Integrating ROS 2 and Physical AI: Architecture and Challenges

Abstract

Physical Artificial Intelligence (Physical AI) embeds sensors, actuators, and AI algorithms to enable robots to learn and adapt in real-world settings. ROS 2, with its DDS-based publisher-subscriber model and advanced Quality of Service (QoS) features, forms a robust platform for such real-time, distributed applications. Yet, integrating AI within a Cyber-Physical System (CPS) framework raises concerns about resource constraints, safety, and uncertainties. This paper explores how to incorporate Physical AI into ROS 2 by fully automating or partially augmenting robot functionalities, focusing on topic structures, message formats, and key QoS considerations. We highlight major challenges - data availability, data loss, inference latency, and security vulnerabilities - and emphasize the importance of a holistic approach to ensure reliability and safety.