Lirui Zhao

UR-VC: Unsupervised Robotic Value Correction for Time-Derived Progress Proxies

Lirui Zhao, Modi Shi, Li Chen, Qi Liu, Ping Luo, Hongyang Li

Modern robot learning systems increasingly rely on dense progress or value signals to evaluate intermediate states, guide policy learning, and detect task completion, making the quality of these signals critical. Since such dense labels are rarely available at scale, normalized t…Read more