Multi-agent collaborative integrated navigation virtual simulation system and method based on Unity 3D

Abstract

The invention provides a Unity3D-based multi-agent collaborative integrated navigation virtual simulation system and method, belongs to the technical field of integrated navigation, and solves the problems that a traditional integrated navigation experiment method is high in cost and risk and is difficult to simulate various complex actual working conditions, and the simulation system comprises a display layer, a business layer and an underlying service layer; the display layer provides an operation and monitoring platform for a user; the business layer is used for performing control logic, queue management, path planning and artificial intelligence training functions on the agents in the simulation environment; and the underlying service is used for performing agent machine learning model training and project version management. The invention provides a low-cost and high-safety virtual simulation scheme for intelligent agent motion control and collaborative operation efficiency optimization, an improved A * search algorithm and an artificial intelligence training interface are integrated, and an economical and efficient virtual test platform is provided for research and development of an intelligent agent control algorithm.

Description

Technical Field

[0001] The present invention relates to a multi-agent collaborative integrated navigation virtual simulation system and method based on Unity3D, belonging to the technical field of integrated navigation. Background Technique

[0002] With the continuous progress of technology, multi-agent systems play an increasingly important role in the fields of automation and intelligence. These systems are usually composed of multiple interacting agents, which can work together to complete complex tasks. In order to improve the collaborative efficiency of these agents and ensure the safety of operations, it is necessary to conduct in-depth research and testing on their motion control and collaborative strategies. Traditional experimental methods are often costly, risky, and difficult to simulate various complex actual working conditions. Therefore, it is particularly important to develop a test and training platform based on virtual simulation technology.

[0003] In the field of virtual simulation, as a powerful game development engine, Unity3D is widely used in simulation and training systems due to its high interactivity and realistic visual effects. Through Unity3D, complex virtual environments can be constructed to simulate the behaviors and responses of agents in various environments. In addition, Unity3D supports multiple programming languages, especially the C# language, enabling developers to flexibly implement the control logic and collaborative strategies of agents.

[0004] In order to enable agents to effectively avoid obstacles and work together in a virtual environment, it is necessary to integrate advanced path planning and obstacle avoidance algorithms. As a classic path search algorithm, the A* algorithm is widely used in various navigation and path planning problems due to its high efficiency and accuracy. By optimizing the A* algorithm, the obstacle avoidance ability of agents in complex environments can be further improved.

[0005] At the same time, with the development of artificial intelligence technology, especially the application of reinforcement learning in multi-agent systems, agents can autonomously learn optimal control strategies through interaction with the environment. As a machine learning framework based on Unity3D, MLAgent provides an interface integrating TensorFlow, allowing researchers to train and deploy AI models in a virtual environment to achieve the autonomous control of agents. Summary of the Invention

[0006] In order to solve the problems of high cost, high risk, and difficulty in simulating various complex actual working conditions of traditional integrated navigation experimental methods, the present invention further proposes a multi-agent collaborative integrated navigation virtual simulation system and method based on Unity3D.

[0007] The technical solution adopted by the present invention to solve the above problems is as follows: A multi-agent collaborative combined navigation virtual simulation system based on Unity3D proposed by the present invention includes:

[0008] A display layer, a business layer, and an underlying service;

[0009] The display layer provides an operation and monitoring platform for users;

[0010] The business layer is used for control logic, queue management, path planning, and artificial intelligence training functions of agents in the simulation environment;

[0011] The underlying service is used for agent machine learning model training and project version management.

[0012] Optionally, the display layer includes an environment simulation module, a UnityUI interface, and a Unity Editor window;

[0013] The environment simulation module is used to construct a virtual environment required for agent interaction;

[0014] The UnityUI interface provides an operation page for users;

[0015] The Unity Editor window provides a monitoring platform for users.

