the first agent, the second speed of the second agent, and the second displacement of the second agent specifically further includes:

[0031] The displacement vector is determined according to the first displacement and the second displacement, and the displacement vector is specifically expressed as follows:

[0032]

[0033] Among them, n ij is the displacement vector, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent;

[0034] Based on the first displacement and the second displacement, the interaction strength coefficient is determined, which is specifically calculated by the following formula:

[0035]

[0036] Among them, a ij is the interaction strength coefficient, ‖q j -q i ‖ σ It means (q j -q i )’s σ norm, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent, and the σ norm is defined as follows:

[0037]

[0038] Among them, the constant ε>0, ‖x‖ σ represents the σ norm of vector x, ‖x‖ represents the length of the vector, and x is a vector of any dimension.

[0039] Optionally, after determining the first control input of the first agent based on the motion parameter, the obstacle position and the initial dynamics parameter, the method further includes:

[0040] Obtaining a plurality of obstacles observed by the first intelligent agent and a plurality of the second intelligent agents;

[0041] Determining the distance between each of the obstacles and the first agent;

[0042] Determine a target obstacle among a plurality of obstacles, wherein the distance between the target obstacle and the first intelligent agent is shortest;

[0043] Based on the distance between the target obstacle and the first agent, a reward function is established, and the reward function is specifically as follows:

[0044]

[0045] Among them, r i is the reward function, d i is the distance between the first agent and the target obstacle, d1 is the critical safety distance, μ is the repulsion coefficient, and d0 is the reward adjustment scale.

[0046] Optionally, the distance function is determined based on the distance between the first agent and the second agent, which is specifically expressed by the following formula:

[0047]

[0048] Among them, Φ α (z) is the distance function with distance z as the independent variable, z = || q j -q i ||, r a is the normalized scale, d a is a constant offset.

[0049] Optionally, performing obstacle avoidance simulation verification on the first agent according to the first control input specifically includes:

[0050] Constructing a simulation environment, wherein the simulation environment includes at least two virtual obstacles and a virtual target point;

[0051] constructing a virtual agent model based on the first control input;

[0052] Allowing the virtual agent to interact in the simulation environment;

[0053] Determining whether the virtual agent can avoid the virtual obstacle and reach the virtual target point;

[0054] If it is determined that the virtual agent avoids the virtual obstacle and reaches the virtual target point, the first control input is determined to be an obstacle avoidance strategy.

[0055] In a second aspect of the present application, a multi-agent cluster obstacle avoidance device is provided, the device comprising an acquisition module, a judgment module, a processing module and a simulation module, wherein:

[0056] The acquisition module is used to acquire initial dynamic parameters of a plurality of agents in an agent cluster, wherein the initial dynamic parameters include displacement, velocity and initial control input;

[0057] The judgment module is used to determine a plurality of second agents within a preset range of the first agent according to the initial dynamic parameters, wherein the first agent is any one of the plurality of agents;

[0058] The acquisition module is used to acquire the motion parameters of the first agent and the observed obstacle position;

[0059] The processing module is used to determine a first control input of the first agent based on the motion parameter, the obstacle position and the initial dynamics parameter;

[0060] The simulation module is used to perform obstacle avoidance simulation verification on the first intelligent agent according to the first control input.

[0061] In the third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes any one of the methods described above.

[0062] In a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions, and when the instructions are executed, any of the methods described above is executed.

[0063] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0064] 1. This application can effectively cope with complex and dynamic environments by dynamically acquiring and analyzing the initial dynamic parameters of the intelligent agent, the positions of other intelligent agents and obstacles within a preset range, and combining the motion parameters of the intelligent agent to dynamically determine the control input of the intelligent agent. By verifying the obstacle avoidance effect of the control input in a simulation environment, static presets and dependence on the global environment are avoided, greatly improving the adaptability and flexibility of the algorithm. In addition, by adaptively adjusting the behavior of the intelligent agent, the dependence on the environmental model is reduced, thereby improving the obstacle avoidance performance and practicality of the multi-agent cluster in uncertain and dynamic environments.

[0065] 2. The obstacle avoidance method proposed in the present invention is applicable to situations where the environment is not completely known, avoiding the need to model the global environment. By having the agent observe in the local environment around itself and limiting the observation space to a smaller range, this method facilitates rapid learning and training, and improves efficiency. At the same time, this method uses the speed change of the agent as the action space for reinforcement learning, which meets actual control requirements. In the process of interaction between the agent and the local environment, the reinforcement learning strategy enables the agent to gradually explore and effectively avoid obstacles, thereby improving the applicability and scalability of the system.

