Enhancing Robot Task Planning: Integrating Environmental Information and Feedback Insights through Large Language Models

Utilizing knowledge derived from large language models (LLMs) has been established as an effective strategy for task planning and providing action plans for agents. In this paper, we put forth EnviroFeedback Planner, a novel approach for generating action plans with LLMs. Specifically, EnviroFeedback Planner integrates environmental information into prompt construction and considers available actions, introducing feedback corrections to enhance the agent's execution capabilities. To validate our proposed method, we systematically conducted experiments in the Virtualhome environment, comparing it against baseline methods. Compared to baseline methods, the action plans generated by EnviroFeedback Planner exhibit a 26.46% improvement in executability and an 11.06% enhancement in correctness.