reinforcement-learning

October 2, 2026

• Ataraxos beats a top Stratego player with a small training budget

  A Nature paper describes Ataraxos, an AI that won 15 of 20 games against four-time world champion Pim Niemeijer. The system combines self-play, a belief model for hidden pieces, and test-time search, and cost a few thousand dollars to train.

  researchreinforcement-learninggamesagents