6G-XR

The MechEye project establishes technical foundations for AI-driven industrial video analysis by developing and evaluating a proof-of-concept system that uses advanced Vision Language Models (VLMs) to improve industrial monitoring, safety, and operational efficiency. Conducted under the 6G-XR Open Call 3, this project integrates video capture, frame sampling, object detection, captioning, and reasoning to detect anomalies. It was designed to allow flexible deployment and experimentation of components across edge and cloud environments. The experiments were implemented on the 5G Test Network (5GTN) at the University of Oulu to assess performance under varying network conditions and compute placements.

The project validated the feasibility of AI-driven industrial monitoring over the 6G-XR infrastructure, provided an initial implementation, and produced empirical data on the compute communication trade-offs for future distributed AI systems. Although current inference times (∼1 second per frame) still limit real-time operation, immediate practical value for a system such as MechEye can be achieved in, e.g.: (1) periodic safety compliance audits, (2) training assessment and procedure verification, and (3) post-incident investigation. MechEye establishes a foundation for future work, including expanding datasets and enhancing reasoning and visualization to enable AI-assisted interactive industrial environments. Our experiments also confirmed that network slicing maintains QoS under congestion, and this is important for multi-camera setup.