Why humanoids keep crashing into barriers: onboard vision can't keep up with sprinting legs

2C_ornot2C · x · 2026-08-19

Why are fast humanoids running into barriers and drifting all over the track this year, when it wasn't obvious last year? The big change is autonomy: last year most high-speed runs were remotely controlled, while this year more robots actually see the track and decide where to run.

The problem is that hardware has outpaced the perception stack. Some humanoids already sprint at or beyond human speed in a straight line, but on curves onboard vision must recognize curvature, estimate position and yaw, predict the track, and adjust foot placement before the body arrives. At 8–10+ m/s, a 30 fps camera yields a frame only every 33 ms — the robot travels 25–35 cm between frames, before adding image processing and state-estimation latency.

Original post →

More from Embodied

Embodied channel →