Path Planning and Control of Drones using Deep Learning through Cloud Computing 2019-01-0920
Drones or unmanned aerial vehicles (UAVs) are increasingly becoming a mode of choice for several applications such as security, surveillance, rescue operations, package delivery, agricultural monitoring and so on. Drones capable of navigating themselves equipped with functionalities of path planning and obstacle detection and avoidance can assist such operations with less human supervision. Vision-based navigation is found to be a promising solution for UAV navigation in recent years. Application of deep learning algorithms to navigation tends to give good results due to its excellent capabilities to interpret complex data acquired in a real environment. This paper aims to compute path planning for drones taking obstacle detection and avoidance into account using deep learning algorithms. Considering the limitation of resources in a drone for storage and computation the learning framework and its associated functionalities can be done in a computing cloud infrastructure. Firstly, a suitable deep learning network is selected and trained to get the required results in drones in a cloud computing platform. Performance of networks like Convolutional Neural Networks (CNN), Recurrent Neural Networks (RCNN) or Deep Neural Networks (DNN) will be examined in accordance with path planning and object detection. Secondly, we discuss the challenges of using deep learning algorithms in drones through cloud computing in accordance with performance and latency. Finally, this paper discusses the scope of combining deep learning results with model predictive controllers for motion control of drones.