Multi-robot task allocation and path planning method based on swarm intelligence and large model

Abstract

The invention discloses a multi-robot task allocation and path planning method based on swarm intelligence and a large model. The method comprises the following steps: acquiring map information, robot information and task information; configuring and agenting an initialized large language model, creating a robot list and a task list through the large language model according to the robot information and the task information, and performing task allocation according to the robot list and the task list to obtain a task allocation strategy; and according to the task allocation strategy and the map information, performing path planning on the robot by adopting a swarm intelligence algorithm to obtain a path planning result, controlling the robot to perform task execution according to the path planning result, and feeding back the task execution result to the large language model. And marking and updating task completion through the large language model.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot control and path planning, and in particular relates to a multi-robot task allocation and path planning method based on swarm intelligence and a large model. Background Art

[0002] The efficiency of robot task allocation can impact energy consumption and economic benefits. For environments with known states, dynamic task allocation for multiple robots can be achieved by combining optimal task allocation with path planning algorithms. Swarm intelligence optimization algorithms are a commonly used adaptive algorithm in path planning and task allocation applications. Common dynamic programming algorithms used in swarm intelligence optimization include particle swarm optimization, genetic algorithms, and ant colony algorithms. Artificial intelligence frameworks, such as reinforcement learning, are also used for multi-robot task allocation.

[0003] However, traditional path planning algorithms face the problem of being stuck in local optimality and the challenge of balancing local and global search capabilities. For online multi-robot path planning and task allocation, the algorithm framework needs to make decisions quickly, but traditional methods lack decision-making efficiency. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a multi-robot task allocation and path planning method based on swarm intelligence and large models, and proposes an improved particle swarm optimization algorithm combined with large model technology to realize multi-robot online path planning and task allocation to solve the problems existing in the above-mentioned existing technologies.

[0005] To achieve the above objectives, the present invention provides a multi-robot task allocation and path planning method based on swarm intelligence and a large model, comprising:

[0006] Obtain map information, robot information, and mission information;

[0007] Configure and initialize a large language model, create a robot list and a task list based on robot information and task information through the large language model, assign tasks based on the robot list and task list, and obtain a task assignment strategy;

[0008] According to the task allocation strategy and map information, a swarm intelligence algorithm is used to plan the path of the robot to obtain a path planning result. The robot is controlled to perform the task according to the path planning result. The task execution result is fed back to the large language model, and the task completion is marked and updated through the large language model.

[0009] Optionally, the process of obtaining the map information includes:

[0010] According to the execution scenario of the current task, map information is generated, wherein the map information is a binary grid map, and the grid map is stored in a two-dimensional matrix.

[0011] Optionally, the process of configuring and proxying the initialization of a large language model includes:

[0012] The large language model is configured with a model type. After the configuration, the large language model is initialized as an agent. During the agent initialization process, a dialogue agent instance is created for the large language model, wherein the dialogue agent instance configuration corresponds to task processing, wherein the task processing includes task creation, task allocation, and task completion.

[0013] Optionally, the large language model interacts with the swarm intelligence algorithm and external input information through AP communication, and performs interactive conversation simulation on the interactive information of the communication through group chat management to obtain corresponding interactive information, and performs task processing through the interactive information.

[0014] Optionally, the large language model creates a robot list and a task list based on the robot information and task information by creating tasks, wherein the robot list includes the robot's ID, capacity and name; the task list includes the task ID and the building, floor, and department attributes.

[0015] Optionally, the large language model allocates tasks according to the robot list and task list, wherein the tasks are allocated according to the buildings and floors in the tasks and the capacity of the robots in the robot list, to obtain a task allocation strategy, and provide it to the server.

[0016] Optionally, the process of using a swarm intelligence algorithm to plan the robot's path includes:

[0017] The objective function of the swarm intelligence algorithm is constructed, and the path planning of the robot is performed by the swarm intelligence algorithm with the goal of minimizing the objective function, wherein the swarm intelligence algorithm is an optimization algorithm combining particle swarm and simulated annealing algorithm.

[0018] Optionally, the objective function F path (x t ,y t )for:

[0019]

[0020] Among them, (x t ,y t ) is the current position of the robot, w1 and w2 are the weight coefficients of the objective functions of path length and collision respectively;

[0021] Path length and collision The objective function is calculated as follows:

[0022]

[0023] in, is the path point p at time t k , For waypoints and distance, For waypoints The location coordinates of . for The collision cost, r c is the radius of the obstacle, j is the total number of obstacles, is the distance from the center of the obstacle to the particle.

