AI2's MolmoB0T: Zero-Shot Real-World Manipulation from 1.7M Simulated Trajectories
chris_j_paxton · x · 2026-10-20
AI2 and collaborators (UW, Princeton, UCLA) released the CoRL 2026 MolmoB0T work challenging the assumption that sim-to-real requires real-world data. Key points:
- Fully open-source MolmoBot-Engine pipeline for procedural data generation across robots, tasks, and simulated environments
- MolmoBot-Data: 1.7M expert trajectories for articulated-object manipulation and pick-and-place
- Three policy classes: MolmoBot (Molmo2 VLM + flow-matching action head), MolmoBot-Pi0 (π₀ replication for controlled comparison), and lightweight MolmoBot-SPOC for edge deployment
- Zero-shot transfer to the real world without any real-world fine-tuning, for both static and mobile manipulation
Models, data, and code are all open-sourced; a RoboPapers podcast episode is coming soon.
More from Embodied
- Ex-NVIDIA team's Dream Labs launches Dream-1, zero-shot control of unseen bimanual robots — gharik · 2026-10-20
- Dream-1 claims new Pareto frontier for robot control: 16x cheaper, 1.7x faster, 59 vs 48 on RoboDojo — scott_e_reed · 2026-10-20
- Surgeons Complete 100+ Cataract Surgeries Using Apple Vision Pro and ClearSphere App — asymco · 2026-10-20
- What makes human video generation good for robot manipulation? A constrained-optimization analysis — mangahomanga · 2026-10-20
- Figure's F.03 humanoid walks 348 miles and 1.2M steps over 9 straight days on a treadmill — adcock_brett · 2026-10-20
- NVIDIA joins Gemma hardware hackathon with 30 Jetson kits at AGI House on Oct 24 — agihouse_org · 2026-10-20