High-resolution data streams-such as images, Li-DAR point-clouds-are common in robotic communication, yet Robot Operating System 2 (ROS 2) struggles to transmit these streams due to increased average latency. On lossy wireless links, the default DDS communication stack in ROS 2 suffers significant performance degradation. This paper presents the first comprehensive network-layer analysis of ROS 2’s DDS stack operating over wireless links with large payloads. We analyze three network bottlenecks that emerge during large-payload data transfers and presents DDS-level optimizations for each one. The proposed solutions are exposed through an XML-based QoS configuration interface, allowing them to be easily tuned. Experiments demonstrate that our approach transmits large-payload data successfully while maintaining lower latency than existing methods.

The integration of Cyber-Physical Systems (CPS) and AI presents both opportunities and challenges. AI operates on the principle that “good things happen probabilistically,” while CPS adheres to the principle that “all bad things must not happen,” requiring uncertainty-awareness. Furthermore, the difference between AI’s resource accessibility assumption and CPS’s resource limitations highlights the need for resource-awareness. We introduce Cyber-Physical AI (CPAI), an interdisciplinary sub-field of AI and CPS research, to address these constraints. To the best of our knowledge, CPAI is the first research domain on CPS-AI integration. We propose a 3D classification schema of CPAI: Constraint (C), Purpose (P), and Approach (A). We also systematize the CPS-AI integration process into three phases and nine steps. By analyzing 104 studies, we highlight nine key challenges and insights from a CPAI perspective. CPAI aims to unify fragmented studies and provide guidance for reliable and resource-efficient integration of AI as a component of CPS.