Data-driven approaches to self-driving
Self-driving technology promises to improve safety on roads, reduce costs of transportation, and, last but not least, save us a lot of time behind the wheel. Even though first real-world deployments exist, the technology is still far from ubiquitous. I believe that the reason for this is that the decision-making capability of autonomous vehicles, referred to as planning, still relies mostly on complex hand-crafted rules. An alternative would be to embrace machine learning methods, that could improve with an increasing amount of data. In the presentation, I would like to share some examples of my and my colleagues’ work in data-driven approaches to autonomous driving, based on reinforcement learning and imitation learning. What is more important, I will share resources (such as open-source code and dataset), that can enable you to work on these important problems yourself.
Bio
Błażej Osiński is a staff research scientist at Lyft Level 5 and a Ph.D. candidate at the University of Warsaw. His professional experience includes working at Google, Google Brain, Microsoft, and Facebook. He was also the first software engineer at a Berlin-based startup Segment of 1, and senior data scientist at deepsense.ai, an machine learning consulting company. His research explores reinforcement learning, both its fundamentals and applications, in particular for autonomous driving.