Same VLA, different data: Human Archive's 100-hour dataset beats rivals on task success

chris_j_paxton · x · 2026-10-17

A controlled VLA experiment pre-trained the same model on 100 hours of Human Archive's OTS dataset versus 100 hours each from microagi's and MeckaAI's datasets on EgoVerse.

Original post →

More from Embodied

Embodied channel →