HiP-CL: High-Level Planning Based on Common Language for Multi-Robot Teaming

Coordinating a multi-robot team using natural language presents significant challenges due to the inherent ambiguity of language and the complexity of multi-agent tasks. To fully harvest the structural and semantic richness of natural language, this paper proposes a common-language based framework that can generate comprehensive plans for multi-robot missions. The framework enables effective multi-robot planning by handling spatial and resource constraints through a dual-stage architecture: (1) common language translation, where a supervised fine-tuned language model eliminates linguistic ambiguities by converting natural-language instructions into a structured common language that we defined; and (2) multi-robot task generation, which employs a semantic-driven approach to produce deterministic plans while resolving semantic constraints. By integrating large language models, the framework provides high flexibility in parsing linguistic inputs and generating deterministic plans. Compared to Large Language Model (LLM) based approaches across three different corpora, the framework achieves superior executability and improved performance 65.66% in simple corpus and 70.00% in complex corpus.