By Kevin McElwee
When requesting a ride-hailing service, you may soon notice something missing: the driver. Fleets of autonomous electric vehicles could someday replace human-powered ride-sharing.
Programming obstacles still stand in the way of this happening on a large scale, but associate research scholar Lina Al-Kanj is tackling the unique challenges with Warren Powell, a professor of operations research and financial engineering.
“Driverless cars are becoming a reality, and we already have electric cars on the market,” Al-Kanj said. Uber and Google’s Waymo each have deals in place for 20,000 or more autonomous cars to hit the roads over the next few years.
Al-Kanj has adapted an algorithm, originally developed by Powell for the trucking industry, to coordinate the actions of thousands of driverless electric cars. One challenge she addressed is the need for autonomous cars to make forward-looking decisions. Currently, ride-hailing apps are nearsighted — they connect the user to the closest car. Al-Kanj’s system takes into consideration the charge level of the battery, the length and destination of the trip, when best to recharge the car, and where to position cars to optimize use of the fleet.
Consider a car near Philadelphia with a battery at half charge that gets pinged for a trip to New York City, roughly 100 miles away. Al-Kanj’s approach, based on Powell’s research on making decisions in the presence of uncertainty, calculates the downstream value of each car based on the projected charge level of the battery after the trip.
Because the vehicles are electric and autonomous, the algorithm must address many other factors that a human driver normally handles: Should the car park and wait for more customers? Should it move to a location with many potential customers? Should it take time to recharge?
These considerations are hard enough to calculate on a case-by-case basis, but when expanded to thousands of cars, the decision-making becomes quite complicated. To handle this complexity, Al-Kanj and Powell developed a framework that scales along with the number of cars in the fleet.
Al-Kanj’s model allows companies to identify the best locations for recharging facilities, the value of faster charging devices, and the optimal size of batteries for fleet operators.
Al-Kanj hopes that ride-hailing companies will take note of her research, which is funded by the Andlinger Center for Energy and the Environment, so that her logic as well as her strategic simulator can assist fleet operators in real-time dispatch.