Connectivity-Aware Trajectory Planning for UAVs considering Cellular Networks

Sammanfattning

Trajectory planning is an essential element of robotics and autonomous systems. This thesis investigates trajectory planning for Unmanned Aerial Vehicles connected to cellular networks. It is desirable to find trajectories of high quality, such as being smooth and ensuring good network connectivity in terms of bit rates. However, state-of-the-art algorithms might be slow for computing trajectories in the complex and dynamic environments of cellular networks. Furthermore, obtaining a precise model of the cellular network is a challenging task. To address these issues, this work explores the use of a generative artificial intelligence model known as diffusion modelling. A cellular network simulator is employed to create bit rate environment maps, which are utilized in three key steps. First, to create a dataset of trajectories to train the diffusion model, using a good but computationally demanding state-of-the-art trajectory planner. Second, when generating new trajectories using the diffusion model, and third when evaluating the generated trajectories. To consider the non-deterministic nature of cellular networks and to challenge the diffusion model, we perform experiments of varying complexity and realism by altering the cellular network environments in the three described steps. The performance of the diffusion model is compared to a good but computationally demanding state-of-the-art trajectory planner. The results show that the diffusion model consistently generates trajectories 10 times faster than the state-of-the-art trajectory planner, while maintaining a comparable quality level. Utilizing generative artificial intelligence, in the form of a diffusion model, is demonstrated to be a promising approach for fast trajectory planning in complex environments, such as those encountered by Unmanned Aerial Vehicles operating while connected to cellular networks.