FACE: contact-factorized imitation learning lets one policy handle unseen friction and geometry

KyleMorgenstein · x · 2026-11-17

Researchers from Kyung Hee University, Ewha Womans University, and UCF introduce FACE, a contact-factorized imitation learning framework that separates intended task behavior from environment-dependent contact factors. A learned contact-normal estimator and an online friction estimator let a single fixed policy adapt to unseen surface geometry and friction (wrinkled surfaces, PVC tape, etc.) on real robots, without updating policy weights. Paper on arXiv; code coming soon.

Original post →

More from Embodied

Embodied channel →