Reproduce “Superior Computer Chess with Model Predictive Control, Reinforcement Learning, and Rollout” 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

Atharva Gundawar, Yuchao Li, Dimitri Bertsekas

Researchers from Arizona State University developed MPC-MC, a meta-algorithm integrating Model Predictive Control and Reinforcement Learning principles to enhance the performance of any existing computer chess engine. This framework consistently improved various engines, including top-tier ones like Stockfish, by applying a theoretically grounded one-step lookahead for move selection.

Key insights

  • Model Predictive Control and Reinforcement Learning principles, specifically the exact execution of the first lookahead step, can provide an additional layer of intelligence to even state-of-the-art chess engines.
  • A meta-algorithmic approach, which treats powerful existing AI systems as components, can significantly enhance overall performance through an overarching, theoretically guided control layer.
  • The successful integration of classical control theory with modern deep learning-based evaluators validates the potential of hybrid AI systems for complex sequential decision-making.

Summary by alphaXiv.

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