Drone-Aided Autonomous Navigation

Abstract

Recent studies indicate that Autonomous systems will replace most of manual driving by 2020. Proposed systems suffer from delay in decision-making and lack of information, two primary components required to operate safely. For example, Uber’s self-driving car struck a pedestrian while autonomous systems were engaged [6]. In this work, we use a paired drone to relay a larger traffic scene by capturing scenes with a larger field-of-view. We make significant improvements to several Artificial Neural Networks for Lane Detection and Object Detection to increase speed of operation and accuracy. We utilize container virtualization to decouple software and hardware to overcome the limitations of the hardware to run those frameworks. We also remove the need to store an autonomous map completely due to memory limitations of a Raspberry Pi 3 by processing video feed in real-time. In this project, we implement a faster and safer Level 3 Autonomous system by increasing the field-of-view and decreasing the latency in the decision-making process.