China Mobile open-sources Open-RAIL middleware to link VLA/WAM models with robot hardware

China Mobile released Open-RAIL (Apache-2.0) on GitHub, a server-client software that connects vision-language-action and world-action models to real robots. The middleware reduces joint acceleration variance from 10 rad/s² to 0.1 rad/s², bridging a 30–50x control frequency gap without retraining the model.

• Three-thread asynchronous pipeline: observation, inference, and control
• Joint acceleration variance drops from 10 to 0.1 rad/s² without retraining
• Supports 10 models: ACT, GR00T N1, RDT, SmolVLA, GO-1, π0/π0.5, TAO
• Adapted for AgiBot G1, Lingxi Ti5 T170C, NAVIAI-WA2, and LeRobot mock