Reproduce “Structured LLM Reasoning for Zero-Shot Human--Robot Coordination Under Hidden Goals” 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
Dong Hae Mangalindan, Anand Gokhale, Francesco Bullo, Vaibhav Srivastava
A structured Large Language Model (LLM) architecture facilitates zero-shot human-robot coordination in cooperative construction tasks with hidden goals, demonstrating improved task efficiency and human trust compared to reinforcement learning baselines. This system integrates approximate inference and planning with rule-based physical verification, ensuring safe and effective collaboration.
Key insights
- Large Language Models can act as tractable surrogates for computationally intensive inference and planning within formal decision-theoretic frameworks.
- Structured decomposition of robot decision-making, including explicit Theory-of-Mind (ToM) for inferring human goals, enhances coordination efficiency and human trust in robot partners.
- Combining LLM-driven semantic reasoning with deterministic, rule-based physical verification is crucial for ensuring the safety and feasibility of robot actions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.