“Realworld reinforcement learning in multiagent systems: deploying cooperative autonomy at scale”
ABSTRACT
The everincreasing penetration of level2 autonomous vehicles (AVs) offers an opportunity to reshape the energy efficiency and throughput of our highways. Even at current low penetration rates (15%), we have observed in small settings that adopting different driving behaviors from humans can sharply decrease fuel consumption by eliminating ubiquitous stopandgo waves from traffic. We examined this idea at scale, showing that we can use reinforcement learning to design AV behaviors that operate cooperatively to smooth traffic in large, realistic simulators. We performed a largescale road test, the first of its kind, in which we deployed a hundred of these cruise controllers onto a highway to show traffic smoothing at scale. Finally, we discuss ongoing efforts to benchmark and test autonomous vehicles by building fast simulators populated by RL agents modeling human behaviors.
PRESENTER
Eugene Vinitsky is an assistant professor in Transportation Engineering at NYU, a member of the C2SMARTER consortium on congestion reduction, and a parttime research scientist at Apple. He works primarily on multiagent learning with a focus on its potential use in transportation systems and robotics. At UC Berkeley, where he was advised by Alexandre Bayen, he received his PhD in controls engineering with a specialization in reinforcement learning and received an MS and BS in physics from UC Santa Barbara and Caltech respectively. During his PhD he spent time at DeepMind, Tesla Autopilot, and FAIR.
Details