University of Glasgow
Task-oriented communication for edge intelligence enabled connected robotics systems
Abstract
dc:description.abstractTraditional digital communication systems are built on the principle of source-channel separation, guided by rate-distortion theory and channel coding. This reconstruction-oriented communication paradigm served as a cornerstone through multiple generations of communication technologies. However, with the rise of machine-to-machine communications and human-to machine interactions, task-specific representations are often more compact and more efficient than full-scale reconstructions, and End-to-End (E2E) trained communication systems have demonstrated superior task performance over traditional communications. This thesis explores task-oriented communication as a paradigm shift from traditional reconstruction-oriented transmission, focusing on optimizing data exchange for machine-driven decision-making rather than full data fidelity. We develop a Task-Oriented Source-Channel Coding (TSCC) framework designed for edge-enabled autonomous driving. By integrating deep learning-based Joint Source-Channel Coding (JSCC) with an end-to-end autonomous driving agent, TSCC minimizes communication overhead while maintaining high inference accuracy, ensuring robustness against noisy channels. Our results demonstrate a 98.36% reduction in communication bandwidth while maintaining driving performance under low Signal-to-Noise Ratio (SNR) conditions. To enhance compatibility with existing digital communication infrastructures, we propose Aligned Task- and Reconstruction-Oriented Communication (ATROC), which bridges task-oriented communication with traditional reconstruction-oriented paradigms. By leveraging an information reshaper and variational information bottleneck (VIB) theory, ATROC improves AI-driven inference on edge servers while ensuring seamless integration with digital communication standards. Experimental results validate that ATROC reduces 99.19% of the communication load while preserving autonomous driving efficiency. Recognizing the need for a holistic approach, we introduce a task-oriented co-design of communication, computing, and control framework tailored for edge-enabled industrial Cyber-Physical Systems (CPS). This framework jointly optimizes data transmission, computational efficiency, and control decisions, and integrates task-oriented JSCC with Delay-aware Trajectory-guided Control Prediction (DTCP) to reduce E2E delay. Experimental results in autonomous driving simulations demonstrate that our co-design approach significantly improves driving performance under high latency scenarios.
Degree
thesis:*- Level dc:type.qualificationlevel
- PhD
- Grantor dc:publisher.institution
- University of Glasgow
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Diao, Yufeng
Subjects
dc:subject × 1Rights
- Language dc:language
- en
Identifiers
dc:identifier.*- DOI dc:identifier.doi
- https://doi.org/10.5525/gla.thesis.85353