映像と操作命令を分けて送る、物体検出付きロボットアーム遠隔操作

A Modular AIoT Framework for Low-Latency Real-Time Robotic Teleoperation in Smart Cities

This paper presents an AI-driven IoT robotic teleoperation system designed for real-time remote manipulation and intelligent visual monitoring, tailored for smart city applications. The architecture integrates a Flutter-based cross-platform mobile interface with MQTT-based control signaling and WebRTC video streaming via the LiveKit framework. A YOLOv11-nano model is deployed for lightweight object detection, enabling real-time perception with annotated visual overlays delivered to the user interface. Control commands are transmitted via MQTT to an ESP8266-based actuator node, which coordinates multi-axis robotic arm motion through an Arduino Mega2560 controller. The backend infrastructure is hosted on DigitalOcean, ensuring scalable cloud orchestration and stable global communication. Latency evaluations conducted under both local and international VPN scenarios (including Hong Kong, Japan, and Belgium) demonstrate actuator response times as low as 0.2 seconds and total video latency under 1.2 seconds, even across high-latency networks. This low-latency dual-protocol design ensures responsive closed-loop interaction and robust performance in distributed environments. Unlike conventional teleoperation platforms, the proposed system emphasizes modular deployment, real-time AI sensing, and adaptable communication strategies, making it well-suited for smart city scenarios such as remote infrastructure inspection, public equipment servicing, and urban automation. Future enhancements will focus on edge-device deployment, adaptive routing, and integration with city-scale IoT networks to enhance resilience and scalability.