Open-RAIL: China Mobile's open-source middleware boosts humanoid execution 2.09x
CyberRobooo · x · 2026-09-23
- China Mobile open-sourced Open-RAIL, a plug-and-play asynchronous inference middleware for VLA/WAM models and heterogeneous robots, with paper and code (GitHub/Gitee) available.
- Mechanism: an async pipeline decouples 5-10 Hz VLA inference from 200-500 Hz motor control; two-stage trajectory smoothing achieves near-C² continuity, cutting acceleration std 100× (10+ → 0.1 rad/s²).
- Compatibility: 20+ VLA/WAM models and 4 robot types; adapting a new model takes only 50-100 lines of code without retraining; cloud-edge-end deployment latencies of 3-5 ms / 35-45 ms / 80-120 ms.
- Results: up to 2.09× speedup over raw policy output (no data recollection), success rates up to 0.95 on π0.5 and GR00T-N1.5 with absolute gains of +0.10 to +0.725.
- Closed loop: automatic LeRobot-format real-robot episode recording with time-aligned human-in-the-loop tele-correction for continuous improvement.
More from Embodied
- UBTECH sells 400 U1 Pro humanoids to Yongda Auto in ~$10M deal — CyberRobooo · 2026-11-12
- Baidu World 2026 to showcase Apollo Go robotaxi rides and Apollo World physical AI model — Baidu_Inc · 2026-11-12
- NVIDIA's open-source Alpamayo 2 Edge VLA driving model lands in Autoware — drmapavone · 2026-11-12
- A closer look at Astra's robot gripper design (and a joke at its expense) — _Stocko_ · 2026-11-12
- Robot startup passes first presale milestone with 10+ units claimed — R4nsack · 2026-11-12
- Runway launches Praxis-1, a 27B MoE world action model, with a 73ms Fast variant — ChongZzZhang · 2026-11-12