IST Lunch Bunch

Tuesday October 22, 2019 12:00 PM

Scalable Autonomous Driving in the Urban Core

Speaker: Peter Carr, Argo AI
Location: Annenberg 105

Level 4 self-driving cars have the potential to transform society as we know it, once they reach a level of maturity which allows operating at scale without safety drivers. Their benefits include increased safety and accessibility while allowing driving time to become productive time.

Argo AI was founded at the end of 2016 to develop a scalable self-driving system. Working closely with our partners, such as Ford Motor Company, we have grown the team into a 500+ person company with four development offices and daily testing and data collection with our third-generation fleet of more than 100 cars in five cities. 

We test in these five cities — some of the most challenging urban cores in the US  — in order to avoid overfitting our system to any one location. For example, in Pittsburgh, our cars deal with narrow streets and complex geometries representative of many older US cities in the north east. In Miami, our cars negotiate an extraordinarily diverse mix of international driving styles, including wheeled and walking pedestrians. In fact, Florida State Senator Brandes called Miami the “double black diamond of driving.” In Washington D.C., our cars face heavy traffic conditions with complicated road signals and changing traffic patterns. 

In this talk, we will give an overview of our learnings from operating in these cities and we will explore the core challenges to self-driving in the urban core. We will also give our view of end-to-end learning versus a modular autonomy stack as well as the challenges of scalable development, including live operation versus simulation and how these ultimately relate to testing and validation, which is as much a part of the technical challenge as algorithm development is. I will also highlight our efforts to help advance the core technology towards solving the general problem of enabling cars to drive anywhere, anytime. This includes our publicly released dataset www.argoverse.org, and the upcoming competition we are sponsoring focused on perception and forecasting.

Series IST Lunch Bunch

Contact: Diane Goodfellow diane@cms.caltech.edu