Mengzhou Wu, Yuzhe Guo, Yuan Cao, Haochuan Lu, Songhe Zhu, Pingzhe Qu, Xin Chen, Kang Qin, Zhongpu Wang, Xiaode Zhang, Xinyi Wang, Wei Dai, Gang Cao, Yuetang Deng, Zhi Gong, Dezhi Ran, Linyi Li, Wei Yang, Tao Xie
View original ↗Create an open-source sandbox environment that allows training GUI agents using physics-based feedback rather than just pixel mimicry. This will help bridge the gap between static datasets and real-world agent reliability.
Suggested repo: oceanus-sim
"Teach your agent physics, not just pixels, to master complex GUI interactions."
Estimated effort: 120h