PuduFM 1.0

Pudu Robotics built PuduFM 1.0 as an embodied-intelligence foundation model that gives robots genuine physical intuition. Rather than treating navigation and manipulation as separate jobs, the model unifies them inside a single cognitive framework, so a robot can move to a spot and act on the world through one continuous decision loop. The system pairs two parts: a Physical Intuition Model that reasons about how objects behave, and a multimodal Vision-Language-Action layer that turns that understanding into movement. As a result, robots no longer just detect objects — they predict consequences, grasp physical constraints, and act with real foresight. Pudu frames this leap as the shift from simple execution tools toward what it calls General Physical Agents.

What truly sets PuduFM 1.0 apart is its scale and its self-improving design. The model learns from one of the industry’s largest real-world datasets, drawn from over 130,000 robots deployed across 80+ countries, and it keeps getting better through a flywheel that spans both simulation and reality. Because Pudu trains it in a high-fidelity world simulator and then refines it with live human feedback, the model sharpens its decisions with every interaction. Crucially, it follows a “One Brain, Multiple Embodiments” approach, so delivery, cleaning, industrial, and humanoid robots can all share the same intelligence. This single, evolving brain points toward a future where robots understand the physical world and create lasting value within it.