Title: Adaptively Serving XR Workloads from Resource-constrained Edge 

Date: Tuesday, Oct 29, 2024

Time: 11:30 AM – 1:00 PM EST 

Location (Hybrid): KACB 3402 and MS Teams (link)  

Jin Heo

Ph.D. Student

School of Computer Science

College of Computing 

Georgia Institute of Technology 

Committee: 

Dr. Ada Gavrilovska (advisor) - School of Computer Science, Georgia Institute of Technology

Dr. Alexey Tumanov - School of Computer Science, Georgia Institute of Technology

Dr. Ashutosh Dhekne - School of Computer Science, Georgia Institute of Technology

Dr. Ketan Bhardwaj - School of Computer Science, Georgia Institute of Technology

Abstract: 

Extended reality (XR) applications demand high computational power for perception and 3D rendering, often exceeding the capabilities of modern mobile devices. Edge computing offers a solution by allowing the offloading of intensive tasks to nearby servers. However, the resource constraints of edge infrastructures present significant challenges in delivering high-quality, multi-user XR experiences. The diverse deployment environments of XR applications often make the current static workload distribution methods suboptimal. Additionally, compression methods for emerging sensor data that do not consider real-time XR constraints compromise the benefit of edge assistance. 

This research proposes that to deliver high-quality, multi-user mobile XR experiences via edge computing, the XR serving system at the edge needs to be adaptive and coordinate the distribution and optimization of the XR workloads jointly with the data transfer, using quality metrics that capture the end-user experience. To address the challenges in edge computing for XR, I introduce a framework for flexibly distributing XR workloads, a fast and lightweight LiDAR compression method for offloading online perception tasks to the edge, and propose adaptive workload optimization techniques for multi-user serving. These include solutions for 3D rendering optimization and an adaptive system for perception scheduling. 

The introduced framework facilitates efficient edge-assisted XR across diverse deployment environments by enabling runtime adaptation of workload distribution. The LiDAR compression method balances compression quality with latency, enhancing real-time LiDAR perception offloading to the edge. The proposed adaptive workload optimization techniques, which consider the user-side quality of service (QoS), improve the multi-user serving capabilities of resource-constrained edge servers while maintaining the QoS. This research demonstrates edge computing's potential to enable the widespread adoption of responsive, high-quality XR applications across diverse scenarios.