[0024] Optionally, in the process of performing path planning for the robot using a swarm intelligence algorithm, particles are iteratively updated, and an optimal path, i.e., a path planning result, is obtained based on the iteratively updated particles; wherein the particles are path planning information;

[0025] The process of iteratively updating particles includes: and location The update formula is used to update particles, and the probability ρ is used to select whether to accept the current new particle: when the probability ρ is greater than the generated random number, the new particle is accepted:

[0026] in,

[0027]

[0028] in is the velocity of particle i at time t+1, ω is the inertia weight, c1 and c2 are the individual and social learning factors, r1 and r2 are random numbers in the interval [0,1]. pbest is the individual optimal value, and gbest is the group optimal value. The position of particle i at time t+1. is the fitness value of particle i at time t.

[0029] On the other hand, the present invention also provides a multi-robot task allocation and path planning system based on swarm intelligence and large models, which is used to execute the above method.

[0030] Compared with the prior art, the present invention has the following advantages and technical effects:

[0031] This paper studies how to achieve multi-robot collaborative scheduling based on swarm intelligence and large models. For multi-robot path planning and task allocation, existing mainstream solutions are based on swarm intelligence algorithms or reinforcement learning models. However, traditional swarm intelligence algorithms face local optimality issues, and task allocation is separated from path planning, relying on preset rules, resulting in low flexibility and efficiency. This method, however, leverages the collaborative mechanism of large models and swarm intelligence, employing distributed communication to innovatively integrate the task understanding capabilities of large models with the optimization efficiency of swarm intelligence, improving the method's adaptability and computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0033] Figure 1 Schematic diagram of the overall framework of the method according to an embodiment of the present invention;

[0034] Figure 2 Flowchart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0036] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0037] This invention relates to a multi-robot task allocation and path planning system that combines swarm intelligence algorithms with a large language model to achieve multi-robot collaborative control. This system generates and assigns tasks based on the location and available robots within an environment. It then uses an improved particle swarm optimization algorithm to generate paths based on task information. The improved particle swarm optimization algorithm balances global and local exploration capabilities, avoiding local optima, and ultimately provides updates on task completion status.

[0038] The purpose of the present invention is to provide a multi-robot task allocation and path planning method based on swarm intelligence and large models. Dynamic task allocation is performed through a large model, and then a swarm intelligence algorithm is used to perform multi-robot path planning, thereby realizing collaborative control of a multi-robot system, improving computational efficiency and reducing the fitness value of the objective function, overcoming the dilemma of swarm intelligence algorithms easily falling into local optimality, and balancing the capabilities of local search and global search. The overall process is shown in the figure below.

[0039] The present invention adopts the following technical solutions: Figure 2 As shown:

[0040] Step 1: Generate a binary raster map based on the current application environment.

[0041] Step 2: Large model configuration and agent initialization, create a group of dialogue agent instances, and initialize the particle swarm for the improved particle swarm algorithm.

[0042] Step 3: The large model creates a list of tasks and robots and assigns tasks; the improved particle swarm algorithm updates the position and velocity of particles based on the assigned tasks, and decides whether to accept the new solution based on the objective function and probability.

[0043] Step 4: Continuously update the position and velocity of the particle swarm. If the fitness value of the current particle objective function is lower, update the individual optimum; if the fitness value of the current particle is lower than the global optimum, update the global optimum until the optimal path is obtained.

[0044] Step 5: The large model updates the task list and robot list according to the obtained path until the task is completed.

[0045] The detailed description of the above steps is as follows:

[0046] In step 1 above, a binary grid map is generated according to the current application scenario, with 0 representing a passable area and 1 representing an impassable area.

[0047] In step 2 above, during the configuration of the large model and agent initialization, each instance has different roles related to task processing, namely task creation, task assignment, and task completion. Each agent is configured with a specific system message to guide its interactions and the GPT-4 model.

[0048] In step 3 above, the large model's "Create Task" function uses an HTTP POST method to communicate with the Task Management API to initialize a new task and list of robots. It returns a JSON response if successful, or an error message if the request fails. The large model generates the task and available robots and returns them as a JSON response.

[0049] Then, the "assignment task" of the large model groups the tasks according to the building and floor, allocates them according to the capacity of the AGV, formulates a structured task allocation strategy, and communicates the allocation strategy to the server. The improved particle swarm algorithm uses the fitness of the objective function to calculate and obtain the optimal path. The objective function F considering path length and collision path (x t ,y t ) is as follows, where (xt ,y t ) is the current position of the robot, w1 and w2 are the weight coefficients of the objective functions of path length and collision, respectively.

