泡茶不洒、冲咖啡不慌……中国移动开源Open-RAIL，让机器人动作精度提升两个数量级

Brief (coffeeyo.com): Since September 17, crowds have gathered at China Mobile’s booth at the China-ASEAN Expo: a tea-ceremony robot pours Liubao tea while a barista robot hands over a fresh latte. Both run on Open-RAIL, an open-source embodied-AI framework China Mobile released that lifts robot motion precision by two orders of magnitude.

Since September 17, the B2 hall of the Nanning International Convention and Exhibition Center has been packed in front of China Mobile’s booth. One Chinese-style tea-ceremony robot lifts a pot to pour water, brews, and serves tea — the cup holding Guangxi’s local Liubao tea; a few steps away, another robot grabs a cup, places it, presses a button, and hands a freshly ground latte to the visitor. Running behind both robots is the very same open-source framework: Open-RAIL.

On September 16, China Mobile officially open-sourced Open-RAIL to the world — the industry’s first general engineering base connecting Vision-Language-Action (VLA) / World Action Model (WAM) models with different robot bodies. It went live simultaneously on the “Huanxin Community,” GitHub, and Gitee. Enterprises and developers can freely use, modify, and integrate the framework into their own products.

17 Steps of Tea Art and 90 Seconds of Coffee: From “Performing” to “Working”

At this year’s China-ASEAN Expo, demonstrations of embodied robots expanded from performance-style motions to actual work. The two robots at China Mobile’s booth are a live sample of that direction.

Take tea brewing first. Tea leaves are granular — scooping must be accurate yet spill nothing and preserve leaf shape; water must be controlled to the milliliter, a drop too much and the cup overflows; from adding leaves and brewing to pouring and serving, the whole process runs 17 steps, and any tiny pose deviation accumulates downstream. Reporters at the scene saw the full tea ceremony finish in about 2 minutes with no human intervention, spillage kept below 5%.

Look at the coffee next. Grasping the cup, placing it, pressing, waiting for output, and transferring it — all done by a single robot, with no control logic written separately for any one step. From order received to cup produced takes about 90 seconds.

While waiting for extraction, the robot doesn’t sit idle — it makes ice cream, fetches ice, and returns just as the coffee finishes extracting, so serving naturally speeds up.

Tea That Doesn’t Slosh, Coffee That Doesn’t Spill: Open-Source Open-RAIL Stabilizes Every Step

The model “thinks” slowly while the robot moves urgently: a large model outputs actions by the second, yet the robot’s control loop issues commands continuously at the millisecond level — a gap of dozens of times. Worse, the model hands over discrete action points with no transition between them, which on the real machine shows up as joint jitter and impact.

Open-RAIL solves exactly this. It lets observation, inference, and control each run independently and wait for none; it reconstructs discrete action points into a continuous trajectory, then smoothly “connects” the next segment from the robot’s current pose and velocity — so the water in the cup doesn’t slosh and the coffee won’t spill on the way to the visitor. After smoothing, the standard deviation of joint acceleration dropped from 10+ rad/s to 0.1 rad/s, down two orders of magnitude, for smoother motion and smaller impact.

Currently Open-RAIL adapts four heterogeneous robot types and supports ten mainstream VLA/WAM models; the adaptation code to bring on a new model is usually within 50–100 lines.

According to China Mobile, the company has formed a complete technology layout across models, data, bodies, platforms, and networks in embodied intelligence. Next, China Mobile will keep improving the framework around model adaptation, data collection, and online evaluation, so embodied AI reaches real scenarios faster and serves China-ASEAN artificial-intelligence cooperation.

Reprint Note & Disclaimer

This article is compiled and edited by coffeeyo.com. The materials and images herein are sourced from public channels and the copyright belongs to the original authors. This site reprints them solely for information sharing within the coffee industry and does not represent our stance; all data and views are subject to the original source. We thank the original authors for their work and sharing.
For matters concerning content authorization, attribution, copyright, or any other issues, please contact us promptly. We will verify and cooperate in handling them without delay. We also welcome coffee-growing regions, brands, and industry media to exchange and contribute articles with us.
Contact coffeeyo.com — Tel: 400-0000-641 (+86-18288016826) | Email: 405854472@qq.com | Address: Jinghong City, Xishuangbanna, Yunnan / Tieling City, Liaoning

Reprint Notice & Disclaimer

This article has been edited and compiled by China Coffee Network. The information and images therein come from public sources and the copyright belongs to the original author. This reprint is intended solely for information sharing within the coffee industry and does not represent the views of this website; all data and opinions are subject to the original article. Our thanks to the original author for creating and sharing it.
For matters concerning content licensing, attribution, copyright or otherwise, please contact us promptly and we will verify and cooperate in handling them as soon as possible. We also welcome coffee origins, brands and industry media to exchange ideas and contribute articles.
Contact China Coffee Network: Tel 400-0000-641 (+86-18288016826) | Email 405854472@qq.com | Address: Jinghong, Xishuangbanna, Yunnan Province / Tieling, Liaoning Province