Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Predictability-Aware Motion Prediction for Edge XR via High-Order Error-State Kalman Filtering

Zhong, Ziyu LU ; Landfeldt, Björn LU ; Alce, Günter LU and Caltenco, Héctor (2025)
Abstract
As 6G networks evolve, offloading extended reality (XR) applications emerges as a key use case, leveraging reduced latency and edge processing to migrate computationally intensive tasks, such as rendering, from user devices to the network. This enables lower battery consumption and smaller device form factors in cellular environments.

However, offloading incurs delays from network transmission and edge server queuing, particularly under multi-user concurrency, resulting in elevated motion-to-photon (MTP) latency that degrades user experience. Motion prediction techniques, including deep learning and Kalman filter (KF), have been proposed to compensate, but deep learning struggles with scalability at resource-constrained edges amid... (More)
As 6G networks evolve, offloading extended reality (XR) applications emerges as a key use case, leveraging reduced latency and edge processing to migrate computationally intensive tasks, such as rendering, from user devices to the network. This enables lower battery consumption and smaller device form factors in cellular environments.

However, offloading incurs delays from network transmission and edge server queuing, particularly under multi-user concurrency, resulting in elevated motion-to-photon (MTP) latency that degrades user experience. Motion prediction techniques, including deep learning and Kalman filter (KF), have been proposed to compensate, but deep learning struggles with scalability at resource-constrained edges amid growing user loads, while traditional KF exhibits vulnerability in handling complex motions and packet loss in 6G’s high-frequency interfaces.

To address these challenges, we introduce a context-aware error-state Kalman filter (ESKF) framework for forecasting user head motion trajectories in remote XR, integrating a motion classifier that categorizes movements by predictability to minimize prediction errors across classes. Our results show that this optimized ESKF outperforms conventional Kalman filters in positional and orientational accuracy, while demonstrating superior robustness and resilience to packet loss. (Less)
Please use this url to cite or link to this publication:
author
; ; and
organization
publishing date
type
Chapter in Book/Report/Conference proceeding
publication status
published
subject
host publication
VRST '25: Proceedings of the 2025 31st ACM Symposium on Virtual Reality Software and Technology
pages
10 pages
publisher
Association for Computing Machinery (ACM)
ISBN
979-8-4007-2118-2
project
6G-FOX: Computing Continuum för att möjliggöra XR applikationsavlastning i 6G nät
language
English
LU publication?
yes
id
14b68cdc-1f25-4f2d-8ab3-5cf7c9a43073
alternative location
https://dl.acm.org/doi/10.1145/3756884.3765973
date added to LUP
2025-12-15 11:28:35
date last changed
2025-12-19 09:16:09
@inproceedings{14b68cdc-1f25-4f2d-8ab3-5cf7c9a43073,
  abstract     = {{As 6G networks evolve, offloading extended reality (XR) applications emerges as a key use case, leveraging reduced latency and edge processing to migrate computationally intensive tasks, such as rendering, from user devices to the network. This enables lower battery consumption and smaller device form factors in cellular environments.<br/><br/>However, offloading incurs delays from network transmission and edge server queuing, particularly under multi-user concurrency, resulting in elevated motion-to-photon (MTP) latency that degrades user experience. Motion prediction techniques, including deep learning and Kalman filter (KF), have been proposed to compensate, but deep learning struggles with scalability at resource-constrained edges amid growing user loads, while traditional KF exhibits vulnerability in handling complex motions and packet loss in 6G’s high-frequency interfaces.<br/><br/>To address these challenges, we introduce a context-aware error-state Kalman filter (ESKF) framework for forecasting user head motion trajectories in remote XR, integrating a motion classifier that categorizes movements by predictability to minimize prediction errors across classes. Our results show that this optimized ESKF outperforms conventional Kalman filters in positional and orientational accuracy, while demonstrating superior robustness and resilience to packet loss.}},
  author       = {{Zhong, Ziyu and Landfeldt, Björn and Alce, Günter and Caltenco, Héctor}},
  booktitle    = {{VRST '25: Proceedings of the 2025 31st ACM Symposium on Virtual Reality Software and Technology}},
  isbn         = {{979-8-4007-2118-2}},
  language     = {{eng}},
  month        = {{12}},
  publisher    = {{Association for Computing Machinery (ACM)}},
  title        = {{Predictability-Aware Motion Prediction for Edge XR via High-Order Error-State Kalman Filtering}},
  url          = {{https://lup.lub.lu.se/search/files/235903749/3756884.3765973.pdf}},
  year         = {{2025}},
}