[0016] Optionally, the business layer includes an Agent control module, an Agent queue control module, a path planning module, and an AI environment module;

[0017] The Agent control module is responsible for the control of a single agent. The control of a single agent includes the management of power data, torque data, path data, following data, mode control, and algorithm selection;

[0018] The Agent queue control module is used to manage the data, target data, and queue control of the agent queue;

[0019] The path planning module is used to calculate the optimal path for the agent to reach the target position from the current position;

[0020] The AI environment module is used to perform artificial intelligence training on the agent and predict the best action plan according to the environmental feedback.

[0021] Optionally, the Agent control module includes a single Agent control module and a multi-Agent control module;

[0022] The single Agent control module is used for agent power control, target path planning, and action state control;

[0023] The multi-Agent control module is used for queue mode switching, agent queue management, and queue motion simulation.

[0024] Optionally, the underlying services include the TensorFlow training interface service module and the Plastic SCM project version control module;

[0025] The TensorFlow training interface service is used to convert the data collected by the agent into training signals to guide the behavior adjustment of the agent;

[0026] Plastic SCM project version control is used for project version management.

[0027] A multi-agent collaborative combined navigation virtual simulation method based on Unity3D, comprising:

[0028] Step 1: Construct a Unity3D virtual environment required for agent interaction based on the environment simulation module;

[0029] Step 2: Combine the single-Agent control module, the path planning module, and the multi-Agent control module, and set the control strategy of the agents in the Unity3D virtual environment through C# scripts;

[0030] Step 3: Utilize the improved A* search algorithm and the RVO2 simulator idea, combined with the control strategy of the agents, to control the multi-agents to avoid obstacles;

[0031] Step 4: Train and deploy an AI model in the Unity3D virtual environment through the AI environment module to control the agents. By connecting to the TensorFlow training interface service and using the MLAgent plugin to train the control strategy of the agents, the agents can successfully avoid obstacles and complete the multi-agent collaborative combined navigation task.

[0032] Optionally, Step 1 specifically includes:

[0033] Step 1.1: Extract the information of all objects in the navigation environment from the multi-angle images, photos, and video materials of the objects through computer vision technology, and calculate and construct the 3D models of all objects by using one of the methods of structured light scanning, stereo vision, and structure from motion;

[0034] Step 1.2: Import the 3D models of all objects into the Unity3D environment, and optimize the 3D models of all objects during the import process. Among them, optimizing the 3D models of all objects includes simplifying polygons, optimizing textures and materials, and adjusting the scale and proportion of the models for the 3D models of the objects;

[0035] Step 1.3: Place the 3D models of all optimized objects at the corresponding positions in the Unity3D environment, and set the lighting, climate effects, and physical properties in the Unity3D environment to complete the construction of the Unity3D virtual environment.

[0036] Optionally, Step 2 specifically includes:

[0037] Step 2.1: Use the sensors and physical engine of Unity3D to enable the agent to perceive changes in the surrounding environment, and collect data on the agent's position, speed, and distances to surrounding objects through the integrated API;

[0038] Step 2.2: The path planning module uses the improved A* search algorithm to analyze the Unity3D virtual environment data, and combines the collected agent position, speed, and distances to surrounding objects to calculate the optimal path for the agent to reach the target position from the current position;

[0039] Step 2.3: Through the single-agent control module and multi-agent control module, enable multiple agents to coordinate their actions to maintain the formation according to the preset formation and dynamic adjustment strategy, and by continuously monitoring the surrounding environment of the agents, when a potential collision risk is detected, calculate the optimal collision avoidance path and adjust the speed and direction of the agents to complete the setting of the agent control strategy.

[0040] Optionally, Step 3 specifically includes:

[0041] Step 3.1: After receiving a path planning request, the agent activates the integrated control script of the A* algorithm module and the RVO2 simulator idea to calculate the optimal path from the current position to the target position;

[0042] Step 3.2: Calculate the repulsive force relationship between agent i and agent j in the x-axis direction;

[0043] Step 3.3: Combine Step 3.2 to calculate the repulsive force relationships between agent i and agent j in the y-axis and z-axis directions;

[0044] Step 3.4: Superimpose the repulsive force relationships between agent i and agent j in the xyz-axis directions to obtain the expected velocity components of the ith agent at time t in the xyz-axis directions in the inertial coordinate system;

