Reproduce “Integrating Action Knowledge and LLMs for Task Planning and Situation Handling in Open Worlds” 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
Yan Ding, Xiaohan Zhang, Saeid Amiri, Nieqing Cao, Hao Yang, Andy Kaminski, Chad Esselink, Shiqi Zhang
The COWP framework, developed by researchers from SUNY Binghamton and Ford Motor Company, integrates classical task planning with Large Language Models to enable robots to dynamically handle unforeseen situations in open-world environments. It achieved a 67.8% task completion rate and 36.9% situation handling rate in dining-focused tasks, outperforming state-of-the-art baselines in simulations and a real-world mobile manipulator demonstration.
Key insights
- Combining the logical consistency and domain grounding of classical planning with the broad commonsense reasoning of LLMs effectively overcomes the limitations of both isolated approaches.
- Dynamically augmenting a robot's action knowledge (preconditions and effects) at execution time via LLM interaction is crucial for robust open-world adaptability.
- A systematically crowdsourced dataset of execution-time situations is essential for thoroughly evaluating and advancing open-world robot planning systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.