Fang, Z., Wang, J., Ma, Y., Tao, Y., Deng, Y., Chen, X., & Fang, Y. (2024). Robust Task-Oriented Communication Framework for Real-Time Collaborative Vision Perception. IEEE Journal on Selected Areas in Communications.
This paper aims to develop a robust communication framework for multi-unmanned ground vehicle (UGV) systems to achieve accurate and timely collaborative vision perception, particularly in object detection tasks, under the constraints of limited bandwidth and dynamic environments.
The authors propose a Robust Task-Oriented COMmunication (R-TOCOM) framework that operates in three phases: idle, calibration, and streaming transmission. R-TOCOM utilizes a Re-ID-based self-calibration technique during the deployment phase to address extrinsic parameter variations. In the streaming phase, it employs an Information Bottleneck (IB)-based encoding method to optimize data transmission based on task relevance. Additionally, an adaptive scheduling mechanism reduces redundancy, and a multi-view fusion network with channel-aware filtering enhances robustness against data loss.
The R-TOCOM framework effectively addresses the challenges of calibration inaccuracies and communication constraints in multi-UGV collaborative perception systems. It demonstrates significant improvements in object detection accuracy and communication efficiency compared to conventional methods, highlighting its potential for real-world applications.
This research significantly contributes to the field of collaborative perception in multi-robot systems. The proposed R-TOCOM framework offers a practical solution for enhancing the performance and robustness of real-time object detection in challenging environments with limited bandwidth, paving the way for advancements in autonomous driving, surveillance, and other related applications.
The paper primarily focuses on pedestrian detection as a representative task. Future research could explore the framework's applicability and performance in more complex scenarios involving diverse object types and dynamic environments. Additionally, investigating the integration of other communication technologies, such as 5G/6G and edge computing, could further enhance the framework's capabilities.
A otro idioma
del contenido fuente
arxiv.org
Ideas clave extraídas de
by Zhengru Fang... a las arxiv.org 10-08-2024
https://arxiv.org/pdf/2410.04168.pdfConsultas más profundas