An Edge-Fog-Cloud-based Hierarchical Adaptive Model for Human-Robot Interaction*

With the rapid advancement of intelligent devices and the Internet of Things (IoT), a vast amount of detailed information can now be captured and analyzed from the human body. This progress has led to the development of more comprehensive and versatile Body Area Networks (BANs). Despite this promising landscape, the field lacks mature global applications, and numerous challenges await resolution, particularly in Human-Robot Interaction (HRI).To address the limitations of HRI, this paper focuses on recognizing complex human activities and achieving rapid, stable transmission of high-load data. An Edge-Fog-Cloud-based hierarchical adaptive system is proposed, utilizing multiple IMU sensors for human activity recognition (HAR). The system introduces an adaptive strategy to enhance the accuracy of identifying complex human behavior. By incorporating hierarchical communication and classification methods, the system improves the communication network’s real-time performance and the algorithms’ architectural structure, ensuring real-time capabilities for the HRI system.

I. Introduction

Autonomous systems based on the Internet of Things (IoT) are experiencing rapid growth, entering an era characterized by affordability, robustness, and reliability. This progress is facilitated by faster communication and extensive data development [1]. However, the escalating cost of computing energy burdens existing infrastructure [2]. Rather than solely focusing on enhancing infrastructure capabilities, an EdgeFog-Cloud hierarchical structure is gradually adopted within autonomous systems [3]. This architecture enables data processing across different layers, maximizing device utilization and enhancing real-time capabilities [4]. Consequently, an abundance of Edge-Fog-Cloud structures has been used on HAR. For instance, a Docker-based Edge-Fog-Cloud system was developed specifically for fall detection, demonstrating comparable accuracy and real-time performance to state-of-the-art approaches. Furthermore, its cloud center allows for long-term optimization [5]. Similarly, a deep learning model utilizing time-series data for HRI addresses the challenges posed by movement variations among individuals. By leveraging the Edge-Fog-Cloud infrastructure, this model effectively resolves such issues [6]. Both of these hierarchical systems showcase notable capabilities in the field of HAR.