[0045] Step 3.5: Dynamically update the search tree of the improved A* search algorithm in combination with the data collected by the sensors of Unity3D, and adjust the maximum turning angular velocity and acceleration of the agent to ensure that the agent moves along the optimal collision avoidance path and avoids obstacles;

[0046] The expression for the repulsive force relationship between agent i and agent j in the x-axis direction is:

[0047]

[0048] In formula (1), η is the repulsive force scale factor, l ij is the distance between agents i and j, and Rdetect is the influence radius of the repulsive force field;

[0049] The calculation formula for the expected velocity components of the ith agent at time t in the xyz-axis directions in the inertial coordinate system is:

[0050]

[0051] In formulas (3)-(5), is the expected velocity component of the ith agent at time t in the x-axis direction, is the expected velocity component of the ith agent at time t in the y-axis direction, is the expected velocity component of the ith agent at time t in the z-axis direction, and Δt is the interval for system state update.

[0052] Optionally, step 4 specifically includes:

[0053] Step 4.1: In the Unity3D virtual environment, receive real-time data from sensors through the MLAgent plugin, where the real-time data includes the positions, velocities of the agents, and the distances to surrounding obstacles, and convert the real-time data into training signals;

[0054] Step 4.2: By connecting to the Tensorflow training interface service, the AI model is trained with the assistance of the MLAgent plugin, predicts the best action plan based on the sensor data feedback, and completes the multi-agent collaborative combined navigation task.

[0055] The beneficial effects of the present invention are:

[0056] By providing a multi-agent collaborative combined navigation virtual simulation solution based on Unity3D, the present invention realizes low-cost and high-security agent motion control and optimization of collaborative operation efficiency. At the same time, it integrates the advanced A* search algorithm and artificial intelligence training interface, enabling the agents to perform effective obstacle avoidance and path planning training in a realistic virtual environment, improving the adaptability and intelligence level of the agents, and providing an economical and efficient virtual test platform for the research and development of agent control algorithms. Description of the Drawings

[0057] Figure 1 It is a schematic structural diagram of a multi-agent collaborative combined navigation virtual simulation system provided by the present invention;

[0058] Figure 2 It is the functional module diagram of the multi-agent collaborative integrated navigation virtual simulation system provided by the present invention;

[0059] Figure 3 It is the schematic flowchart of a multi-agent collaborative integrated navigation virtual simulation method based on Unity3D provided by the present invention;

[0060] Figure 4 It is the schematic diagram of agent collision avoidance provided by the present invention;

[0061] Figure 5 It is the data flow diagram of AI training provided by the present invention;

[0062] Figure 6 It is the flowchart of Agent life cycle provided by the present invention. Detailed implementation manners

[0063] Detailed implementation manner one: In combination with Figure 1 and Figure 2 to illustrate this implementation manner. As Figure 1 shown, the structure of a multi-agent collaborative integrated navigation virtual simulation system based on Unity3D described in this implementation manner includes: a display layer, a service layer, and an underlying service;

[0064] The display layer includes an HDRP virtual simulation environment, a UnityUI interface, and a Unity Editor window, which together provide an intuitive operation and monitoring platform for users. The service layer is the core of the simulation system, handling key service functions such as agent control logic, queue management, path planning, and artificial intelligence training. This layer is further divided into two modules: an Agent control module and an Agent queue control module. The Agent control module is responsible for the control of a single agent, including the management of power data, torque data, path data, following data, as well as mode control and algorithm selection. The Agent queue control module manages the data, target data, and queue control of the agent queue to ensure that multiple agents can work together. The underlying service provides the underlying support required by the simulation system, including the Tensorflow training interface service and Plastic SCM project version control, ensuring the training of the machine learning model of the agent and the version management of the project.

