Strands Agents: 자연어 명령으로 로봇을 제어하는 오픈소스 에이전트 프레임워크

Physical AI represents the next evolution in robotics, shifting from rigid, pre-programmed tasks to autonomous systems capable of natural language interaction. By integrating Vision-Language-Action (VLA) models with agentic frameworks, robots can interpret complex human intents and translate them into physical actions. The presentation introduces Strands Agents, an open-source SDK designed to facilitate this transition by treating robot sensors and actuators as standard AI tools. A hybrid architecture is employed to balance these requirements, where low-latency control tasks are executed locally on edge devices, while complex reasoning and planning are delegated to the cloud. This approach significantly reduces the training burden, as robots can learn from fewer samples in simulation and self-correct during runtime, paving the way for more flexible and human-collaborative robotic systems.