ShengShu Technology Introduces Motubrain, a Unified AI Model for Robotics

ShengShu Technology has introduced Motubrain, a world action model that functions as a unified brain for robotics, announced in a press release. The system replaces task-specific models with a single architecture capable of perception, reasoning, prediction, and action across various robot types.

Motubrain ranked among the top performers on key embodied AI benchmarks, scoring 63.77 on WorldArena and 96.0 on RoboTwin 2.0 across 50 tasks. It remains the only model to exceed 95.0 in randomized environments. The model integrates video, action, and language within a single multimodal framework, using a Mixture-of-Transformers structure to manage vision, language, and control simultaneously.

The model learns from large-scale unlabelled video and multi-robot data, allowing it to generalize across environments and robot types. In testing, robots trained with Motubrain completed complex multi-step tasks, adapting to changing conditions and retrying actions when outcomes failed.

Motubrain is already being used in active robot training programs through partnerships with Astribot, SimpleAI, and Anyverse Dynamics. The model builds on ShengShu’s earlier work with its Vidu video generation platform and is supported by a 293 million dollar Series B funding round led by Alibaba Cloud, alongside investors including Baidu Ventures and the China Internet Investment Fund.