Selected Publications

Flow is a new computational framework for training autonomous vehicles to optimize traffic conditions. We show that deep RL controllers trained in this framework exceed the performance of hand designed controllers in stabilizing human driving on a ring.

Recent Publications

. Emergent behaviors in mixed-autonomy traffic. Emergent behaviors in mixed-autonomy traffic, 2018.


. Flow: Architecture and Benchmarking for Reinforcement Learning in Traffic Control. 2018.




Flow is a pythonic simulator/control environment for assessing the effects of vehicle/traffic light control strategies on traffic. It interfaces SUMO (a traffic simulator) with RLlib/RLlab (reinforcement learning libraries) to enable researchers to train controllers for improving traffic objectives such as average speed or throughput.

Recent Posts

TLDR; Create more opportunities for open discussion. Help with heuristics to identify who you want to talk to/listen to.


TLDR; If you’re writing a textbook, add summary sections illustrating clearly what is and isn’t essential to memorize, indicate which exercises are useful and in what way they’re useful, and attempt to add a narrative.


Past Projects

Ion Trap Dynamics

We examine the motions of particles in quadrupole ion traps as a function of damping and trapping forces, including cases where nonlinear damping or nonlinearities in the electric field geometry play significant roles. We report the discovery of a new collective behavior in damped 2D microparticle ion traps, where particles spontaneously assemble into a remarkable knot of overlapping, corotating diamond orbits, self-stabilized by air currents arising from the particle motion.

Phase Evolution of BaReH9 under Pressure

We present a study of the high-pressure behavior of BaReH9, a novel hydrogen-rich compound, using optical, Raman, and infrared spectroscopy as well as synchrotron x-ray diffraction. The x-ray diffraction measurements demonstrate that BaReH9 retains its hexagonal structure on room temperature compression up to 40 GPa. Optical absorption shows the absence of a gap closure to 80 GPa. Raman and IR spectra reveal the pressure evolution of a newly observed phonon peak, and large peak broadening with increasing pressure. These data constrain the disorder present in the material following the P-T paths explored.