Data-driven Learning and Control - DDLC

Capabilities of self-driving cars has surged in the last 20 years, propelled by the promise that, in a very near future, commercial self-driving cars will be safe and perform well. Academia is spurring ground-breaking research (e.g. deep learning) and industry is validating software and hardware extensively with millions of miles being driven on the roads and in simulation.  Yet, final adoption has slowed - primarily due to challenges such as weather and scaling. In this talk, I will present recent research at Cornell addressing these challenges. I will give an overview of the Ithaca 365 dataset, with repeated traversals over diverse scene and weather conditions; and a recent data collection from Hanoi Vietnam, with many scooters. I will then show several research directions that have built on these datasets, including learning via repeated traversals; transforming snowy scenes to sunny; and uncertainty quantification in deep learning.