[0065] As Figure 2As shown in the figure, the functional modules of the multi-agent collaborative combined navigation virtual simulation system include an environment simulation module, a single Agent control module, a multi-Agent control module, a path planning module, and an AI environment module. The environment simulation module is used to construct the virtual environment required for agent interaction. The single Agent control module is used for agent power control, target path planning, and action state control. The multi-Agent control module is used for queue mode switching, agent queue management, and queue movement simulation. The path planning module is used to calculate the optimal path for the agent to reach the target position from the current position. The AI environment module is used to perform artificial intelligence training on the agent and predict the best action plan based on environmental feedback.

[0066] Specific Embodiment 2: In combination with Figures 3-6 This embodiment is described as follows. As Figure 3 shown, the steps of a multi-agent collaborative combined navigation virtual simulation method based on Unity3D described in this embodiment include:

[0067] S1: Virtual environment construction;

[0068] S101: The first step in constructing a virtual environment in Unity3D is to obtain the 3D models of all the objects with which the agents in the scene will interact. The acquisition of these models can be automatically completed by advanced computer vision techniques that can extract information from multi-angle images, photos, or video materials of the objects and calculate and construct the three-dimensional models of the objects using methods such as structured light scanning, stereo vision, or structure from motion (SFM). For models that cannot be obtained by automated methods, this embodiment relies on professional 3D modeling software such as AutoCAD, 3ds Max, Maya, etc., and professional personnel manually construct them according to the design blueprints or reference photos of the objects.

[0069] S102: In this embodiment, these finely constructed 3D models are imported into the Unity3D environment. During the import process, it is crucial to perform necessary optimizations on the models, which include simplifying polygons, optimizing textures and materials, and adjusting the scale and proportion of the models to ensure that they can maintain high-quality visual effects in the virtual environment and can efficiently perform physical simulations and interactions. The optimization of these models is crucial for ensuring the smooth interaction of agents in the virtual environment.

[0070] S103: In this embodiment, the scene layout is built in Unity3D, and the optimized 3D models are placed at appropriate positions in the simulation space to simulate the object distribution and relationships in the real world. In addition, the lighting, climate effects, and physical properties such as gravity and friction in the environment need to be set in this embodiment to enhance the realism of the environment. Through these meticulous construction steps, this embodiment provides an agent with a visually realistic and physically accurate operating space, laying a solid foundation for the subsequent simulation of agent collaborative operations.

[0071] S2: Control logic implementation;

[0072] S201: In the Unity3D environment, the control logic of the agent is precisely implemented through C# scripts. As Figure 3 shown, this process covers multiple key aspects of the agent's interaction with the virtual environment. First, the information acquisition module uses the sensors and physical engine of Unity3D to enable the agent to perceive changes in the surrounding environment, such as terrain undulations, climate conditions, and the effects of various resistance sources. Through the integrated API, the agent can collect data on its position, speed, distance to surrounding objects, etc., which are the basis for the agent's decision-making;

[0073] S202: The path planning module adopts advanced algorithms, such as the A* search algorithm, to calculate the optimal path for the agent to reach the target position from the current position. This involves the analysis of environmental data, including the passability of the terrain, the positions of obstacles, and possible dynamic changes. The path planning algorithm will generate a path considering all these factors to ensure that the agent can navigate to the target in the most efficient way while avoiding unnecessary energy consumption and potential collision risks;

[0074] S203: Combining the single-agent control module, path planning module, and multi-agent control module, the formation control and automatic obstacle avoidance functions are realized through collaborative algorithms and environmental perception technologies. The formation control module enables multiple agents to coordinate their actions according to preset formations or dynamic adjustment strategies to maintain the integrity and efficiency of the formation. The automatic obstacle avoidance function monitors the surrounding environment of the agent in real time. When a potential collision risk is detected, it can quickly calculate an obstacle avoidance path or adjust the speed and direction to ensure the safety of the agent. The implementation of these control logics provides solid technical support for the efficient and safe operation of the agent in the virtual environment.

