Vesoma
“VioLA's core innovation is using pretrained motion encoders that map both human and robot motion into the same latent space, allowing human video recordings to directly supervise a robot policy's action outputs.”
Source→“A human recording is therefore labeled in the policy's action space, and the training demonstration pool contains 140.6 million frames, 93.2% of them human”
Source→“GR00T N1.7 and Ψ0 reach 16.7% and 0%, respectively for locomotion; GR00T N1.7 completes 0/35 trials and Ψ0 completes 3/35”
Source→“Founder of SMPL/MANO body and hand models, which are foundational to VioLA's human motion representation.”
Source→“His presence at Vesoma suggests the company is pursuing humanoid control seriously.”
Source→“Head of the Institute for Machine Learning, one of Europe's leading ML groups. His involvement signals serious academic backing.”
Source→“Lead author, ETH Zürich / MPI-IS / Vesoma. Conducted work during internship at Vesoma. Co-lead on the project.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.