Trajectory Scheduling of Multi-UAV System in Infrastructure Deficient Environment

Abstract

The use of unmanned aerial vehicles (UAVs) is rapidly growing in research, particularly for surveillance and communication in areas without developed infrastructure. Their versatility allows for a wide range of applications, including remote sensing, traffic monitoring, and target tracking. By deploying a network of multiple UAVs, extensive areas can be covered efficiently, enabling synchronized operations that are both quick and cost-effective. This is especially crucial for real-time monitoring tasks where consistent and reliable communication is key to maintaining high-quality service. However, when it comes to real-time monitoring tasks, uninterrupted and reliable communication channels become crucial to maintain a high Quality of Service (QoS). This continuous connectivity is essential for the effective and seamless functioning of UAV-based systems, especially in scenarios that demand constant and accurate data transmission. This thesis introduces a multi-UAV system designed for efficient data collection in resource-limited settings. The multi-UAV system is comprised of two types of UAVs: the Access UAV (A_UAV) and Inspection-UAVs (I_UAVs). These UAVs differ in terms of their operational capabilities and maneuverability in the environment. The A_UAV serves as a central access platform, coordinating the data collection efforts of I_UAVs, each equipped with a visual sensor for capturing and relaying data to the cloud. This system is engineered to optimize the trajectory of both A_UAV and I_UAV , ensuring data is collected from designated points in a decentralized fashion. For optimizing the trajectories of the UAVs, this thesis introduces the Distance and Access Latency Aware Trajectory (DLAT) optimization specifically for the A_UAVs. This optimization method plays a crucial role in balancing the trajectory planning with the need to minimize the consumption of total system energy for end to end data offloading from I_UAVs to the base stations. In addition, a Lyapunov-based online optimization strategy is employed to ensure the stability of the system, particularly focusing on the average queue backlogs that is critical for dynamic data collection. To facilitate effective coordination between the I_UAV and A_UAV, the system incorporates a message-based mechanism. This aspect is essential for ensuring that data collection and transmission are synchronized and efficient. Further, the thesis delves in the aspect of Age-of-Information (AoI) of the data being collected. A Deep Reinforcement Learning (DRL) framework-based model is conceived utilizing an actor-critic deep network for learning the optimal policy for the A_UAV s to minimize the AoI of the data. The AoI problem is mapped to the Markov Decision Processes (MDP) with a curated reward function to solve trajectory scheduling for the A_UAV. RL provides a robust framework for modeling the decision-making process, considering the stochastic nature of UAV environments and various parameters of the state space such as location, battery levels, and environmental factors. Experiments are performed against multiple baselines with different parameter settings and multiple seeds. The proposed approaches in this thesis have shown improved performances against the available baselines and the methods prevalent in the literature.