The Gaussian is enough: Toyota study finds non-Gaussian priors don't help fine-tuning LBMs
_krishna_murthy · x · 2026-09-25
A negative-result study from Toyota Research Institute, Woven by Toyota and Cornell ("The Gaussian Is Enough") shows that closer-to-target non-Gaussian action priors—which help when training imitation policies from scratch—offer no benefit when fine-tuning pretrained Large Behavior Models. Across 100K+ simulation rollouts on three LBM architectures (LBM 1.0, π0.5, GR00T N1.5), 40+ tasks, and 1250 hardware rollouts on five bimanual tasks, the standard Gaussian prior performed statistically indistinguishably or better, with a possible exception at very low fine-tuning data fractions. Diagnostic analyses indicate the observation encoder, not the action prior, governs fine-tuning performance.
More from Embodied
- BracketBot robot autonomously navigates via video-built maps to scan shelves — broodsugar · 2026-09-25
- Smart light fixtures as an always-powered AI sensor platform for the home — K-enthusiast24 · 2026-09-25
- Meta is putting its Muse agent into AI glasses, aiming to make the interface disappear — TansuYegen · 2026-09-25
- Agents turn stock ESP32 kits into anything: why familiar forms guide what to build — genmon · 2026-09-25
- FANUC cobot takes plain-language commands, trained in Isaac Sim before hardware existed — lukas_m_ziegler · 2026-09-25
- AR Yu-Gi-Oh built on Snap's Specs glasses uses AI to battle with physical cards — TheZachMueller · 2026-09-25