During my PhD, I focused on problems that require real-time decision making under uncertainty, with a particular interest in electric vehicle routing and logistics.

Google's DeepMind team has done some impressive work showing that an AI can successfully learn to play lots of Atari games better than humans. We wondered if there was something special about that original set of games, or if we could design our own "Atari games" that this AI could learn to play too.

We designed a game to test this, based on a classical vehicle routing problem.

A short write-up of the work is available on arXiv, and my presentation on the work at the 2019 INFORMS annual meeting is also available on YouTube.

We're a few years down the road, and ridehail companies like Lyft and Uber are operating fleets that consist entirely of autonomous electric vehicles. It's a Thursday night, and bar patrons are starting to request rides home. How should a ridehail company decide which vehicle to assign to the new requests? When should they send vehicles to recharge, which vehicles should be moved around to better serve future requests, and where should they be moved to?

I've proposed AI techniques (deep reinforcement learning) to solve this problem in the work "Dynamic Ridehailing with Electric Vehicles."

A preprint of the work is available here, and my presentation on this topic at the 2019 INFORMS annual meeting is on YouTube here.

If an EV needs to stop and recharge while traveling about and serving customers, where and when should it stop to recharge? And if all the chargers are occupied when it shows up to recharge, should it wait, or should it bail and try its luck somewhere else?

This is the topic of our work EV Routing with Public Charging Stations, which has been accepted for publication in Transportation Science.

A preprint of the work is available here.

Before working in transportation, I used multi-objective optimization to study how climate change might impact natural resource management. For instance, how does climate change impact the US Forest Service's ability to reduce fire hazard without impacting local species? Will more severe climate change mean more harm to local species in order to achieve the same amount of fire hazard reduction?

This problem also served as a platform on which to more generally study how we measure conflict among objectives in multi-objective optimization.

More details about this research are available in my masters thesis here.