Announcement
OmniGuide steers frozen robot AI in 3D, thoma_gu says
According to thoma_gu, OmniGuide steers a frozen robot model using vision-language models, geometry and human guidance.
TLDR
OmniGuide applies test-time guidance to diffusion-based vision-language-action models, thoma_gu says, converting guidance from vision-language models, geometry and humans into a differentiable 3D energy function that steers action sampling. The model stays frozen; the guidance does the steering. KostasPenn says TimSong52005757 will present poster #32 at CoRL 2026, with thoma_gu listing November 9, 5–6:30 p.m.
Combined views
—
3 Sources, first seen ago
