Dynamically optimized task-offloading in a UAV cloud-IoT environment

Unmanned Aerial Vehicles (UAVs) are increasingly used within a Mobile Edge Computation (MEC) network where the UAV is made intelligent by equipping it with an edge device (NVIDIA Jetson devices for example). Tasks such as autonomous navigation and obstacle avoidance, but also real-time deep learning or next-best-view applications, can be increasingly applied on the edge in a real-time fashion. In a global strive to make UAVs more autonomous, a variety of tasks are ideally carried out simultaneously. However, edge devices are computationally restricted in terms of CPU, GPU and storage and, therefore, co-existing tasks are competing for scarce computation resources. Moreover, tasks are becoming more time-sensitive and, therefore, more sensitive to communication failures or fluctuating bandwidth capacities. Efficient task offloading is when tasks are carried out either locally or on an (edge) cloud based on user-defined optimization criteria.

The aim of this topic is to design an optimized task-offloading framework that allows multiple tasks to be carried out during a UAV mission. Here, particular attention should be given to resource consumption on the UAV (in terms of battery and computational overhead), on the cloud (in terms of computational consumption) and the speed of execution. In addition, special attention can be given to dynamically changing conditions during a flight, such as the detection of objects of interest, obstacles, or changes in the flight path.