Ran Gong

Efficient Sim-to-Real Transfer of World-Action Models from Synthetic Priors

Zixing Wang, Kausik Sivakumar, Jinghuan Shang, Yafei Hu, Zhaoming Xie, Ran Gong, et al.

Bridging the sim-to-real gap is a core challenge in deploying learned manipulation policies. Sim-to-real learning is attractive because it can replace expensive real robot demonstrations with scalable synthetic data, yet world-action models have not previously been shown to trans…Read more