Pudu Robotics Unveils Embodied Large Model PuduFM
量子位 · wechat · 2026-07-17
At WAIC, Pudu Robotics launched its embodied intelligence large model PuduFM and the embodied AI operating system PuduAgent. This aims to solve data reuse and intelligence transfer across different robot forms and scenarios via a "One Brain, Many Forms" strategy.
Core Technology & System
- PuduFM: Consists of three main components. PIM (Physics Oracle), based on a causal attention Transformer, extracts physical dynamics like gravity and friction to give the robot physical intuition. VLA Multimodal Alignment integrates vision, voice, and mobile manipulation, allowing one brain to coordinate continuous actions. The World Model simulation engine generates massive adversarial scenario data from real-world assets, combining it with limited real feedback to achieve a "virtual-real dual closed-loop."
- PuduAgent: Features a core runtime base (OS), a standardized atomic capability library, and safety mechanisms. It handles unified scheduling, memory management, and cross-form skill reuse, acting as the "Android/iOS" for robots.
Commercialization & Data Flywheel
The report identifies commercial services as the primary battleground for scaling embodied intelligence. Pudu has deployed over 130,000 robots globally, generating tens of millions of hours of real navigation and operation data annually. Acquiring data through massive real-world deployments feeds back into model evolution, which is then expanded to more scenarios, forming a complete embodied AI data flywheel.
Related event: Pudu Unveils PuduFM and PuduAgent for Embodied AI(2 posts)→
More from Embodied
- LIBERO-MAX: Robot Policies Drop 11–26 Points When the World Changes Mid-Task — drmapavone · 2026-10-27
- Photos from ROSCon UK: BBC's robotic whale, da Vinci surgical robot, PR2 and Baxter — kamathsblog · 2026-10-27
- Agility's Digit robot to expand into two more GXO US warehouses from early 2027 — jonstephens85 · 2026-10-27
- Real2Sem2Real turns one phone video into 2,000+ robot demos, matching 150 teleop demos — Scobleizer · 2026-10-27
- Wayve CEO argues end-to-end deep learning should power driving, safety and world models — alexgkendall · 2026-10-27
- Mundo turns a few room photos into an explorable, simulatable 3D world — at 1.5 FPS — mark_k · 2026-10-27