Huawei Cloud sets Nov. 16 open-source launch for robot world model WM1.2

Huawei Cloud has scheduled an open-source release session for CloudRobo-WM1.2, its new world model for robots, at a November 16 launch event. The agenda on Huawei Cloud's event page puts the session at 2:50 p.m. China Standard Time, after the company introduces the model and describes how it works.

The question for robotics teams is what they will be able to obtain. The agenda identifies the model and reserves time for an open-source release. The files covered by that release, the license for using them and any access requirements remain unspecified. A release of model weights would give developers different options from a release that also includes inference code or training data.

A separate slot for the release

Early coverage of the November 12 event announcement by Shanghai Securities News described a world model built for Huawei's Ascend computing platform. It listed a model introduction, a technical discussion and sessions about industry partners. The more detailed agenda on Huawei Cloud's event site sets aside a separate five minutes for the formal launch and open-source release. The page does not give a date for the agenda revision.

The November 16 stream is scheduled to begin at 2 p.m. in China. Wu Xiaodong, who leads the company's embodied intelligence product line, is listed first. Model specialist Liu Liu is scheduled to start the architecture and capabilities discussion at 2:20 p.m. The page directs viewers to sign in with a Huawei Cloud account for the stream. It gives no download link for the model itself.

What the existing platform already offers

Huawei Cloud introduced its current CloudRobo development platform at a June event. The service lets teams bring in data, evaluate models in simulation and test them on supported physical robots. The November event concerns a specific model that carries the CloudRobo name. An agenda entry leaves open when that model might become available for selection inside the existing service.

One established use of the platform is recording robot movements in a simulated task. The trajectory generation guide describes running a task with a virtual robot and saving the movements that completed it successfully. The feature is documented for one cloud region in Guiyang. It has a limited free allocation in a shared pool, and dedicated resources are billed. That pricing applies to the existing data generation feature.

Moving a task to physical hardware adds its own integration work. Huawei Cloud's examples list a pen placement task with the SO101 arm as separate simulated and physical entries. Its documentation also says teams can adapt their own robots by providing code that connects the device to the cloud platform. Those resources describe the existing service and its examples, so WM1.2's own hardware support remains open.

CloudRobo already manages simulation scenes, datasets and models as separate kinds of assets. Teams assembling a test in CloudRobo can choose the environment and robot separately from the model. A checkpoint would give them the model parameters. Reproducing a specific task would also require the corresponding robot and objects and the settings for the test environment. The event agenda has yet to list supported devices or provide test results for WM1.2.

After the release session, Huawei Cloud plans to discuss partner work using its Ascend infrastructure. The detailed agenda also lists a preview of a partner's product. The download timing and the availability of WM1.2 in existing CloudRobo workspaces depend on the release materials.