Astribot releases Lumo-2 tech report: three-stage aligned training, label-free skill transfer
CyberRobooo · x · 2026-09-30
Astribot published the technical report for Lumo-2, its next-gen Latent World-Action Model, with three core upgrades over Lumo-1:
- Implicit predictive reasoning in a physically grounded latent space replaces explicit textual planning, enabling world-model-style control without dense pixel generation
- A three-stage progressive pre-alignment paradigm (action-world dynamics, then vision-language, then co-training on VLM, video, and robot data) replaces single-stage joint training
- Native multi-source data support: Lumo-2 learns from egocentric human videos and multi-robot demonstrations with or without action labels, making human-to-robot and cross-robot skill transfer more scalable
The model shows consistent gains on temporal reasoning, physical understanding, and long-horizon dexterous manipulation tasks.
Related event: Astribot Debuts Dual-Robot Collaboration and Lumo-2 at IROS(2 posts)→
More from Embodied
- Duke's MARBLE sealed sphere robot rolls by shifting internal weight, works on land and water — lukas_m_ziegler · 2026-09-30
- Astribot demos two humanoid robots collaborating with its Lumo-2 world-action model — CyberRobooo · 2026-09-30
- Fully open-source VLA MolmoAct2 lands CoRL 2026 Spotlight, matching π0.5 performance — jmin__cho · 2026-09-30
- Apple's HomePad smart home hub reportedly launches October 13th alongside new HomePod mini and Apple TV — tomwarren · 2026-09-30
- Image Lab Tops Robot Leaderboard: FLUX 3 Action Runs on a Desk-Size Arm — lmoroney · 2026-09-30
- Open-source self-driving guide adds VFH: binning LiDAR into 5° obstacle histograms — 4310sy · 2026-09-30