Machenta robot reacts and recovers in real time, trained 100% in simulation
Ed__Johns · x · 2026-10-28
Machenta's Instant Robot Learning shows one real demo (placing a blue object in a grey box and closing it) is enough for the robot to react in real time to moved objects/boxes, re-grasp dropped items, restore removed objects, and re-close opened lids — via a closed-loop end-to-end policy. None of this reactivity was demonstrated; it emerges from the pre-trained model, which was trained 100% in simulation and had never seen these objects before.
More from Embodied
- VLA Insiders: Fast Policies Are 90% About Skipping Teleop Data — Kinematics Helps Too — andrew_n_carr · 2026-10-28
- UniPred, accepted at IEEE T-RO 2026, invents task concepts from demonstrations for long-horizon robot planning — tomssilver · 2026-10-28
- A 1-inch robot roach that crawls into pipes and smells rust — Scobleizer · 2026-10-28
- Ex-Apple team launches Silica: joint actuators 50x more sensitive, pre-orders open — ycombinator · 2026-10-28
- edwinarbus refuses to wear AI glasses on the train: "Don't look at me" — edwinarbus · 2026-10-28
- Engineer wires Claude Code to Altium with ~600 tools — it designs a full 4-layer PCB from a spec — Salitronic · 2026-10-28