The global race for 'Embodied AI'—intelligence that can perceive, reason, and act within the physical world—has a new frontrunner. Wujie Dongli, a rising Chinese AI firm, has officially released its MWA™ (Embodied General Brain), the world’s first latent space world model based on a long-sequence bidirectional physical causal chain. This breakthrough aims to solve one of the most persistent hurdles in robotics: the ability to execute complex, multi-step tasks with the same fluid continuity as a human.
Technically, the MWA™ model utilizes a 'Temporal Chunk-level' inverse dynamics modeling mechanism. By outputting continuous 'Latent Action Chunks,' the system allows robots to navigate diverse and complex environments while maintaining long-term task coherence. This shift from simple reactive movements to predictive, causal-based planning represents a significant leap in how machines internalize the laws of physics and temporal progression.
The model’s prowess was recently validated on the RoboCasa GR1 TableTop leaderboard, a premier benchmark for embodied intelligence co-founded by Stanford University. Wujie Dongli’s MWA™ claimed the top spot, notably outperforming industry heavyweights and high-profile rivals. Its performance eclipsed Nvidia’s GR00T-N1.6, as well as significant domestic competitors including DexForce’s ACE-EGO-0, Xpeng’s DIAL, and Agibot’s ABot-M0.
This technical milestone follows a massive vote of confidence from the capital markets. On June 26, Wujie Dongli announced it had secured over $200 million in an Angel funding round, led by funds associated with e-commerce giant JD.com. The scale of this initial investment, coupled with the benchmark success, positions the company as a central player in China’s broader strategy to lead the next generation of artificial general intelligence applied to hardware and manufacturing.