Engineer Ports NVIDIA's Open-Source SONIC Model to AGIBOT X2 in 6 GPU-Hours
Scobleizer · x · 2026-11-03
A financial engineer in San Francisco ported NVIDIA's open-source SONIC whole-body control model to the AGIBOT X2, a robot body it was never trained for. His pipeline: train thousands of virtual X2s in parallel in Isaac Lab, re-test in MuJoCo as a second physics engine, then deploy on the real robot. Early policies fell within 6 seconds in sim; the fix was sphere colliders for the feet plus the right rotation ordering. He also scanned his garage into the sim so X2 could rehearse in a digital twin. A month later he transferred a motion planner trained 2 million steps on other platforms to the X2 — training from scratch would take 10,000 H100 GPU-hours, but the transfer took 6 hours on a single workstation GPU. The robot now walks, dances, and mirrors his full-body motion.
More from Embodied
- Xiaomi open-sources Robotics-1, a robot foundation model trained on 100K hours of trajectories — CyberRobooo · 2026-11-03
- Factory photos show desktop robot modules in volume assembly, project hits production stage — victormustar · 2026-11-03
- Machenta demos instant robot learning from passersby at London King's Cross — Ed__Johns · 2026-11-03
- PRIMA animal 3D mesh recovery accepted to WACV2027 with 9K pseudo-3D dataset — TrackingActions · 2026-11-03
- Robotics startup Dimensional hosts SF warehouse party showcasing SOTA robots — Scobleizer · 2026-11-03
- Video appears to show a fleet of Figure F.03 humanoid robots walking across a campus — TechPreacher · 2026-11-03