2021/04/22 by Davide Callegaro, Marco Levorato, Callegaro, Davide +3
Computer Science · #Advanced Neural Network Applications #Age of Information Optimization #Distributed #FOS: Computer and information sciences #FOS: Electrical engineering #IoT and Edge/Fog Computing #Networking and Internet Architecture (cs.NI) #Parallel #Systems and Control (eess.SY) #and Cluster Computing (cs.DC) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2105.15105
openalex publication_date 2021/04/22 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Edge computing enables Mobile Autonomous Systems (MASs) to execute continuous\nstreams of heavy-duty mission-critical processing tasks, such as real-time\nobstacle detection and navigation. However, in practical applications, erratic\npatterns in channel quality, network load, and edge server load can interrupt\nthe task flow execution, which necessarily leads to severe disruption of the\nsystem's key operations. Existing work has mostly tackled the problem with\nreactive approaches, which cannot guarantee task-level reliability. Conversely,\nin this paper we focus on learning-based predictive edge computing to achieve\nself-resilient task offloading. By conducting a preliminary experimental\nevaluation, we show that there is no dominant feature that can predict the\nedge-MAS system reliability, which calls for an ensemble and selection of\nweaker features. To tackle the complexity of the problem, we propose SeReMAS, a\ndata-driven optimization framework. We first mathematically formulate a\nRedundant Task Offloading Problem (RTOP), where a MAS may connect to multiple\nedge servers for redundancy, and needs to select which server(s) to transmit\nits computing tasks in order to maximize the probability of task execution\nwhile minimizing channel and edge resource utilization. We then create a\npredictor based on Deep Reinforcement Learning (DRL), which produces the\noptimum task assignment based on application-, network- and telemetry-based\nfeatures. We prototype SeReMAS on a testbed composed by a drone, mounting a\nPixHawk flight controller, a Jetson Nano board, and three 802.11n WiFi\ninterfaces. We extensively evaluate SeReMAS by considering an application where\none drone offloads high-resolution images for real-time analysis to three edge\nservers on the ground. Experimental results show that SeReMAS improves task\nexecution probability by 17 % with respect to existing reactive-based\napproaches.\n