[0066] 3. This application improves the obstacle avoidance capability of multi-agent clusters in unknown and complex environments by dynamically generating control inputs based on the local observation space, motion parameters, and obstacle positions of the agent. The complexity and uncertainty of global environment modeling are avoided, allowing each agent to make real-time decisions based on local information. Combining collision avoidance control, initial control input, and control obstacle avoidance input based on observation space, this method effectively improves the reaction speed and adaptability of the agent in a dynamic environment. At the same time, through the design of reward functions and distance functions, the obstacle avoidance behavior of the agent is further optimized to ensure stability and safety in complex environments, and has strong practicality and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flowchart of a multi-agent cluster obstacle avoidance method disclosed in an embodiment of the present application;

[0068] Figure 2 It is a schematic diagram of the initial process of simulating a multi-agent cluster obstacle avoidance method disclosed in an embodiment of the present application;

[0069] Figure 3 It is a schematic diagram of a simulation obstacle avoidance process of a multi-agent cluster obstacle avoidance method disclosed in an embodiment of the present application;

[0070] Figure 4 It is a schematic diagram of a simulation obstacle avoidance result of a multi-agent cluster obstacle avoidance method disclosed in an embodiment of the present application;

[0071] Figure 5 It is a module schematic diagram of a multi-agent cluster obstacle avoidance device disclosed in an embodiment of the present application;

[0072] Figure 6 It is a structural schematic diagram of an electronic device disclosed in an embodiment of the present application.

[0073] Explanation of the reference numerals: 501, acquisition module; 502, judgment module; 503, processing module; 504, simulation module; 601, processor; 602, communication bus; 603, user interface; 604, network interface; 605, memory. DETAILED DESCRIPTION

[0074] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0075] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0076] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0077] Unmanned system swarms are composed of multiple autonomous or semi-autonomous agents that complete complex tasks through network collaboration and have independent decision-making capabilities. In multi-agent swarm control, obstacle avoidance is one of the core challenges. The three principles proposed by Reynolds - aggregation, collision avoidance and speed matching - provide the basis for swarm motion control. However, although the traditional flocking algorithm is robust in maintaining cluster aggregation and speed consistency, it has limitations in obstacle avoidance in dynamic and complex environments. It usually relies on artificial potential field methods for global environment modeling and has poor adaptability. Therefore, it is necessary to combine new methods such as reinforcement learning to enhance the flexibility and practicality of swarm obstacle avoidance algorithms.

[0078] This embodiment discloses a multi-agent cluster obstacle avoidance method, referring to Figure 1 , including the following steps:

[0079] S110, obtaining initial dynamic parameters of an agent cluster including multiple agents.

[0080] The dynamic modeling of multi-agent clusters is the basis of the entire system, which describes the motion state of each agent and its changing rules. Assume a cluster composed of N agents, each of which has its own motion state (displacement, speed) and control input. Agents can interact with each other through perception or communication, and each agent has a certain perception of the surrounding environment (including other agents and obstacles).

[0081] For the i-th agent, the following two dynamic equations are used to describe its motion state:

[0082]

[0083] Among them, q i represents displacement, p i Indicates speed, u i represents the control input of the ith agent.

[0084] The displacement q of the agent i is its speed p i In discrete time, this equation can be expressed as:

[0085] q i (t+1)=q i (t)+p i (t)·Δt

[0086] Where Δt is the time step. Similarly, the speed of the agent p i The control input u i Similarly, in discrete time, it can be expressed as:

[0087] p i (t+1)=p i (t)+u i (t)·Δt

[0088] The state of each agent is composed of its displacement qi and velocity pi, so the state of the i-th agent can be expressed as x i =[q i , p i ] T The state space of the entire multi-agent system is a high-dimensional space composed of the states of all agents. The state vector of the system can be expressed as X = [x1, x2, ..., x N ] T .

[0089] The control input of each agent is determined by many factors, including the status of other agents, target location, obstacle avoidance requirements, etc. The control input design can use traditional control methods (such as PID control) or combine artificial intelligence methods (such as reinforcement learning) to design more intelligent control laws. In multi-agent systems, a distributed control strategy is usually adopted, that is, each agent independently calculates its own control input without relying on global information. This strategy can not only improve the robustness of the system, but also better adapt to dynamically changing environments.

[0090] S120, determining a plurality of second agents within a preset range of the first agent according to the initial dynamic parameters.

[0091] In the initial state of the multi-agent cluster, the dynamic parameters (including displacement, velocity, control input, etc.) of each agent need to be obtained through perception or communication. For example, for any of the agents, the first agent, the initial dynamic parameters include its position, velocity, and control input.

[0092] Each intelligent agent obtains its current position through GPS or other positioning systems, obtains its current speed through an inertial measurement unit (IMU) or other speed measurement devices, shares information with other intelligent agents through wireless communication networks, and obtains status data of adjacent intelligent agents.

[0093] The agent determines multiple neighboring agents within a preset range through the observation space. The observation space is a collection of environmental information that the agent can perceive or obtain. This space usually includes its own state information and observation information of other agents and obstacles. Here, the observation space mainly includes the following three parts: its own state information, including the displacement and speed of the agent. Observation information of other agents, including the position and speed of neighboring agents. Observation information of obstacles, including the position of obstacles, etc.

[0094] In the embodiment of the present application, it is assumed that the observation range of each agent is large enough, but in order to simplify the processing, each agent only observes its three neighboring agents. The process of selecting neighboring agents is as follows:

[0095] For the first agent, we first need to calculate the distance between it and other agents. Let the position of the jth agent be q j , then the Euclidean distance d between the first agent and the jth agent ij for:

[0096] d ij =||q i -q j ||

[0097] Based on the calculated distance dij, multiple agents with the closest distance to the i-th agent are selected to form the neighboring agent set of the agent. All agents are sorted from small to large distances to obtain a list of nearest neighbor agents. Multiple nearest agents are selected from the sorted results to obtain multiple second agents, which are agents within the preset range of the first agent among the multiple agents.

