Live page · Day archive

World Action Learning via Interaction-Centric Spectral Latent Guidance

Read the original at HF Daily Papers

Summary

Researchers introduce WING, a framework that transfers interaction knowledge from egocentric videos to robot policies by isolating hand-object movements and aligning low-frequency temporal structures, achieving a 99.20% success rate on the LIBERO benchmark.

Carried by: HF Daily Papers. First seen: .