galbenecho/echo-foundry-vla

Echo Foundry — VLA Dataset TL;DR 10,133 episodes (60 h) of physics-verified robot manipulation where every tick explains itself — a reasoning sentence with live measured distances on all 4.3M frames, every object's ground-truth pose, metric depth, typed outcomes with the evidence attached, and the planner's rejected alternatives with the margins that killed them. One robot (UR5e), five tabletop tasks, simulation registered to a real robot cell whose calibration… See the full description on the dataset page: https://huggingface.co/datasets/galbenecho/echo-foundry-vla.