[0075] S3: Integration of obstacle avoidance algorithms;

[0076] The integration of obstacle avoidance algorithms is one of the key technologies in the agent virtual simulation system. As Figure 4The core idea of obstacle avoidance and collision avoidance is to determine a conflict area and try to stay away from it, which ensures that the agent can effectively perform autonomous navigation in the face of complex environments and diverse sources of resistance. In the Unity3D environment, this implementation uses the A* search algorithm as the core obstacle avoidance mechanism, and optimizes the algorithm to meet the dynamic obstacle avoidance needs of the agent by adjusting the heuristic function and cost evaluation strategy. This includes optimizing the preprocessing stage of the algorithm, such as identifying possible obstacles in advance and evaluating the passability of the environment, so as to reduce the computational burden of the agent in the simulation environment.

[0077] When adjusting the heuristic function and cost evaluation strategy, in order to further avoid collisions between multiple agents, the repulsive force field in the artificial potential field is considered to be introduced. The repulsive force frep can represent the gradient of Urep, and the repulsive force relationship between agent i and agent j in the x-axis direction can be expressed as the following formula relationship.

[0078]

[0079] In formula (1), η is the repulsive force scale factor, which is taken as 10 in the simulation, l ij is the distance between agent i and j, and Rdetect is the influence radius of the repulsive force field; when the agents are far apart, there is no repulsive force influence between them. Similarly, the repulsive force relationship between multiple agents in the y-axis direction can be obtained. Finally, the expected velocity components of the i-th (i>0) agent at time t in the xyz-axis directions in the inertial coordinate system are expressed as the following formula:

[0080]

[0081] In formulas (3)-(4), is the expected velocity component of the i-th agent at time t in the x-axis direction, is the expected velocity component of the i-th agent at time t in the y-axis direction, is the expected velocity component of the i-th agent at time t in the z-axis direction, and Δt is the interval of system state update.

[0082] For two-dimensional motion, only the xy axes are considered. This implementation writes the operations for the z-axis following formulas (3) and (4) as follows:

[0083]

[0084] The path search process of the A* search algorithm is refined. By integrating the sensor data of the agent into the physics engine of Unity3D, the algorithm can obtain the relative position and distance information between the agent and surrounding obstacles in real time. This data is used to dynamically update the search tree of the A* search algorithm, enabling the agent to quickly respond to environmental changes during movement and immediately adjust the path to avoid newly emerging obstacles. In addition, the algorithm also takes into account the dynamic characteristics of the agent, such as the maximum steering angular velocity and acceleration limits, ensuring that the generated obstacle avoidance path is not only safe but also feasible.

[0085] At the implementation level of Unity3D, the A* search algorithm is tightly integrated with the control script of the agent. When the agent receives a path planning request, the A* algorithm module is activated and starts calculating the optimal path from the current position to the target position. The output of the algorithm is a path containing key turning points and speed suggestions, and the agent adjusts its motion state according to this information to achieve smooth and efficient obstacle avoidance navigation. In this way, the actions of the agent in the virtual environment are more similar to real-world behavior patterns, improving the overall authenticity and reliability of the simulation system.

[0086] S4: Integration of artificial intelligence training interface;

[0087] The integration of the artificial intelligence training interface is an innovation point in the virtual simulation system, which provides the agent with the ability to learn and self-optimize. In the Unity3D environment, this embodiment implements a flexible artificial intelligence training interface through the MLAgent plugin, which allows researchers to design and implement complex training scenarios. As Figure 5 shown in the data flow diagram, MLAgent, as an intermediate layer, receives real-time data from the Unity environment, such as the position of the agent, speed, distance to surrounding obstacles, etc., and converts this data into training signals to guide the behavior adjustment of the agent.

[0088] Through the integration of TensorFlow, researchers can design deep learning models to process the large amount of data collected by the agent. These models are trained with the assistance of MLAgent to learn how to predict the best action plan based on environmental feedback. For example, when the agent encounters a source of resistance, the AI model can identify the obstacle avoidance pattern through previous training and automatically adjust the path planning strategy of the agent to avoid collisions and maintain the smooth execution of the task. This training mechanism enables the agent to gradually adapt to the uncertainty and complexity in the virtual environment, improving its autonomous decision-making ability.

