Shengshu unveils Motus 2, a self-evolving world model hinting at recursive self-improvement
生数科技 · wechat · 2026-09-10
At the 2026 Bund Summit, Shengshu Technology CEO Luo Yihang unveiled Motus 2, a self-evolving general world model for robotic dexterous manipulation, alongside a five-level L1–L5 roadmap for general world models: generating worlds, interacting, acting, autonomous world agents, and world organizers.
- Motus 2 unifies policy, world simulation, and value evaluation in one closed loop: it proposes actions, predicts outcomes, evaluates whether actions approach goals, and refines strategy from feedback — an early form of recursive self-improvement (RSI).
- The model adds tactile feedback and learns from large-scale egocentric human manipulation data, with skills transferable across robots and tasks.
- Luo summarizes the vision as "mind, then dexterity": reasoning before action plus fine control in the real world.
The team says more experimental results and technical details will follow from Sept 12. Open problems toward L4/L5 remain long-horizon memory, physical/causal modeling, online learning, and low-latency deployment.
More from Companies & People
- Glean Is Worth $7.2B, but What's Actually Its Moat? — yogthinks · 2026-09-11
- Replit CEO joins backlash against AI doomers, says focus should be on cybersecurity not sci-fi — amasad · 2026-09-11
- Replit CEO Amjad Masad: AI existential-risk talk is a distraction from real cybersecurity work — amasad · 2026-09-11
- SF hiring of wet-lab biologists for RL environments draws bio-safety concerns — beffjezos · 2026-09-11
- Schmidhuber appears on the "Europe on Edge" podcast — SchmidhuberAI · 2026-09-11
- VC advice for Indian founders: stop pitching that you'll be in SF, go where customers are — vaibhavbetter · 2026-09-11