Towards Sustainable Next-Generation UxV Networks: Joint Optimisation of Communication, Control, and Energy Management

Empowered by exceptional generalisability and autonomy, Unmanned Vehicles (UxVs), including unmanned aerial, terrestrial, surface, and aquatic systems, have emerged as promising tools capable of performing complex tasks ubiquitously. They are highly valuable in various applications, such as environmental monitoring, search and rescue, and logistics. Due to their autonomous nature and exceptional flexibility, UxVs possess the capability to independently access remote and hazardous locations that are unreachable by humans. However, individual or homogeneous UxVs are insufficient for performing complex tasks due to their limited onboard resources and restricted coverage capabilities. Hence, cooperation among multiple heterogeneous UxVs is essential to optimise task execution performance, while improving adaptability and robustness. This cooperation fundamentally depends on reliable inter-UxV communications, while providing robust and stable network functions for heterogeneous UxVs is extremely challenging for the following reasons. First, the inherent dynamic nature and high mobility of UxVs induce frequent changes in network topology and connectivity. These fluctuations impose significant network management overhead and hinder network robustness and scalability. To address this challenge, this thesis firstly proposes the SDUxVN architecture, which enables flexible network management. The separation of the control and data plane allows for more efficient network configuration and dynamic traffic management. The proposed distributed unmanned control plane significantly improves the QoS of the uplink communication under this architecture. Extensive simulations across diverse scenarios demonstrate that SDUxVN substantially improves network autonomy, robustness, and scalability. Second, different types of data within UxV networks possess heterogeneous optimisation objectives, resulting in distinct metrics, such as latency, AoI, and throughput. Meanwhile, energy efficiency and task-related metrics. Meanwhile, preferences among energy and QoS dynamically evolve with network states, further complicating the optimisation. Finding a dynamic trade-off among multiple objectives in such complex network environments significantly exacerbates the complexity of network optimisation. To tackle such optimisation complexity, this thesis introduces advanced reinforcement learning, specifically MODDPG and its generalised MAMODDPG. These algorithms effectively achieve dynamic trade-offs among multiple objectives in linear time under any given preference setting with both single- and multi-agent contexts. These two learning algorithms provide adaptive and efficient MO optimisation for dynamic UxV network management. Finally, UxVs are frequently deployed in infrastructure-limited and resource-constrained scenarios. This highly dynamic and resource-constrained environment often leads to network disruptions and increased data transmission latency. To address this problem, the thesis further explores two specific UxV network optimisation challenges, categorised into uplink and downlink contexts. For uplink optimisation, the focus lies on managing control and data traffic to enhance control traffic reliability and data throughput. For downlink optimisation, given the energy-constrained nature of UxVs and the difficulty in replenishing their energy in harsh deployment environments, this thesis aims at achieving sustainable and energy-efficient data transmission. Multi-objective learning algorithms are applied to dynamically optimise data communication and energy harvesting, enabling a more sustainable and resilient UxV network. In summary, this thesis establishes fundamental methodologies and technological pathways towards autonomous, robust, and energy-efficient unmanned vehicle networks, opening numerous avenues for future research and practical applications.<p></p>

Author dc:creator

• Zhuhui Li (21057956)

• UxV Network
• Software-defined Network
• Reinforcement Learning
• Multi-Agent Systems
• Multi-Objective Optimisation
• Simultaneous Wireless Information and Power Transfer
• Energy–Information Trade-off

Statement dc:rights

• All rights reserved
• Open Access after 2027-10-20

Identifier

10779/exe.32064678.v1

OAI identifier oai:identifier

oai:figshare.com:article/32064678