ShengShu Technology's world model team has started livestreaming the full training process of Motubrain 2.0, the next version of its embodied AI model for robots. The broadcast began on October 21 and is planned to run for about two weeks. The company says Motubrain 2.0 will be open-sourced soon, and it chose to show the training process before that release.
Motubrain is ShengShu's family of robot foundation models, aimed at giving robots a general-purpose brain that can carry skills across tasks, scenes and robot bodies. The first Motubrain was introduced in April as a world action model. ShengShu is better known for its Vidu video generation models.
The livestream is organized into four stages. Pretraining looks at how training scale affects performance and zero-shot generalization. Post-training targets specific tasks to improve execution. An agent task stage tests whether the model can transfer what it has learned to other tasks. A final stage explores recursive self-improvement, asking whether the model can judge its own actions from execution feedback and improve them. The live page shows training steps and evaluation success rates as they update.
The pitch is transparency. Outsiders usually judge embodied AI progress through polished robot demos, benchmark tables and papers, which reveal little about how capability forms during training or what fails along the way. ShengShu says the broadcast will show failures, trial and error and adjustments as well as gains, and that it hopes to share open questions about scaling, cross-task generalization and autonomous improvement with other researchers.
The move follows a broader trend of Chinese AI teams sharing training progress publicly. Earlier, Luo Fuli, who leads Xiaomi's MiMo models, posted details from the reinforcement learning stage of MiMo-V2.6.
There are limits to what a livestream proves. The weights have not yet been released, and the company itself says the research is still exploratory and that results will need later experiments and evaluation to confirm. The announcement was supplied by ShengShu and carried by Chinese tech media. The open-source release, and whether its benchmarks match what the stream shows, will be the real test.
SEE ALSO: Shengshu Technology Launches Motubrain · Shengshu Technology Releases Motus2