[0098] Although it is assumed that the observation range of the agent is large enough, in actual applications, due to the limitations of sensors and communications, the observation range of the agent is limited. If the number of neighboring agents of an agent exceeds a predetermined limit (such as 3), the 3 closest agents can be selected first. If the number of neighboring agents is less than 3, only the actually observed agents are selected.

[0099] S130, obtaining motion parameters of the first agent and the observed obstacle position.

[0100] To avoid collisions, the agent needs to sense the location of obstacles in its surroundings. Obstacle observation can be achieved in the following ways: The agent's LiDAR generates distance and direction information of obstacles by emitting laser beams and receiving reflected signals, which is usually used to generate high-precision 2D or 3D environment maps. This is a widely used obstacle detection sensor. Alternatively, the agent's ultrasonic sensor measures the distance between the obstacle and the agent by emitting and receiving ultrasonic signals, which is often used in low-cost, low-complexity obstacle avoidance systems.

[0101] After observing the obstacle, the position of the obstacle needs to be estimated. Common obstacle position estimation methods include: dividing the environment into grids, each grid indicating whether a certain area is occupied by an obstacle. The agent updates the grid state based on sensor data to estimate the position of the obstacle. Use vision-based object detection algorithms (such as YOLO, SSD) to identify obstacles in the image, and use methods such as Kalman filtering to track the movement trajectory of the obstacle to continuously estimate its position. In a dynamic environment, the position of obstacles may change over time. The agent needs to continuously observe and update the obstacle position information in real time. This can be achieved by periodically scanning the environment and updating the obstacle map. For dynamic obstacles (such as moving pedestrians or vehicles), the agent also needs to predict their future positions to plan safe obstacle avoidance paths.

[0102] S140, determining a first control input of the first agent based on the motion parameters, the obstacle position and the initial dynamic parameters.

[0103] The control input u of the first agent i It consists of three parts: obstacle avoidance control input f i β, collision avoidance control input f i g and the initial control input f i n , the formula is as follows:

[0104]

[0105] Among them, u i is the first control input, f i g Collision avoidance control input to avoid collisions between multiple agents, used to avoid collisions between agents; f i n is the initial control input, used to make the agent move to the target point; f i β To control obstacle avoidance input, used to avoid collision with obstacles.

[0106] According to the first speed of the first agent, the first displacement of the first agent, the second speed of the second agent, and the second displacement of the second agent, the collision avoidance control input of the first agent is designed. The specific formula is as follows:

[0107]

[0108] Among them, f i g is the collision avoidance control input, q i is the first displacement of the first agent, p i is the first speed of the first agent, q j is the second displacement of the jth second agent, p j is the second speed of the jth second agent, c1 and c2 are the control parameters of the first agent, Φ α is the distance function, n ij is the displacement vector of the first agent pointing to the jth second agent, a ij is the interaction strength coefficient.

[0109] This formula is used to avoid collisions between multiple agents, mainly considering the relative position and speed differences. The first term is calculated by the position difference q of the adjacent agents. j -q i Calculate the displacement vector n ij , and then through the distance function Φ α Measures the influence of neighboring agents on the current agent. This part actually simulates a kind of "repulsive force". When the distance between agents is too close, the repulsive force increases, forcing them to move away from each other and avoid collision. The second term is calculated by the speed difference p of the neighboring agents. j -p iTo adjust the speed inconsistency between them. a ij is an interaction intensity coefficient to ensure that the speed difference does not cause collisions. This is a speed adjustment term that avoids collisions by guiding the speed of the agent closer to the speed of its neighboring agents.

[0110] In the formula, Φ α (||q j -q i || σ ) is a distance function used to measure the influence of the distance between agents, usually represented by a Gaussian function or other forms. Φ α (z) is a distance function with the distance z as the independent variable:

[0111]

[0112] The distance function is used to measure the mutual force between two agents. It is calculated by the product of a window function ρ h and another potential function Φ(z - d α ). It remains 1 when the distance is close, that is, the full force acts; when the distance is moderate, it gradually weakens as the distance increases; when the distance is too far, the force is 0. This design ensures that agents only have a significant impact on nearby neighbors. The potential function Φ(z - d α ) represents the intensity of the force, z represents the distance between two agents, and the constant offset d α is a threshold that controls the zero point of the force. By adjusting d α , the minimum allowable distance between agents can be defined.

[0113] Among them, the normalization scale r a and the constant offset d a are calculated as follows respectively:

[0114]

[0115] Among them, the potential function Φ(z) is expressed as follows:

[0116]

[0117] Among them, σ1(z) is an S-shaped function used to smooth the influence of distance on the force, and the constants a and b satisfy 0 < a ≤ b, and c satisfies the following conditions:

[0118]

[0119] This entire function construction ensures that when the distance is small, the interaction between agents is strong, and as the distance increases, the force gradually weakens until it approaches zero.

