DRL-based Trajectory Planning and Sensor Task Scheduling for Edge Robotics

Abstract

Mobile Edge Computing (MEC) and Edge Robotics have recently emerged as transformative technologies, revolutionizing industries by enabling real-time processing, decision-making, and automation at the network edge. However, the dynamicity induced by the system’s conditions and specifically the mobility poses a challenge for optimally deciding where to execute a given computational task. As a response, we develop an intelligent algorithm for dynamic sensor task offloading tailored to the unique requirements of MEC-enabled robotic environments. Specifically, we first introduce the environmental dynamics including a sensor task’s end-to-end delay and the robots’ mobility and energy consumption and provide mathematical formulations to model these dynamics. Then, we mathematically formulate the optimization problem and its MDP counterpart and we propose a Deep Reinforcement Learning (DRL)-based computational offloading strategy to jointly optimize Quality of Service (QoS) and energy consumption through robot trajectory planning and fine-grained task allocation. Through hand-picked representative simulation scenarios, we demonstrate the superiority of our proposed mechanism in enhancing the overall system performance, specifically in optimizing task execution, reducing energy consumption, and mitigating transmission delays, compared to various baseline approaches.