The authors extended a computer-use agent with a shared action space that includes relative pointer motion, sustained inputs, and composite actions.
Synthetic experience improves AI game control beyond supervised training
Training on 16 synthetic environments boosts Player-1-9B's MineDojo success from 10.5% to 38.3%, with gains transferring to unseen games
Chinese Tech
Mingyu Ouyang · Yiran Wang · Siyuan Hu · Xingyilang Yin · Huiqiang Sun · Guangzhi Wang · +6 more
National University of Singapore · Tencent IEG · Nanyang Technological University
Research Digest··2 min read
Thread:Synthetic Environment RL
The authors introduce Player-1, a post-training framework that first fine-tunes a vision-language model on 6,044 curated gameplay demonstrations, then applies reinforcement learning in 16 synthetic game environments designed to cover common control demands.
Why this paper
From Tencent IEG and 2 others
In one line
Learning from synthetic experience in 16 environments improves game control beyond supervised initialization.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§