Friday, October 12, 2018

OSCAR Student Robert Hitt Turns Ambiguous Traffic Scenarios into Autonomous Vehicle Intelligence

My name is Robert Hitt and I’m a senior Computer Science student. My project, which I conducted as a member of an OSCAR Summer Impact Grant team, focused on turning ambiguous traffic scenarios into autonomous vehicle’s intelligence. Autonomous vehicles, like many systems that rely on large-scale sensor data, require extensive training sets to become viable. To that end, we focused on creating a robust, diverse dataset for autonomous vehicles, as well as a labeler to create such a dataset. I applied for this project as I had been interested in the study of AVs and I wanted to contribute what I could to their development. 
In the beginning, we primarily focused on investigating the relevant literature and technologies. We spent several weeks experimenting with various classification techniques, as well as database software. Upon finding a gap in preexisting labeling software, for the next few weeks we focused our efforts on creating our own. Several weeks were then spent just on manually labeling video. I gained a lot of useful technical experience from this project, and I also gained a greater appreciation for the amount of work it takes to train a good neural network.

Throughout this project, beyond technical skills, there were a few main areas I gained some great exposure to. First and foremost, as this was my first research opportunity, I learned a lot about the research process and its many challenges. As I was working on a team, I also further progressed my ability to work towards a singular goal with others. I hope to take this experience and apply it to my future career, whether that be in the industry or in academia.