Before autonomous vehicles can reach a higher level of functionality, they need the ability to not only quickly and accurately evaluate situations, but also make good predictions for future planning, even in new situations. Current algorithms use explicit definitions of traffic rules and models for pedestrian prediction, but in order for autonomous driving to become more robust, the vehicle must be able to extract these rules from the scenes it is exposed to. The goal of this research is to develop an algorithm for scene understanding and prediction that enables the car to determine the rules and relationships between traffic objects in real time.

“I have been doing research at MIT since the spring semester of freshman year, and I have recently become interested in autonomous car research. When I learned about SuperUROP, I thought it was a fantastic opportunity to apply the knowledge I have gained through my machine learning coursework to the field of autonomous driving. I hope to do some great work, and I am excited by the opportunity to possibly publish a paper about my research.”