Abstract
Autonomous mobile robot (AMR) platforms have emerged as practical solutions for logistics operations in dense industrial IoT environments. This paper presents a remote data-processing architecture for legacy mobile robots that offloads compute-intensive perception while preserving existing control software. The system uses ROS and ROSbridge over a secure Tailscale VPN to connect a mobile robot based on an NVIDIA Jetson Orin Nano 4GB to a remote NVIDIA Jetson Orin NX 16GB server across heterogeneous networks. In the studied use case, image topics published by the robot are processed remotely using Ultralytics YOLO, and the resulting information is returned to the robot and external monitoring clients. The work evaluates the feasibility of this architecture for IoT robotics in terms of communication stability, end-to-end latency, resource usage, and system bottlenecks. The results indicate stable connectivity and latency suitable for near-real-time operation, while reducing the computational burden on the robot. These findings suggest that the proposed approach provides a practical basis for upgrading legacy robotic platforms and supporting future multi-robot orchestration through containerized services.