[0089] The agent in Figure 6During the shown life cycle, the behavior of the agent is continuously optimized through interactions with the virtual environment. MLAgent provides rich sensor and actuator APIs, enabling the agent to perceive environmental changes and respond. Researchers can set up reward mechanisms to guide the learning process of the agent. For example, by rewarding the agent's behavior of successfully avoiding obstacles or efficiently completing specified tasks. As the training progresses, the control strategy of the agent gradually matures, being able to handle unknown situations more intelligently, showing human-like learning abilities and adaptability. This integrated artificial intelligence training interface greatly enhances the agent's navigation and operation capabilities in complex environments, providing a solid technical foundation for further research and application of the agent.

[0090] In summary, the present invention provides a virtual simulation solution for optimizing the motion control and cooperative operation efficiency of agents with low cost and high security, integrating the improved A* search algorithm and artificial intelligence training interface, providing an economical and efficient virtual test platform for the research and development of agent control algorithms.

[0091] The above are only the preferred embodiments of the present invention and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the equivalent embodiments by using the above-disclosed technical content within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention and is based on the technical essence of the present invention, any simple modification, equivalent replacement, and improvement of the above embodiments still fall within the protection scope of the technical solution of the present invention.

Claims

1. A multi-agent collaborative combined navigation virtual simulation system based on Unity3D, characterized in that: The structure of the multi-agent collaborative combined navigation virtual simulation system based on Unity3D includes: Presentation layer, business layer and underlying services; The presentation layer provides an operation and monitoring platform for users; The business layer is used for control logic, queue management, path planning and artificial intelligence training functions of the intelligent body in the simulation environment; The underlying service is used for intelligent machine learning model training and project version management.

2. According to the Unity3D-based multi-agent collaborative combined navigation virtual simulation system of claim 1, it is characterized in that: The presentation layer includes an environment simulation module, a UnityUI interface and a Unity Editor window; The environment simulation module is used to construct a virtual environment required for agent interaction; The UnityUI interface provides an operation page for users; The Unity Editor window provides a monitoring platform for users.

3. The multi-agent collaborative combined navigation virtual simulation system based on Unity3D according to claim 1 is characterized in that: The business layer includes an agent control module, an agent queue control module, a path planning module and an AI environment module; The Agent control module is responsible for the control of a single agent, which includes the management of power data, torque data, path data, follow-up data, mode control and algorithm selection; The Agent queue control module is used to manage the data, target data and queue control of the agent queue; The path planning module is used to calculate the optimal path for the agent to reach the target location from the current location; The AI ​​environment module is used to perform artificial intelligence training on the agent and predict the best action plan based on environmental feedback.

4. The multi-agent collaborative combined navigation virtual simulation system based on Unity3D according to claim 1 is characterized in that: The Agent control module includes a single Agent control module and a multi-Agent control module; The single agent control module is used for intelligent body power control, target path planning and action state control; The multi-agent control module is used for queue mode switching, agent queue management and queue motion simulation.

5. The multi-agent collaborative combined navigation virtual simulation system based on Unity3D according to claim 1 is characterized in that: The underlying services include the Tensorflow training interface service module and the Plastic SCM project version control module; The Tensorflow training interface service is used to convert sensor collected data into training signals to guide the behavior adjustment of the intelligent agent; The Plastic SCM project version control is used to perform project version management.

6. A multi-agent collaborative combined navigation virtual simulation method based on Unity3D, applied to a multi-agent collaborative combined navigation virtual simulation system based on Unity3D as described in any one of claims 1 to 5, characterized in that: include: Step 1: Building a Unity3D virtual environment required for agent interaction based on the environment simulation module; Step 2: Combining the single agent control module, the path planning module and the multi-agent control module, setting the control strategy of the intelligent agent in the Unity3D virtual environment through a C# script; Step 3: Use the improved A* search algorithm and RVO2 simulator idea, combined with the control strategy of the intelligent agent, to control the multi-agent to avoid obstacles; Step 4: Train and deploy the AI ​​model in the Unity3D virtual environment through the AI ​​environment module to control the intelligent agent. By connecting to the Tensorflow training interface service, use the MLAgent plug-in to train the control strategy of the intelligent agent so that the intelligent agent can successfully avoid obstacles and complete the multi-agent collaborative combined navigation task.

