Activity and Trajectory Reasoning based on Deep (Reinforcement) Learning
The Distributed Robotics Lab at MIT CSAIL is contributing to the development of self-driving cars within the Toyota-CSAIL joint research center. Our work addresses the full scope of challenges in the development of this new and exciting technology, involving theoretical and applied works on decision making, perception, and control.
Deep learning has been successfully applied to different aspects of the autonomous driving task such as lane and vehicle detection as well as full end-to-end control. In this project, we are interested at enabling safe autonomy by focussing on robust behavior models of agents surrounding an autonomous vehicle (such as other vehicles or pedestrians). The main tasks of this project involve developing novel and implementing existing algorithms for trajectory prediction of heterogeneous traffic agents. The UROP will also involve the implementation and development of full data-pipelines and model evaluation on standard benchmark datasets.