The content discusses the use of large language models (LLMs) as high-level planners to resolve deadlocks in multi-robot systems (MRS). The authors propose a hierarchical control framework where an LLM is used to assign a leader and a direction for the leader to move in order to resolve deadlocks, while a graph neural network-based low-level distributed control policy executes the assigned plan.
The authors first extend their previous work on graph control barrier functions (GCBF) to incorporate connectivity requirements in addition to safety constraints. They then use GPT-3.5 as the LLM and explore various prompting techniques, including the use of in-context examples, to improve the LLM's performance in resolving deadlocks.
The authors conduct extensive experiments on various multi-robot environments with up to 15 agents and 40 obstacles. The results demonstrate that the LLM-based high-level planners are effective in resolving deadlocks in MRS. The authors also provide a detailed discussion on the impact of in-context examples and potential future directions to improve the performance of LLMs as high-level planners for assisting low-level controllers in complex MRS problems.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Kunal Garg,J... at arxiv.org 04-10-2024
https://arxiv.org/pdf/2404.06413.pdfDeeper Inquiries