PhyAI: Real-Time Physical AI at the Edge, Scalable Rollouts in the Cloud

PhyAI is a unified inference engine for physical AI across evaluation, cloud reinforcement-learning rollouts, edge GPU serving, and onboard deployment. Model-specific conditioning, solvers, caches, and output logic remain in adapters, while graph execution, kernels, memory management, and parallel services are shared. The authors report 1.40x-4.65x speedups over official implementations of pi0, pi0.5, GR00T N1.7, and MiniCPM-Robot. For Cosmos3-Nano-Policy-DROID, latency falls from 2.46 to 1.18 seconds on eight H20 GPUs. Profiling also shows that optimal execution policies differ substantially by model and batch size.