China Mobile Open-Sources Open-RAIL to Streamline VLA/WAM Model Integration

On September 16, China Mobile announced the global open-source release of Open-RAIL, which it describes as the industry's first general engineering base connecting VLA/WAM models with robot bodies. The platform integrates model inference, real-world execution, data feedback, and model iteration into a closed loop.

Open-RAIL has been adapted for four heterogeneous robots and supports 10 mainstream VLA/WAM models. New models can be integrated with as few as 50 to 100 lines of code by implementing a standard inference interface. The platform uses a lightweight hardware abstraction layer to standardize control, state reading, and action execution, so that a new robot only needs underlying adaptation and registration while upper-layer logic remains unchanged.

According to China Mobile, action smoothness has improved by two orders of magnitude, with joint acceleration standard deviation dropping from over 10 rad/s² to 0.1 rad/s². Each real-world run automatically accumulates training data in a ready-to-use format at no extra cost. The platform supports flexible edge-cloud deployment without code changes. The open-source repository has been made available.