[0120] Window function ρ h (z) is defined as:

[0121]

[0122] This is a window function with smooth transition, which is used to adjust the range and intensity of interaction between agents. When the distance z is small, that is, when z is in the interval [0,h], the interaction intensity remains at the maximum (i.e. 1). When z is in the transition interval [h,1], the interaction intensity gradually decreases and smoothly transitions to zero. When z>1, the interaction intensity directly becomes zero, indicating that the distance between the agents is too large and they no longer interact with each other. This function simulates the smooth attenuation of the interaction force with distance in reality, ensuring that when the agents are far apart, the interaction gradually disappears, avoiding abrupt changes.

[0123] In a possible implementation manner, a displacement vector is determined according to the first displacement and the second displacement, and the displacement vector is specifically expressed as follows:

[0124]

[0125] Among them, n ij is the displacement vector, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent.

[0126] n ij Represents the vector pointing from the first agent position to the jth second agent position, reflecting the directional relationship between the two agents. j -q i is the displacement difference between agents i and j, representing the direction vector from i to j. The denominator is a normalization factor that ensures that the vector takes into account the influence of the distance between agents when calculating, so that when the distance increases, the length of the direction vector decreases accordingly. This formula can be understood as introducing nonlinear scaling on the basis of traditional vector normalization, adjusting the influence intensity of the vector, especially when the distance is very small, to avoid infinite increase.

[0127] Based on the first displacement and the second displacement, the interaction strength coefficient is determined, which is specifically calculated by the following formula:

[0128]

[0129] Among them, a ij is the interaction strength coefficient, which is used to represent the interaction strength between the first agent i and the jth second agent. The value varies in the interval [0,1]. j -q i ‖ σ It means (q j -qi )'s σ norm q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent, and the σ norm is defined as follows:

[0130]

[0131] Among them, the constant ε>0, ‖x‖ σ represents the σ norm of vector x, ‖x‖ represents the vector length, and x is a vector of any dimension. ||x|| σ It is a norm of nonlinear transformation, which is suitable for scenarios where the small-scale effect of large-distance transformation needs to be considered. This norm definition introduces a small parameter ∈ to nonlinearly scale the standard Euclidean distance ||x||. When ∈ is very small, ||x|| σ Approximately equal to ||x||; when ∈ is large, the norm becomes more sensitive to large distances. When dealing with a wide range of distances, this nonlinear scaling can more accurately describe the impact of long distances without increasing the computational complexity due to excessive distances.

[0132] In one possible implementation, in a multi-agent system, the motion control of the agent needs to be dynamically adjusted based on its perception of the surrounding environment. In this process, the observation space is also the core basis for the agent's decision-making. The observation space Si includes the following three parts: the state information of the agent itself, including its current position (first displacement) and current speed (first speed). This part of information describes the basic state of the agent in the environment. The agent's observation information of other surrounding agents. This part of information may include the relative position and speed with neighboring agents, which is used to judge possible collision risks or coordinated movements. The agent's observation information of other surrounding agents. This part of information may include the relative position and speed with neighboring agents, which is used to judge possible collision risks or coordinated movements. By combining the above information, the agent can fully perceive the spatial layout of itself, other agents and obstacles, and thus make reasonable obstacle avoidance decisions. Based on the first displacement and the first speed, the observation space is determined, and the observation space is expressed as follows:

[0133]

[0134] Among them, s i is the observation space, s i self is the first displacement and the first velocity, s i other is the observation of the first agent on the second agent, S i obstacle is the first agent’s observation of the obstacle.

[0135] After obtaining the observation space Si, the first agent determines the change in its speed according to the following logic. If an obstacle is detected, the agent calculates a new motion strategy based on the position and speed of the obstacle. This usually involves adjusting the direction or magnitude of the speed to avoid collision with the obstacle. Specifically, it includes calculating the required avoidance direction based on the distance and direction of the obstacle. At the same time, the speed is adjusted to ensure that the agent does not collide with the obstacle during the obstacle avoidance process, while trying to maintain the target speed. If no obstacle is detected, the agent maintains the original speed and continues to perform the given task without adjusting the speed.

[0136] The obstacle avoidance control input is adjusted based on the agent's observation of obstacles. The specific rules are as follows: If an obstacle is observed, the obstacle avoidance control input will be equal to the speed change Δp i The change is the result of the agent's adjustment based on the obstacle information, ensuring that the agent can safely avoid obstacles. The formula is:

[0137]

[0138] If no obstacle is observed: When the agent does not detect an obstacle, the obstacle avoidance control input will be zero, that is, no additional speed adjustment will be made. At this time, the agent only needs to continue to execute the original motion plan. The formula is expressed as:

[0139]

[0140] In a possible implementation, after determining the first control input of the first agent based on the motion parameters, the obstacle position and the initial dynamic parameters, the method further includes: obtaining multiple obstacles observed by the first agent and multiple second agents; determining the distance between each obstacle and the first agent; determining a target obstacle among the multiple obstacles, the target obstacle being closest to the first agent; and establishing a reward function based on the distance between the target obstacle and the first agent, the reward function being specifically as follows:

[0141]

[0142] Among them, r i is the reward function, d i is the distance between the first agent and the target obstacle, d1 is the critical safety distance, μ is the repulsion coefficient, and d0 is the reward adjustment scale.

