Physical AI Runtime is the lightweight deployment layer for Physical AI. It runs exported policies with a unified API and CLI, and provides camera/robot interfaces for real‑world execution.

• Unified Inference API - Load and run exported policies with InferenceModel
• Runtime CLI - physicalai run/serve/validate for deployment workflows
• Hardware Interfaces - Camera and robot interfaces with clean boundaries
• Benchmarking - NumPy‑only benchmarking runner and protocols
• Format‑agnostic - Works with exported models from physicalai-train, LeRobot, and custom frameworks

pip install physicalai

from physicalai import InferenceModel

policy = InferenceModel("./exports/act_policy")
action = policy.select_action(observation)

physicalai run --model ./exports/act_policy --robot robot.yaml

We welcome contributions! See CONTRIBUTING.md for guidelines.