Kuavo-linked Dimensional AI proposes "PhysicalToken economics" to measure marginal cost of robot skills
机器之心 · wechat · 2026-08-20
At the Physical AI Leaders Forum during WRC2026 in Beijing, Dimensional Intelligence (跨维智能) founder Jia Kui delivered a keynote on "PhysicalToken economics," proposing that the viability of general physical intelligence be measured by the marginal cost of acquiring new capabilities: if every new skill requires roughly the same data, compute and engineering, the business is still project delivery; only a declining cost curve creates true economies of scale.
The company frames its stack in four stages: Experience (unified data representation with Dexterity-BEV alignment to make learning reusable), Leverage (leveraging small amounts of real data with the DexVerse engine and GS-World generative simulation, cutting environment setup from days to 10 minutes), Condensation (efficient post-training on a foundation-model base), and Expansion (feeding deployment data back into the model as a flywheel). Three demos were shown: phone assembly inspection, autonomous retail scanning, and ice cream/popcorn manipulation.
The core argument: zero-shot may not be a precondition for scale but the end state of sustained economies of scale. The company raised a 1-billion-yuan Series B in June at a post-money valuation over 10 billion yuan.
More from Embodied
- Weekly Robotics Roundup: Updates from Unitree, Generalist, Gravis, and More — lukas_m_ziegler · 2026-08-20
- DaxAI's Rideable Robot Horse Qigi Now Available: X1 at $43K, Wheel-legged XS at $53K — Scobleizer · 2026-08-20
- JPMorgan: Tesla FSD v15 launching this year to boost Robotaxi deployment — JOBhakdi · 2026-08-20
- TMLR Survey: The Embodiment Gap in Robot Foundation Models — HirokatuKataoka · 2026-08-20
- 'Humanoids are becoming more human': robot evolution clip circulates — HeyZoyaKhan · 2026-08-20
- Autonomous Vehicle Navigates Tight Indian Traffic in Driverless Demo — sanjeevs_iitr · 2026-08-20