CORTEXA
← Browse

Kang Chen

2 papers indexed

arxivcs.RO2026-06-29

STEAM: Self-Supervised Temporal Ensemble Advantage Modeling for Real-World Robot Learning

Zhihao Liu, Qiuyi Gu, Yitao Wang, Dongming Qiao, Yixian Zhang, Shuaihang Chen, et al.

Real-world robot learning increasingly relies on heterogeneous data, but demonstrations and rollouts often mix useful progress with stalls, corrections, and suboptimal behavior. Effective policy learning therefore requires frame-level advantages that distinguish reliable local pr…

View free PDFSource page