[0143] Specifically, first, it is necessary to obtain the observation results of multiple agents on obstacles in the environment. Obstacle information is obtained for each agent (including the first agent and multiple second agents). Each agent will form an obstacle list, recording the location information and distance of multiple obstacles. For each observed obstacle, the Euclidean distance between them and the first agent is calculated. Through this process, a distance list between multiple obstacles and the first agent can be obtained.

[0144] From the list of obstacles, select the obstacle closest to the first agent, called the target obstacle. The target obstacle is the obstacle with the smallest distance to the agent and is also the obstacle most likely to affect the agent's obstacle avoidance behavior. By comparing the distances of all obstacles, select the target obstacle with the smallest distance.

[0145] In the multi-agent obstacle avoidance task, the reward function is one of the core parts of designing agent behavior in reinforcement learning. It guides the agent to learn in the direction of the goal and continuously adjusts the strategy to achieve better performance. For the obstacle avoidance problem, the reward function needs to pay special attention to the safe distance between the agent and the obstacle to avoid collision. The reward function determines the reward value based on the distance between the agent and the obstacle, and is designed with the following two cases: If the distance between the agent and the obstacle is less than or equal to the critical safety distance d1, the agent will receive a negative reward. The size of the negative reward is proportional to the inverse of the distance 1 / d i This means that the closer the distance, the more negative the reward value, that is, the greater the penalty for the agent, which incentivizes the agent to stay away from obstacles as much as possible. The repulsion coefficient determines the agent's sensitivity to obstacle avoidance. The larger the repulsion coefficient, the more motivated the agent is to avoid obstacles. d0 is used as a constant to ensure that the value of the reward function does not become excessively large even when the distance is large (that is, there is a certain safe distance between the agent and the obstacle).

[0146] When the distance between the agent and the obstacle is greater than or equal to the critical safety distance d1, the reward function is zero. This means that within the safe distance, the agent is no longer penalized and will not receive additional rewards, thus encouraging the agent to move freely while ensuring safety.

[0147] This reward function drives the agent to stay away from obstacles and avoid collisions by applying negative rewards within the critical safety distance. Such a design can effectively guide the agent to learn obstacle avoidance behavior in the reinforcement learning framework. In addition, by adjusting the parameters, the agent's sensitivity to obstacles can be adjusted to adapt to different task requirements and environmental complexity. In the process of reinforcement learning, as the agent's understanding of the environment continues to deepen, the reward function will gradually guide the agent to learn to optimize the path, which can effectively avoid obstacles and complete the task objectives more efficiently. In the process of reinforcement learning, as the agent's understanding of the environment continues to deepen, the reward function will gradually guide the agent to learn to optimize the path, which can effectively avoid obstacles and complete the task objectives more efficiently.

[0148] S150, performing obstacle avoidance simulation verification on the first intelligent agent according to the first control input.

[0149] In one possible implementation, an obstacle avoidance simulation verification is performed on the first intelligent agent based on the first control input, specifically including: constructing a simulation environment, the simulation environment including at least two virtual obstacles and a virtual target point; constructing a virtual intelligent agent model based on the first control input; allowing the virtual intelligent agent to interact in the simulation environment; determining whether the virtual intelligent agent can avoid the virtual obstacles and reach the virtual target point; if it is determined that the virtual intelligent agent avoids the virtual obstacles and reaches the virtual target point, determining that the first control input is an obstacle avoidance strategy.

[0150] Specifically, create a simulation environment that contains multiple dynamic and static obstacles to increase the complexity of the obstacle avoidance task. Set a virtual target point in the environment, and the agent needs to avoid obstacles and reach this target point. The position of the target point can be fixed or dynamically changed. Arrange at least two obstacles to simulate the obstacle avoidance task of the agent in a complex environment. Obstacles should have different sizes, shapes, and motion characteristics to increase the difficulty of the agent's perception and decision-making.

[0151] For example, refer to Figure 2 In the Cartesian coordinate system of the XY plane, the obstacle avoidance problem of a cluster with N=50 agents is studied. The initial positions of the agents are randomly and uniformly distributed in the rectangular area of ​​x∈[100,50], y∈[100,100], and the initial velocities are all 0. The cluster needs to cross two circular obstacles with center positions (100,50) and (100,100) to reach the target point with coordinates (300,75). Other relevant simulation parameters: ε=0.1,a=b=4.5,c=0,h=0.2,d=16,r=20,e1=0.5,e2=0,c1=c2=1,μ=30,d0=15,d1=12,simulation step size 0.01s,simulation duration 1s.