7. The multi-agent collaborative combined navigation virtual simulation method based on Unity3D according to claim 6 is characterized in that: Step 1 specifically includes: Step 1.1: Use computer vision technology to extract information about all objects in the navigation environment from multi-angle images, photos, and video materials of the objects, and use one of the methods of structured light scanning, stereo vision, and motion recovery structure to calculate and construct 3D models of all objects; Step 1.2: Import the 3D models of all objects into the Unity3D environment, and optimize the 3D models of all objects during the import process. The optimization of the 3D models of all objects includes simplifying polygons of the 3D models of the objects, optimizing textures and materials, and adjusting the scale and proportion of the models; Step 1.3: Place the optimized 3D models of all objects at the corresponding positions in the Unity3D environment, and set the lighting, climate effects and physical properties in the Unity3D environment to complete the construction of the Unity3D virtual environment.

8. The multi-agent collaborative combined navigation virtual simulation method based on Unity3D according to claim 6 is characterized in that: Step 2 specifically includes: Step 2.1: Use Unity3D's sensors and physics engine to make the agent perceive changes in the surrounding environment, and collect data on the agent's position, speed, and distance to surrounding objects through the integrated API; Step 2.2: The path planning module uses the improved A* search algorithm to analyze the Unity3D virtual environment data, and combines the collected agent position, speed, and distance to surrounding objects to calculate the optimal path for the agent to reach the target position from its current position; Step 2.3: Through the single-agent control module and the multi-agent control module, multiple agents are enabled to coordinate their actions to maintain the formation according to the preset formation and dynamic adjustment strategy, and the surrounding environment of the agents is monitored in real time. When a potential collision risk is detected, the optimal obstacle avoidance path is calculated and the speed and direction of the agents are adjusted to complete the setting of the agent control strategy.

9. The multi-agent collaborative combined navigation virtual simulation method based on Unity3D according to claim 6 is characterized in that: Step 3 specifically includes: Step 3.1: After receiving the path planning request, the agent activates the integrated control script of the A* algorithm module and the RVO2 simulator idea to calculate the optimal path from the current position to the target position; Step 3.2: Calculate the repulsive force relationship between agent i and agent j in the x-axis direction; Step 3.3: Combined with step 3.2, calculate the repulsive force relationship between agent i and agent j in the y-axis and z-axis directions; Step 3.4: Superimpose the repulsive force relationship between agent i and agent j in the xyz axis direction to obtain the expected velocity component of the i-th agent in the xyz axis direction in the inertial coordinate system at time t; Step 3.5: Dynamically update the search tree of the improved A* search algorithm based on the data collected by Unity3D's sensors, and adjust the maximum steering angular velocity and acceleration of the agent to ensure that the agent moves along the optimal obstacle avoidance path and avoids obstacles; The expression of the repulsive force relationship between agent i and agent j in the x-axis direction is: In formula (1), η is the repulsive force scale factor, l ij is the distance between agents i and j, Rdetect is the influence radius of the repulsive field; The calculation formula for the expected velocity component of the i-th agent in the xyz axis direction in the inertial coordinate system at time t is: In formulas (3)-(5), is the expected velocity component of the ith agent in the x-axis direction at time t, is the expected velocity component of the ith agent in the y-axis direction at time t, is the expected velocity component of the ith agent in the z-axis direction at time t, and Δt is the interval for updating the system state.

10. The multi-agent collaborative combined navigation virtual simulation method based on Unity3D according to claim 6, characterized in that: Step 4 specifically includes: Step 4.1: In the Unity3D virtual environment, receive real-time data from sensors through the MLAgent plug-in, where the real-time data includes the position, speed, and distance of surrounding obstacles of the agent, and convert the real-time data into training signals; Step 4.2: By connecting to the Tensorflow training interface service, the AI ​​model is trained with the assistance of the MLAgent plug-in, predicting the best action plan based on sensor data feedback to complete the multi-agent collaborative combined navigation task.

Images