Robot training fix: cutting noise injection from 2.5% to 0.25% of joint range removes jerkiness

DominiqueCAPaul · x · 2026-09-28

Developer DominiqueCAPaul traced jerkiness in robot learning to over-aggressive noise injection: sigma was 2.5% of each joint's range, far too high. Reducing it to a tenth made training noticeably better — learning is still slow and the robot does dumb things, but success-rate curves across rollouts now show a clear learning effect.

Original post →

More from Embodied

Embodied channel →