How to Build Physical AI Agents: Natural Language for Real-World Robotics

Today's language models are trained on text, multimodal models are trained on text, audio, and video, and Vision-Language-Action models (VLAs) are trained on sensor state and the physical world. With Strands Agents, we made building digital agents accessible and are now bringing that same developer experience to the edge and physical AI. In this demonstration we will connect an SO-101 3D printable robotic arm with a tuned NVIDIA GR00T VLA model on Jetson edge devices, Hugging Face LeRobot for hardware integration, and Strands Agents orchestrating the physical AI agent intelligently through natural language.