Reproduce “LLM-based Robot Task Planning with Exceptional Handling for General Purpose Service Robots” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ruoyu Wang, Zhipeng Yang, Zinan Zhao, Xinyan Tong, Zhi Hong, Kun Qian

Researchers developed a robust task planning method for general-purpose service robots that leverages large language models (LLMs) to interpret natural language commands and generate executable action sequences. Their system, employing a constrained prompting scheme and an exceptional handling module, achieved 83% correct action plans and successfully completed real-world tasks, mitigating issues like LLM hallucinations and unexecutable outputs.

Key insights

  • Constraining LLM outputs with predefined primitive actions and specific formatting through multi-faceted prompting significantly enhances the executability and correctness of generated robot task plans.
  • A reactive exceptional handling module can effectively mitigate LLM hallucinations by detecting and correcting format deviations or the generation of non-existent actions, improving system robustness.
  • Different LLMs exhibit varying performance in robotic task decomposition, with models like ERNIE-Bot 4.0 showing superior accuracy, especially for complex interaction tasks, though all struggle with visually-dependent questions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.