PhD Proposal: Towards Autonomous Driving in Dense, Heterogeneous, and Unstructured Traffic Environments

Rohan Chandra
11.29.2021 15:00 to 17:00

IRB 4105

The current success of autonomous driving is limited to highway and sparse urban traffic. This proposal lays out a blueprint for an autonomous driving stack for dense, heterogeneous, and unstructured surroundings. This includes developing new algorithms and systems to perceive, predict, and plan among human drivers in traffic typical of developing nations. There are several characteristics that elevate the difficulty of autonomous driving in such environments. For instance, there is a lack of clear lane markings, presence of non-standard objects e.g. animals, high heterogeneity, and high traffic density area. Furthermore, human drivers in such regions act unpredictably. This includes drivers weaving through traffic as opposed to lane driving as well as drivers breaking traffic rules such as running traffic lights, driving in the wrong lanes, and so on.I will present my work on agent tracking (IROS’19, ICRA’20), trajectory prediction (CVPR’19, RAL/IROS’20), and driver behavior modeling (ICRA’20, IROS’20) in such traffic environments. I will then go over the ideas of behavior-driven planning and navigation in such traffic environments. Collectively, these disjoint ideas can be composed to design a new autonomous driving pipeline specifically intended for dense, heterogeneous, and unstructured traffic.Examining Committee:

Chair:Department Representative:Members:

Dr. Dinesh Manocha Dr. Yiannis AloimonosDr. Pratap Tokekar Dr. Mac Schwager