Sudu Tech's Sim-to-Real Embodied AI Approach
量子位 · wechat · 2026-07-20
This in-depth article details Sudu Tech's embodied AI showcases at WAIC, including zero-shot grasping in simulation, dual-arm precision assembly, flexible packing, mobile manipulation, and a four-machine battery module production line collaboration with CATL.
The core focus is their "simulation-first, sim-to-real fusion" technical route. The team argues that real-world robot data is hard to scale, making it more effective to train a generalized foundation model using large-scale simulation data first, then fine-tune with minimal real-world demonstrations to achieve a deployable 99%+ success rate. The article also introduces their proposed high-granularity "fourth type of world model," which goes beyond rendering and simulation to accurately predict material properties, friction, geometric constraints, and post-contact state changes.
Additionally, the article reveals that Sudu, founded just 14 months ago, has reached a valuation of approximately 20 billion RMB and shares the founder's insights on robotic data, world models, and the path to industrial deployment.
Related event: Sudu Tech Showcases Embodied AI Skills and CATL Line at WAIC(2 posts)→
More from Companies & People
- World Labs brings SceniX in as it pushes spatial intelligence into the physical world — XRarchitect · 2026-07-22
- RSS launches under OMSF to push structural biology data modeling at scale — MoAlQuraishi · 2026-07-22
- Moonshot AI reportedly targets a $50B round ahead of Hong Kong listing — ctjlewis · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- AI industry astroturfing roundup tracks the sector’s fake-grassroots problem — ShakeelHashim · 2026-07-22
- Enterprise AI’s real value is in agentic workflows, not generic chatbots — irfan__77 · 2026-07-22