[0152] Based on the first control input, the dynamic model of the virtual agent is constructed. The agent obtains obstacle information and the location of the target point in the environment through sensors or perception modules, thereby determining its own obstacle avoidance strategy. The motion model of the agent determines how it changes its displacement and speed according to the control input. The model can use simplified dynamics or more complex physical simulation (such as considering factors such as friction and gravity). Figure 3 , the agent makes real-time decisions based on the environmental information (obstacles and target points) it observes, generates new control inputs, and performs corresponding motion behaviors. The agent needs to continuously update its state during the motion process and interact with obstacles and target points in the simulation environment. Dynamic environments in the real world can be simulated by introducing dynamic obstacles (such as moving vehicles or other agents) to test the adaptability of the agent in complex environments.

[0153] The simulation system monitors the distance between the agent and the obstacle in real time. If the agent always maintains a safe distance, it is considered that the agent successfully avoids the obstacle. If the agent collides with the obstacle, it is considered that the obstacle avoidance fails. Figure 4 , under the premise that the agent successfully avoids obstacles, if it can reach the target point within the set time, the agent is considered to have successfully completed the task. The criterion for reaching the target point can be based on the distance between the agent and the target point being less than a certain threshold. If the agent can successfully avoid obstacles and reach the target point in multiple simulation tests, the first control input adopted can be considered to be an effective obstacle avoidance strategy. By testing the same obstacle avoidance strategy in different scenarios, its versatility and robustness in multiple environments can be verified. In addition, based on the simulation results, the performance of the obstacle avoidance strategy is evaluated, including indicators such as obstacle avoidance success rate, time to reach the target point, and path efficiency. If there is a failure in the simulation, further adjust the reward function, control input or algorithm parameters, and iteratively optimize the obstacle avoidance strategy.

[0154] This application can effectively cope with complex and dynamic environments by dynamically acquiring and analyzing the initial dynamic parameters of the agent, observing the positions of other agents and obstacles in the space in real time, and combining the motion parameters of the agent to dynamically determine the control input of the agent. By verifying the obstacle avoidance effect of the control input in a simulation environment, static presetting and dependence on the global environment are avoided, greatly improving the adaptability and flexibility of the algorithm. In addition, by adaptively adjusting the behavior of the agent, the dependence on the environmental model is reduced, thereby improving the obstacle avoidance performance and practicality of the multi-agent cluster in uncertain and dynamic environments.

[0155] This embodiment also discloses a multi-agent cluster obstacle avoidance device, referring to Figure 5 The device includes an acquisition module 501, a judgment module 502, a processing module 503 and a simulation module 504, wherein:

[0156] The acquisition module 501 is used to acquire the initial dynamic parameters of the agent cluster including multiple agents, and the initial dynamic parameters include displacement, speed and initial control input.

[0157] The judgment module 502 is used to determine a plurality of second agents within a preset range of the first agent according to the initial dynamic parameters, wherein the first agent is any one of the plurality of agents.

[0158] The acquisition module 501 is used to acquire the motion parameters of the first agent and the observed obstacle position.

[0159] The processing module 503 is used to determine the first control input of the first agent based on the motion parameters, the obstacle position and the initial dynamic parameters.

[0160] The simulation module 504 is used to perform obstacle avoidance simulation verification on the first agent according to the first control input.

[0161] In a possible implementation, the processing module 503 is used to determine the collision avoidance control input of the first agent according to the first speed of the first agent, the first displacement of the first agent, the second speed of the second agent, and the second displacement of the second agent. The collision avoidance control input is expressed as follows:

[0162]

[0163] Among them, f i g is the collision avoidance control input, q i is the first displacement of the first agent, p i is the first speed of the first agent, q j is the second displacement of the jth second agent, p j is the second speed of the jth second agent, c1 and c2 are the control parameters of the first agent, n ij is the displacement vector of the first agent pointing to the jth second agent, a ij is the interaction strength coefficient, Φ α is the distance function.

[0164] The acquisition module 501 is used to acquire the target point position and target point speed of the agent cluster.

[0165] The processing module 503 is used to determine the initial control input according to the target point position and the target point speed. The initial control input is expressed as follows:

[0166] f i n =-e1(q i -q target )-e2(pi -p target )

[0167] Among them, f i n is the initial control input, q i is the first displacement, p i is the first speed, q target is the target point position, p target is the target point speed, e1 and e2 are control parameters.

[0168] The processing module 503 is used to determine a first control input based on the collision avoidance control input and the initial control input. The first control input is expressed as follows:

[0169]

[0170] Among them, u i is the first control input, f i g To avoid collisions between multiple agents, f i n is the initial control input, f i β To control the obstacle avoidance input.

[0171] In a possible implementation, the processing module 503 is used to determine an observation space based on the first displacement and the first speed, where the observation space is expressed as follows:

[0172]

[0173] Among them, s i is the observation space, s i self is the first displacement and the first velocity, s i other is the observation of the first agent on the second agent, S i obstacle is the first agent’s observation of the obstacle.

[0174] The processing module 503 is used to determine the control obstacle avoidance input based on the observation space. The control obstacle avoidance input is expressed as follows:

[0175]

[0176] Among them, f i β To control the obstacle avoidance input, Δp i is the speed change.

