LLM-assisted Task Decomposition and Generation for Autonomous Mission Planning

In multi-robot systems, task planning is essential for operating multiple robots simultaneously, and a task assignment algorithm assigns tasks to each robot and determines the execution order. This algorithm requires tasks and their specifications as input, and the process of generating these is called mission planning. In real-world environments, mission planning is typically performed manually by operators, which is inefficient and workload-intensive. To address this issue, recent studies have focused on autonomous mission planning using large language models (LLMs) to automate mission planning. In this study, we developed an LLM-based autonomous mission planning system to effeciently operate a multirobot system. This mission planning framework utilizes an LLM to decompose high-level missions into subtasks, generate specifications, and transform them into a format that can be processed by task assignment algorithms. Moreover, it employs a four-stage Actor-Critic structure to improve mission analysis accuracy and consider task dependencies. Additionally, an optimization-based algorithm is utilized for task assignment and scheduling, mitigating the mathematical limitations of the LLM. Specially, we apply the designed framework to a multi-robot system deployed in disaster scenarios, solve task assignment and scheduling, and validate its performance through 3D simulations.