[0050]

[0051] In steps 3 and 4 above, the swarm intelligence algorithm inspired by particle swarm and simulated annealing algorithms is optimized by particle swarm optimization. and location The update formula updates the particles and uses the probability ρ to decide whether to accept the current new solution until the optimal path is obtained. If the probability ρ is greater than the generated random number, the new solution is accepted, rather than only accepting solutions below the individual optimal fitness. This allows the algorithm to escape the local optimal trap and enhance local and global search capabilities. The speed and position update formulas, as well as the probability calculation formula, are as follows:

[0052]

[0053] in is the velocity of particle i at time t+1, ω is the inertia weight, c1 and c2 are the individual and social learning factors, r1 and r2 are random numbers in the interval [0,1]. pbest is the individual optimal value, and gbest is the group optimal value. The position of particle i at time t+1. is the fitness value of particle i at time t.

[0054] In step 5 above, the "complete task" function of the large model marks the task as completed by sending a corresponding HTTP POST request to the server and updates the task status in the system.

[0055] like Figure 1 and Figure 2 As shown, the present invention is a multi-robot task allocation and path planning method that combines swarm intelligence and large models. The specific implementation steps of the invention are as follows:

[0056] Step 1: Based on the current application environment information, the passable area is represented by 0, the impassable area is represented by 1, and the raster map is stored in a two-dimensional matrix.

[0057] Step 2: Configure the Large Language Model. Select the GPT-4 model, which can understand complex queries and provide understandable, relevant input. Then, perform agent initialization, creating a set of conversational agent instances, or AI agents. Each agent is configured with the Large Language Model. Each agent (or instance) is configured with different roles related to task processing, namely, task creation, task assignment, and task completion.

[0058] GPT-4's integration and workflow include group chat management and API communication. Group chat management builds a group chat instance to simulate interactive conversations between agent instances, enabling them to engage in multi-round conversations focused on task execution. Group chat management oversees these interactions and seamlessly coordinates task management between agent instances. For ease of understanding, group chat management provides a communication scenario or space between the different intelligent agents mentioned above. The transmission of data between agents, or the receipt and release of tasks, is achieved through group chat management. Task information and robot information are input into the current group chat management, and the agent instances for task creation, task assignment, and task completion can perform corresponding operations based on these inputs. Each agent instance updates the task status and output information in JSON format in the group chat management, transmitting information through interactive conversations to make real-time decisions and complete tasks.

[0059] The agent communicates with external devices, such as a robot controller or command delivery system, via an API. API communication involves issuing structured HTTP requests to an external API, protected by an API key, and handling responses using standard error handling techniques. Swarm intelligence algorithms, inspired by particle swarm and simulated annealing algorithms, initialize the particle swarm's position and velocity in a random manner.

[0060] Step 3: The "Create Task" function in the large model generates tasks and available robots and returns them as a JSON response. The robot list contains the robot ID, capacity, and name, and the task list contains the task ID, the building and floor where it is located, and the department attributes. The response is as follows:

[0061]

[0062]

[0063] Then, the "assignment tasks" of the big model grouped the tasks according to their buildings and floors, and allocated them according to the capacity of the AGVs. A structured task allocation strategy was developed, in which the robots with the closest distance and the most suitable capacity were assigned to the corresponding floors, and the allocation was communicated to the server, where the server is the server that configured the big model.

[0064] The robot controller or command issuing system is equipped with a swarm intelligence algorithm. The swarm intelligence algorithm, inspired by the particle swarm and simulated annealing algorithms, uses the fitness of the objective function to calculate the objective function F considering the path length and collision. path (x t ,y t ) is as follows, where (x t ,y t) is the current position of the robot, w1 and w2 are the weight coefficients of the objective functions of path length and collision, respectively.

[0065]

[0066] About path length and collision The objective function is calculated as follows:

[0067]

[0068] in, is the path point p at time t k , For waypoints and distance, For waypoints The location coordinates of . for The collision cost, r c is the radius of the obstacle, j is the total number of obstacles, is the distance from the center of the obstacle to the particle.

[0069] Each robot supports multi-angle steering and has current environment information, moves at a constant speed, and has information and task lists of other robots, treating other robots as moving obstacles.

[0070] Steps 3 and 4: Speed of swarm intelligence algorithms inspired by particle swarm and simulated annealing algorithms through particle swarm optimization and location The update formula updates the particles and uses the probability ρ to decide whether to accept the new solution until the optimal path is obtained. A particle is an iterative solution for a path point, and the solution with the minimum target value corresponding to the particle is the path point of the optimal path. If the probability ρ is greater than the generated random number, the new solution is accepted. The speed and position update formulas, as well as the probability calculation formulas, are as follows:

[0071]

[0072] in is the velocity of particle i at time t+1, ω is the inertia weight, c1 and c2 are the individual and social learning factors, r1 and r2 are random numbers in the interval [0,1]. pbest is the individual optimal value, and gbest is the group optimal value. The position of particle i at time t+1. is the fitness value of particle i at time t.

