Remote teleoperation of an anthropomorphic robotic hand with cloud-enabled wearable sensing and lightweight MLP

Abstract

Teleoperation of robotic hands has gained significant attention due to its potential in assistive technologies, remote interaction, and operation in hazardous environments. Recent advances in wearable sensing, embedded intelligence, and cloud-based communication have created new opportunities for enabling real-time, low-cost solutions in this domain. This work explores a framework that integrates a flex-sensor-based wearable glove with a lightweight multilayer perceptron (MLP) model to classify finger bending into discrete postures. The predicted commands are transmitted via a cloud-based database to a tendon-driven robotic hand, enabling remote actuation across independent networks. Experimental evaluation demonstrated high classification accuracy and optimal end-to-end latency, validating the feasibility of combining affordable hardware, embedded learning, and scalable cloud infrastructure for practical remote teleoperation of anthropomorphic robotic hands.