Hang Yang

A history-conditioned, physics-informed flow-matching framework for probabilistic dynamics prediction, using motion and actuation histories to predict the distribution of the next complete joint configuration and recursively sampling next-step configurations under prescribed commands.

GIFT reduces depth hallucinations in non-Lambertian surfaces without needing depth labels, enhancing monocular depth models' practical utility.

TC-SOH reduces prediction errors in battery health estimation by over 50% while enhancing model transparency through innovative representation diagnostics.