Search papers, labs, and topics across Lattice.
University of Illinois Urbana-Champaign
3
0
5
5
LLMs can scalably annotate motion capture data to produce semantically rich descriptions of bimanual interactions, enabling higher-quality generation of dexterous hand motions.
Achieve expert-level imitation learning with just a handful of state-only expert demonstrations by pre-training a dynamics-aware latent space.
Humanoids can now perform complex loco-manipulation tasks from egocentric vision and sparse goals, thanks to a unified controller trained without relying on predefined motion references at test time.
