fast and accurate target search with drones
Department: Electrical & Computer Engineering
Research Institute Affiliation: Center for Machine-Integrated Computing and Security
Faculty Advisor(s): Tara Javidi
Name: Yongxi Lu
Grad Year: 2019
Drones are becoming ubiquitous in our society. Small commercial quadcopter are finding applications in aerial photography, monitoring and surveillance and even e-commerce. However, civilian drones are limited in their size and capabilities. Part of the reason is economics, but safety reason might become a long-term limiting factor. The limited onboard computational powers limits its widespread usage in applications in which high level perceptions tasks are essential. This work addresses integrated control and perception on a mobile platform. As a first step towards a unifying framework, it explores the challenges in the co-design of planning and perception for a drone tasked with finding a small target on the ground. To be successful, the planning algorithm have to recognize the unique characteristics of the latest perception algorithms, namely a convolution neural network. The effect of its non-trivial computational cost as well as persistent noise pattern are considered. This project also integrates the latest developments in efficient perception and test their utility in this application. The future direction of this work is to explore even more efficient planning algorithms that can successfully tackles the challenges revealed, and to explore more planning friendly perception algorithms to enable a broad spectrum of applications.