Reproduce “Stochastic Games for Interactive Manipulation Domains” 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
Karan Muvvala, Andrew M. Wells, Morteza Lahijanian, Lydia E. Kavraki, Moshe Y. Vardi
Researchers at the Toyota Research Institute and Rice University developed a framework using stochastic games to synthesize robot strategies for interactive manipulation tasks. This approach models human-robot interactions by accounting for strategic human behavior and uncertainties in robot actions, leading to strategies that maximize the probability of task completion, demonstrated with substantial computational improvements and in a physical Tic-Tac-Toe game.
Key insights
- Stochastic games effectively model both strategic and probabilistic aspects of human-robot interaction, accommodating a spectrum of human behaviors without restrictive assumptions.
- The proposed computational improvements in model construction, specifically bypassing PRISM-games' built-in methods, are crucial for practical applicability in robotics.
- Applying the framework to a physical Tic-Tac-Toe game reveals emergent strategic robot behaviors, such as exploiting human errors and creating forks, without explicit programming.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.