Xin Yu, Haoyuan Li, Yandong Wang, Simin Li, Rongye Shi, Wenjun Wu
Submitted to IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025
Goal assignment is a critical challenge in multi-robot systems. The emergence of large language models has enabled the use of natural language commands for tackling goal assignment problems. However, applying LLMs directly to these tasks presents two limitations: 1) limited accuracy and 2) excessive decision delays due to their autoregressive nature, hindering adaptability to unexpected changes. To address these issues, inspired by dual-process theory, we propose a framework called Collaborative LLMs for dynamic Goal Assignment (CLGA). Specifically, we leverage LLMs for pre-planning tasks and invoke an external solver to generate an initial goal assignment plan, ensuring solution accuracy. During execution, distributed small-scale models enable real-time adjustments to respond to dynamic environmental changes. This approach integrates the strengths of slow, precise pre-planning and fast, adaptive online adjustments, allowing agents to efficiently handle real-world challenges. Additionally, we introduce a benchmark dataset for multi-robot goal assignment to advance research in this domain. Simulation and real-world experiments demonstrate that CLGA significantly enhances task execution efficiency and flexibility in multi-robot systems.
To further verify the performance of the CLGA framework in actual multi-robot mission scenarios, we designed and implemented four sets of real robot experiments, covering dynamic task reallocation, real-time response after target addition, emergency replacement after agent failure, and dynamic planning in navigation tasks. These experiments intuitively demonstrated the flexibility and efficiency of CLGA in dealing with emergencies and dynamic changes in real environments, further verifying the practical applicability and robustness of this method.
The video shows that when the original intelligent agent fails, the system quickly activates the backup intelligent agent and re-plans the task allocation in real time through the CLGA framework to ensure that the multi-robot team can complete the task efficiently and stably.
The video shows that when a new target appears, the system immediately activates a new agent and quickly plans the path and assigns tasks through the CLGA framework to ensure that the agent reaches the target in time and completes the task.
The video shows that after the original intelligent agent fails in the navigation task, the system quickly activates the backup intelligent agent and re-plans the task allocation in real time through the CLGA framework to ensure the efficient completion of the overall navigation task.
The video shows that when a new target appears in the attack and defense scenario, the system immediately dispatches a new intelligent agent, generates an interception strategy in real time through the CLGA framework, and quickly goes to the target location for effective interception.
The following animation shows the learning behavior of CLGA in different scenarios mentioned in the paper.
2D_Navigation_Emergency_Situaion1
2D_Navigation_Emergency_Situaion2
2D_Defense_Emergency_Situaion1
2D_Defense_Emergency_Situaion2
3D_Navigation_Emergency_Situaion1
3D_Navigation_Emergency_Situaion2
3D_Defense_Emergency_Situaion1
3D_Defense_Emergency_Situaion2
The following picture shows the prompt word templates actually used by each module mentioned in the framework of CLGA in the paper.
Contact
If you have any question please contact nlsdeyuxin@buaa.edu.cn