MMP robotics model demonstrates human-like reasoning and zero-shot household chores
量子位 · wechat · 2026-09-02
A mysterious team has released new demos for an embodied AI model codenamed 'MVP' (Make Veritable People). The videos show the model excelling in physical reasoning and human-like decision-making: it maintains balance while holding a water cup against external pushes and simultaneously stabilizes falling cans, demonstrating dynamic prediction capabilities. In zero-shot household tasks, the robot understands object properties (e.g., hanging rings, throwing cushions) and adopts 'lazy' strategies to optimize efficiency. Two different robots also collaborated on tasks like shaking bed sheets. The team claims the model achieves an 80% zero-shot success rate and possesses self-evolutionary capabilities, distinguishing it from mainstream VLA and WAM models that rely on probabilistic prediction or rigid concatenation.
More from Embodied
- GoatRemote update: AI voice commands for Mac via Siri Remote — mayfer · 2026-09-02
- Supply Chain Bottlenecks: Millions of Robots Need Billions of Precision Parts — Scobleizer · 2026-09-02
- AndroidWorld benchmark exposes static mobile agent flaws — East-Muffin-6472 · 2026-09-02
- Bought a $6,279 RTX 5090 Prebuilt: Was it the right choice for local AI? — Sexyvette07 · 2026-09-02
- Hacking Feetech servos: Tuning registers for better performance — kamathsblog · 2026-09-02
- GoPro sale confirms Shenzhen's dominance in hardware — alexmacgregor__ · 2026-09-02