BROWSE
Detail
Conference Paper
Process-State-Aware Remote MEC GPU Sharing for Vision-Guided Collaborative Robotic Welding
Cited 0 time in 
Share 
Share
- Authors
- Woo-Sung Jung, Tea Hyun Yoon, Dongkoo Shon, Dae Seung Yoo
- Issue Date
- 2026-07
- Citation
- International Conference on Ubiquitous and Future Networks (ICUFN) 2026, pp.1097-1102
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- Vision-guided robotic welding requires GPU-intensive perception for scene understanding, workpiece recognition, and weld-start localization. However, such high-load computation is needed only during short pre-welding phases, while the actual welding phase is dominated by motion execution and feedback control. This mismatch makes per-robot GPU deployment costly and inefficient in multi-robot welding cells. In this paper, we propose a remote MEC-assisted architecture for vision-guided collaborative robotic welding, where GPU resources are shared across robots through a 5G-enabled industrial network. The proposed architecture places sensing and robot execution on the shop floor, while perception inference is offloaded to remote compute resources. To position the proposed architecture, we compare it with prior work on vision-aided robotic welding, cloud-edge robotic welding, shared edge GPU scheduling, and lightweight distributed perception. We further present a lightweight simulation-based feasibility study using three deployment modes, namely Local, Cloud, and Proposed MEC. The evaluation focuses on a large hardware-gap case between robot-side embedded GPUs and remote server-class GPUs, and additionally reflects the latency penalty of a slow and high-jitter industrial backbone in the Cloud mode. The results show that the Proposed MEC mode preserves the compute advantage of remote inference while avoiding the backbone bottleneck of external cloud access. The paper concludes with open research issues on scheduling, reliability, deployment efficiency, and future split-processing extensions.
- Keyword
- robotic welding, MEC, edge computing, GPU sharing, Collaborative robot, computer vision, 5G industrial networks
- KSP Keywords
- Collaborative robot, Computer Vision(CV), Edge Computing, External cloud, Feasibility Study, Feedback control, GPU scheduling, Load computation, Multi-Robot, Open research, Research Issues