To efficiently assign natural language commands to multiple robots, we propose a framework that leverages each robot's unique field knowledge to decompose and assign tasks. Leveraging a large-scale language model (LLM) and spatial concepts, we develop a novel, small-shot prompting strategy that infers the required objects from ambiguous commands and decomposes them into appropriate subtasks. Experimental results demonstrate that the proposed method outperforms alternatives, successfully performing task decomposition, assignment, sequential planning, and execution using a real mobile manipulator.
