The paper presents an interactive robot framework that uses language models for long-horizon task planning. This system can adjust to new goals or tasks during execution, without requiring complex module designs or prompt engineering. The framework incorporates high-level planning and low-level function execution via language. It’s robust at generating new high-level instructions for unseen objectives and can adapt to different tasks by merely changing the task guidelines. It can also replan based on new user requests and previously executed steps.
Publication date: 18 Oct 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2310.10645
