Astribot Releases Lumo-2 Robotics Model
rohanpaul_ai · x · 2026-07-18
Astribot released Lumo-2, a 4B robotics foundation model. Its core approach is predicting task-relevant "future physical states" before deciding on actions. The author clarifies it doesn't generate full future videos but only models task-useful changes like object movement, contact changes, or pouring completion.
Results include:
- Overall inference speed improved by 2.71x
- Supports human videos and multiple robot embodiments
- Performs better on 105 unseen objects
- Achieves the best comprehensive score on 22 real-world manipulation tasks, covering motion prediction, memory, physical reasoning, long-horizon tasks, and fine hand control
These results were validated on real robots across 20+ complex household tasks, not just benchmark scores. Training involves three steps: aligning visual changes with robot actions, establishing semantic correspondence between action tokens and language/vision, and finally jointly training the VLM, video, and robot demonstrations so the model predicts the next physical outcome before outputting actions.
Related event: Astribot launches Lumo-2 with real-robot demos(10 posts)→
More from Embodied
- Embodied AI startups need a different operational discipline, investor says — ___Mufasaa · 2026-10-27
- π0.5 study: raising demos from 10 to 438 lifts Place Ball in Bowl success from 7% to 83% — stingning · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- HandsOn benchmark: 66 verified tasks and 18,000 human demos for dexterous manipulation — stingning · 2026-10-27
- DexPort: a handheld interface with a real robot hand for dexterous-hand data collection — pabbeel · 2026-10-27
- ROS names Alejandro Hernández Cordero new project lead, eyes 2027 release — ___Mufasaa · 2026-10-27