Player-1: A post-training framework for training game agents with synthetic experience
What happened
According to an arXiv paper, Mingyu Ouyang et al. propose Player-1, a post-training framework that lets game agents learn game operations beyond the scope of their demonstration data through synthetic experience. The paper reports that on the MineDojo-300 benchmark, Player-1-9B and Player-1-35B-A3B improve over their respective SFT (supervised fine-tuning) policies by 27.8 and 17.7 percentage points. Player-1-9B reaches a 38.33% success rate, while the strongest baseline reaches 12.63%. The paper says the team built Player-1-Gym, which contains 16 synthetic 2D and 3D environments designed around the shared control requirements of various games, for generating training experience. The paper also reports further gains over SFT on ViZDoom-10 and GameWorld. All of the above results are self-reported by the paper.
AI-written from the coverage · updated 2026-10-15
Coverage timeline
Follow the coverage to see the story from different sides.
- arXiv Game AIPickPlayer-1: Learning to play from synthetic experience
Player-1 is a post-training framework that uses synthetic experience to let game agents learn game operations beyond the scope of their demonstration data. On MineDojo-300, Player-1-9B and Player-1-35B-A3B improve over their respective SFT policies by 27.8 and 17.7 percentage points. Player-1-9B reaches a 38.33% success rate, while the strongest baseline reaches 12.63%.