Preprint
Oct 2026
Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning
Heteroscedastic Student-t action regression (HT-Policies) is introduced, which learns input-dependent residual scales and reduces the influence of heavy tails and offers an efficient direct-regression alternative for a range of architectures and tasks.
Yu-Chen Zhou, Jia-Cheng You, Wei-Kang Wan et al.
· 0 citations