[0177] In a possible implementation, the processing module 503 is used to determine a displacement vector according to the first displacement and the second displacement. The displacement vector is specifically expressed as follows:

[0178]

[0179] Among them, n ij is the displacement vector, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent;

[0180] The processing module 503 is used to determine the interaction strength coefficient based on the first displacement and the second displacement, which is specifically calculated by the following formula:

[0181]

[0182] Among them, a ij is the interaction strength coefficient, ‖q j -q i ‖ σ It means (q j -q i )’s σ norm, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent, and the σ norm is defined as follows:

[0183]

[0184] Among them, the constant ε>0, ‖x‖ σ represents the σ norm of vector x, ‖x‖ represents the length of the vector, and x is a vector of any dimension.

[0185] In a possible implementation, the acquisition module 501 is used to acquire multiple obstacles observed by the first intelligent agent and multiple second intelligent agents.

[0186] The judgment module 502 is used to determine the distance between each obstacle and the first agent.

[0187] The judgment module 502 is used to determine a target obstacle among multiple obstacles, and the distance between the target obstacle and the first intelligent agent is the shortest.

[0188] The processing module 503 is used to establish a reward function based on the distance between the target obstacle and the first agent. The reward function is as follows:

[0189]

[0190] Among them, r i is the reward function, d iis the distance between the first agent and the target obstacle, d1 is the critical safety distance, μ is the repulsion coefficient, and d0 is the reward adjustment scale.

[0191] In a possible implementation, the processing module 503 is used to determine a distance function based on the distance between the first agent and the second agent, which is specifically represented by the following formula:

[0192]

[0193] Among them, Φ α (z) is the distance function with distance z as the independent variable, z = || q j -q i ||, r a is the normalized scale, d a is a constant offset.

[0194] In a possible implementation, the simulation module 504 is used to construct a simulation environment, where the simulation environment includes at least two virtual obstacles and a virtual target point.

[0195] The simulation module 504 is used to construct a virtual agent model based on the first control input.

[0196] The simulation module 504 is used to make the virtual agents interact in the simulation environment.

[0197] The judgment module 502 is used to judge whether the virtual agent can avoid virtual obstacles and reach the virtual target point.

[0198] The judgment module 502 is used to determine that the first control input is an obstacle avoidance strategy if it is determined that the virtual agent avoids the virtual obstacle and reaches the virtual target point.

[0199] It should be noted that: when the device provided in the above embodiment realizes its function, only the division of the above functional modules is used as an example. In actual application, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0200] This embodiment also discloses an electronic device, referring to Figure 6 The electronic device may include: at least one processor 601 , at least one communication bus 602 , a user interface 603 , a network interface 604 , and at least one memory 605 .

[0201] The communication bus 602 is used to realize the connection and communication between these components.

[0202] The user interface 603 may include a display screen (Display) and a camera (Camera), and the optional user interface 603 may also include a standard wired interface and a wireless interface.

[0203] The network interface 604 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0204] Among them, the processor 601 may include one or more processing cores. The processor 601 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 605, and calling data stored in the memory 605. Optionally, the processor 601 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 601 can integrate one or more combinations of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 601, and it can be implemented by a single chip.

[0205] Among them, the memory 605 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory 605 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 605 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 605 may also be at least one storage device located away from the aforementioned processor 601. The memory 605 as a computer storage medium may include an operating system, a network communication module, a user interface 603 module, and an application program for a multi-agent cluster obstacle avoidance method.

[0206] exist Figure 6 In the electronic device shown, the user interface 603 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 601 can be used to call an application program for a multi-agent cluster obstacle avoidance method stored in the memory 605. When executed by one or more processors 601, the electronic device executes one or more methods in the above-mentioned embodiments.

[0207] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application.

[0208] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0210] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0211] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0212] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory 605, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory 605 includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk.

[0213] The present application also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores instructions. When executed by one or more processors 601, the electronic device executes one or more of the methods described in the above embodiments.

[0214] The above is only an exemplary embodiment of the present disclosure, and the scope of the present disclosure cannot be limited thereto. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, it will be easy for those skilled in the art to think of other embodiments of the present disclosure. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the technical field that are not recorded in the present disclosure. The description and examples are regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A multi-agent cluster obstacle avoidance method, characterized in that: The method comprises: Acquire initial dynamic parameters of a plurality of agents in an agent cluster, wherein the initial dynamic parameters include displacement, velocity, and initial control input; Determine, according to the initial dynamic parameters, a plurality of second agents within a preset range of the first agent, wherein the first agent is any one of the plurality of agents; Obtaining motion parameters of the first agent and an observed obstacle position; Determining a first control input of the first agent based on the motion parameters, the obstacle position, and the initial dynamic parameters; According to the first control input, an obstacle avoidance simulation verification is performed on the first agent.

