Learning Driven Elastic Task Multi-Connectivity Immersive Computing Systems

TL;DR

This work addresses elastic VR task offloading in multi-user MEC systems with multi-connectivity by formulating a constrained stochastic optimization that maximizes computational energy-efficiency under QoE deadlines. It introduces three learning-based decision frameworks—CPPG (centralized), IPPG (independent per-user), and DSMAB (decentralized bandit-based)—that leverage task features, past transmission/response times, and energy history to determine where to compute each elastic VR task. Through trace-driven simulations using real 4G/5G/WiGig traces and 360° video data, the study shows CPPG achieves substantial improvements in latency and energy (e.g., reductions of 28% and 78% over IPPG) and discusses the trade-offs between centralized full observability and decentralized scalability. The work demonstrates how elasticity and QoE considerations can be embedded into MARL and MAB frameworks to efficiently manage edge resources in immersive VR scenarios, with practical implications for next-generation mobile edge computing deployments.

Abstract

In virtual reality (VR) environments, computational tasks exhibit an elastic nature, meaning they can dynamically adjust based on various user and system constraints. This elasticity is essential for maintaining immersive experiences; however, it also introduces challenges for communication and computing in VR systems. In this paper, we investigate elastic task offloading for multi-user edge-computing-enabled VR systems with multi-connectivity, aiming to maximize the computational energy-efficiency (computational throughput per unit of energy consumed). To balance the induced communication, computation, energy consumption, and quality of experience trade-offs due to the elasticity of VR tasks, we formulate a constrained stochastic computational energy-efficiency optimization problem that integrates the multi-connectivity/multi-user action space and the elastic nature of VR computational tasks. We formulate a centralized phasic policy gradient (CPPG) framework to solve the problem of interest online, using only prior elastic task offloading statistics (energy consumption, response time, and transmission time), and task information (i.e., task size and computational intensity), while observing the induced system performance (energy consumption and latency). We further extend our approach to decentralized learning by formulating an independent phasic policy gradient (IPPG) method and a decentralized shared multi-armed bandit (DSMAB) method. We train our methods with real-world 4G, 5G, and WiGig network traces and 360 video datasets to evaluate their performance in terms of response time, energy efficiency, scalability, and delivered quality of experience. We also provide a comprehensive analysis of task size and its effect on offloading policy and system performance. In particular, we show that CPPG reduces latency by 28% and energy consumption by 78% compared to IPPG.

Figures (11)

• Figure 1: Multi-user multi-connectivity edge-assisted AR/VR offloading scheme: AR/VR users can compute their elastic tasks locally or offload them to a nearby MEC unit through one of the $C$ wireless technologies available. Wireless technologies operate on different bands with unique characteristics.
• Figure 2: Task Elasticity: Task computational and communication requirements (measured via task size) and quality of experience (measured via PSNR) increase as the video quality increases, introducing challenges for VR communication and computing systems.
• Figure 3: The CPPG agent follows an actor-critic model. The actor makes a joint offloading decision for all the users. The critic estimates the state value for each of the users in the environment.
• Figure 4: The IPPG agent observes the state of each agent and makes an offloading decision for the newly arrived elastic task.
• Figure 5: DSMAB agent observes contextual information of a user's actions, estimates the value of performing each action, and based on the estimation makes an offloading decision for the elastic task.