A Roadmap for Physical AI Foundation Models in Robotics

Abstract

Vision-language-action models excel at semantic generalization, yet real-world robotics imposes latency, jitter, energy, thermal and safety constraints that monolithic models cannot absorb. We argue that Physical AI foundation models must become cloud-edge-robot systems that separate time scales, safety authority and data lifecycles. We propose this architecture, an Action Proposal Contract and Physical Intelligence Metrics as a roadmap for assurance-ready robotics.