2. A multi-agent cluster obstacle avoidance method according to claim 1, characterized in that: The determining, based on the motion parameter, the obstacle position and the initial dynamics parameter, a first control input of the first agent specifically includes: According to the first speed of the first agent, the first displacement of the first agent, the second speed of the second agent, and the second displacement of the second agent, a collision avoidance control input of the first agent is determined, and the collision avoidance control input is expressed as follows: Among them, f i g is the collision avoidance control input, q i is the first displacement of the first agent, p i is the first speed of the first agent, q j is the second displacement of the jth second agent, p j is the second speed of the jth second agent, c1 and c2 are the control parameters of the first agent, n ij is the displacement vector of the first agent’s displacement pointing to the jth second agent’s displacement, a ij is the interaction strength coefficient, Φ α is the distance function; Obtaining the target point position and target point speed of the intelligent agent cluster; According to the target point position and the target point speed, an initial control input is determined, and the initial control input is expressed as follows: f i n =-e1(q i -q target )-e2(p i -p target ) Among them, f i n is the initial control input, q i is the first displacement, p i is the first speed, q target is the target point position, p target is the target point speed, e1 and e2 are control parameters; Based on the collision avoidance control input and the initial control input, the first control input is determined, and the first control input is expressed as follows: Among them, u i is the first control input, f i g To avoid collisions between multiple agents, f i n is the initial control input, f i β To control the obstacle avoidance input.

3. A multi-agent cluster obstacle avoidance method according to claim 2, characterized in that: The determining the first control input based on the obstacle avoidance control input specifically further includes: Based on the first displacement and the first velocity, an observation space is determined, and the observation space is expressed as follows: Among them, s i is the observation space, s i self is the first displacement and the first speed, s i other is the observation of the second agent by the first agent, S i obstacle is the observation of the obstacle by the first intelligent agent; Based on the observation space, the obstacle avoidance control input is determined, and the obstacle avoidance control input is expressed as follows: Among them, f i β is the obstacle avoidance control input, Δp i is the speed change.

4. A multi-agent cluster obstacle avoidance method according to claim 2, characterized in that: The step of determining the collision avoidance control input of the first agent according to the first speed of the first agent, the first displacement of the first agent, the second speed of the second agent, and the second displacement of the second agent specifically includes: The displacement vector is determined according to the first displacement and the second displacement, and the displacement vector is specifically expressed as follows: Among them, n ij is the displacement vector, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent; Based on the first displacement and the second displacement, the interaction strength coefficient is determined, which is specifically calculated by the following formula: Among them, a ij is the interaction strength coefficient, ‖q j -q i ‖ σ It means (q j -q i )’s σ norm, q i is the first displacement of the first agent, q j is the second displacement of the j-th second agent, and the σ norm is defined as follows: Among them, the constant ε>0, ‖x‖ σ represents the σ norm of vector x, ‖x‖ represents the length of the vector, and x is a vector of any dimension.

5. A multi-agent cluster obstacle avoidance method according to claim 1, characterized in that: After determining the first control input of the first agent based on the motion parameter, the obstacle position and the initial dynamics parameter, the method further includes: Obtaining a plurality of obstacles observed by the first intelligent agent and a plurality of the second intelligent agents; Determining the distance between each of the obstacles and the first agent; Determine a target obstacle among a plurality of obstacles, wherein the distance between the target obstacle and the first intelligent agent is shortest; Based on the distance between the target obstacle and the first agent, a reward function is established, and the reward function is specifically as follows: Among them, r i is the reward function, d i is the distance between the first agent and the target obstacle, d1 is the critical safety distance, μ is the repulsion coefficient, and d0 is the reward adjustment scale.

6. A multi-agent cluster obstacle avoidance method according to claim 2, characterized in that: The distance function is determined based on the distance between the first agent and the second agent, and is specifically expressed by the following formula: Among them, Φ α (z) is the distance function with distance z as the independent variable, z = || q j -q i ||, r a is the normalized scale, d a is a constant offset.

7. A multi-agent cluster obstacle avoidance method according to claim 1, characterized in that: The performing obstacle avoidance simulation verification on the first agent according to the first control input specifically includes: Constructing a simulation environment, wherein the simulation environment includes at least two virtual obstacles and a virtual target point; constructing a virtual agent model based on the first control input; Allowing the virtual agent to interact in the simulation environment; Determining whether the virtual agent can avoid the virtual obstacle and reach the virtual target point; If it is determined that the virtual agent avoids the virtual obstacle and reaches the virtual target point, the first control input is determined to be an obstacle avoidance strategy.

8. A multi-agent cluster obstacle avoidance device, characterized in that: The device comprises an acquisition module (501), a judgment module (502), a processing module (503) and a simulation module (504), wherein: The acquisition module (501) is used to acquire the initial dynamic parameters of the agent cluster including multiple agents, wherein the initial dynamic parameters include displacement, velocity and initial control input; The judgment module (502) is used to determine a plurality of second agents within a preset range of the first agent according to the initial dynamic parameters, wherein the first agent is any one of the plurality of agents; The acquisition module (501) is used to acquire the motion parameters of the first agent and the observed obstacle position; The processing module (503) is used to determine a first control input of the first agent based on the motion parameter, the obstacle position and the initial dynamics parameter; The simulation module (504) is used to perform obstacle avoidance simulation verification on the first intelligent agent according to the first control input.

9. An electronic device, characterized in that: The electronic device comprises a processor (601), a communication bus (602), a memory (605), a user interface (603), a network interface (604) and a memory (605), wherein the memory (605) is used to store instructions, the user interface (603) and the network interface (604) are both used to communicate with other devices, and the processor (601) is used to execute the instructions stored in the memory (605) so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images