Lumo-2 Speeds Up Inference with Block-Wise Decoding
rohanpaul_ai · x · 2026-07-18
This part supplements the inference optimization of Lumo-2: it uses block-wise decoding, generating several weakly related tokens at once instead of one by one for 32 action tokens.
On a single NVIDIA RTX 5090, Astribot reports latency reduced from 253.66ms to 93.53ms, achieving a 2.71x speedup.
The post also mentions that the model makes the action space a shared representation, making it easier to transfer skills between different robot embodiments.
Related event: Astribot launches Lumo-2 with real-robot demos(10 posts)→
More from Embodied
- LIBERO-MAX: Robot Policies Drop 11–26 Points When the World Changes Mid-Task — drmapavone · 2026-10-27
- Agility's Digit robot to expand into two more GXO US warehouses from early 2027 — jonstephens85 · 2026-10-27
- Real2Sem2Real turns one phone video into 2,000+ robot demos, matching 150 teleop demos — Scobleizer · 2026-10-27
- Wayve CEO argues end-to-end deep learning should power driving, safety and world models — alexgkendall · 2026-10-27
- Mundo turns a few room photos into an explorable, simulatable 3D world — at 1.5 FPS — mark_k · 2026-10-27
- Zurich Physical AI Hackathon puts attendees in a room with $1.5M of humanoids — ___Mufasaa · 2026-10-27