[0073] Step 5: The large model "Task Completion Agent" marks the task as completed by sending a corresponding HTTP POST request to the server and updates the task status in the system, as shown below:

[0074]

[0075] After updating the task status in the system, return to step 1 to reallocate the tasks.

[0076] This paper studies how to achieve multi-robot collaborative scheduling based on swarm intelligence and large models. For multi-robot path planning and task allocation, existing mainstream solutions are based on swarm intelligence algorithms or reinforcement learning models. However, traditional swarm intelligence algorithms face local optimality issues, and task allocation is separated from path planning, relying on preset rules, resulting in low flexibility and efficiency. This method, however, leverages the collaborative mechanism of large models and swarm intelligence, employing distributed communication to innovatively integrate the task understanding capabilities of large models with the optimization efficiency of swarm intelligence, improving the method's adaptability and computational efficiency.

[0077] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A multi-robot task allocation and path planning method based on swarm intelligence and large models, characterized in that: include: Obtain map information, robot information, and mission information; Configure and initialize a large language model, create a robot list and a task list based on robot information and task information through the large language model, assign tasks based on the robot list and task list, and obtain a task assignment strategy; According to the task allocation strategy and map information, a swarm intelligence algorithm is used to plan the path of the robot to obtain a path planning result. The robot is controlled to perform the task according to the path planning result. The task execution result is fed back to the large language model, and the task completion is marked and updated through the large language model.

2. The method according to claim 1, characterized in that The process of obtaining the map information includes: According to the execution scenario of the current task, map information is generated, wherein the map information is a binary grid map, and the grid map is stored in a two-dimensional matrix.

3. The method according to claim 1, characterized in that The process of configuring and initializing a large language model includes: The large language model is configured with a model type. After the configuration, the large language model is initialized as an agent. During the agent initialization process, a dialogue agent instance is created for the large language model, wherein the dialogue agent instance configuration corresponds to task processing, wherein the task processing includes task creation, task allocation, and task completion.

4. The method according to claim 1, wherein The large language model interacts with the swarm intelligence algorithm and external input information through AP communication, and performs interactive conversation simulation on the interactive information of the communication through group chat management to obtain corresponding interactive information, and performs task processing work through the interactive information.

5. The method according to claim 1, wherein The large language model creates a robot list and a task list according to the robot information and task information by creating tasks, wherein the robot list includes the robot's ID, capacity and name; the task list includes the task ID and the building, floor, and department attributes.

6. The method according to claim 1, characterized in that The large language model allocates tasks according to the robot list and task list, wherein the task allocation is performed according to the building and floor in the task and the capacity of the robots in the robot list, and a task allocation strategy is obtained and provided to the server.

7. The method according to claim 1, characterized in that The process of using swarm intelligence algorithms to plan the robot's path includes: The objective function of the swarm intelligence algorithm is constructed, and the path planning of the robot is performed by the swarm intelligence algorithm with the goal of minimizing the objective function, wherein the swarm intelligence algorithm is an optimization algorithm combining particle swarm and simulated annealing algorithm.

8. The method according to claim 7, characterized in that The objective function F path (x t ,y t )for: Among them, (x t ,y t ) is the current position of the robot, w1 and w2 are the weight coefficients of the objective functions of path length and collision respectively; Path length and collision The objective function is calculated as follows: in, is the path point p at time t k , For waypoints and distance, For waypoints The location coordinates of for The collision cost, r c is the radius of the obstacle, j is the total number of obstacles, is the distance from the center of the obstacle to the particle.

9. The method according to claim 7, characterized in that In the process of path planning for the robot through the swarm intelligence algorithm, the particles are iteratively updated, and the optimal path, i.e., the path planning result, is obtained based on the iteratively updated particles; The particles are path planning information; The process of iteratively updating particles includes: and location The update formula is used to update particles, and the probability ρ is used to select whether to accept the current new particle: when the probability ρ is greater than the generated random number, the new particle is accepted: in, in is the velocity of particle i at time t+1, ω is the inertia weight, c1 and c2 are individual and social learning factors, r1 and r2 are random numbers in the interval [0,1], pbest is the individual optimal value, gbest is the group optimal value, The position of particle i at time t+1, is the fitness value of particle i at time t.

10. A multi-robot task allocation and path planning system based on swarm intelligence and large models, characterized by: Used to execute the method according to any one of claims 1 to 9.

Images