My passion in Robotics would not be so defined if I hadn’t taken part in the 2014 Robotics Institute Summer Scholars (RISS) program. At the Robotics Institute, I worked on the BIRD Multi University Research Initiative (MURI) under the guidance of Dr. Drew Bagnell and Dr. Martial Hebert. The goal of the Bird project is to develop technologies to allow fast and safe flight for small UAVs in cluttered environments. As a team our main goal for the summer was to develop a quadcopter that could fly through cluttered environments like forests in a fast manner while avoiding obstacles with the use of a single camera. In this project, I specifically contributed towards implementing a receding horizon controller to enable the quadcopter to follow trajectories. At the end of the summer we presented a poster on our success with the receding horizon controller.
While I worked on my responsibilities, in my free time I read about and became interested in an algorithm called ConSeqOpt which was developed by my graduate advisor, Debadeepta Dey, and Dr. Drew Bagnell. The purpose of the ConSeqOpt algorithm is to reduce the amount of fatal errors caused by perception uncertainties using machine learning techniques. During the summer, I also took part in weekly meetings with Dr. Drew Bagnell’s research group and started to learn about Machine Learning. This was overall a great research experience since it not only improved my knowledge in controls and computer vision but it introduced me to Machine Learning and its applications in robotics. After the experience I coauthored a paper (Preprint) which is currently in review for the 2015 International Conference on Robotics and Automation (ICRA). Because of this experience I later took an online course from coursera.org on Machine Learning from Dr. Andrew Ng, Stanford University.
Below you can see a video of a previous implementation using AR.Drones.
Previous Page Photo credit: http://robotwhisperer.org/wp-content/uploads/2012/08/teaser2.png