Using data from 3 million taxi rides, the new algorithm works in real-time to reroute cars based on incoming requests, and can also proactively send idle cars to areas with high demand—a step that speeds up service 20%, according to Rus.

Instead of transporting people one at a time, drivers could transport two to four people at once, results in fewer trips, in less time, to make the same amount of money. A system like this could allow drivers to work shorter shifts, while also creating less traffic, cleaner air and shorter, less stressful commutes.

To our knowledge, this is the first time that scientists have been able to experimentally quantify the trade-off between fleet size, capacity, waiting time, travel delay, and operational costs for a range of vehicles, from taxis to vans and shuttles. What’s more, the system is particularly suited to autonomous cars, since it can continuously reroute vehicles based on real-time requests.

—Daniela Rus

While the concept of carpooling has been around for decades, it’s only in the last two years that services like Uber and Lyft have leveraged smartphone data in a way that has made ride-sharing a cheap, convenient option. (In 2015 Lyft reported that half of its San Francisco trips were carpools.)

However, existing approaches are still limited in their complexity. For example, some ride-sharing systems require that user B be on the way for user A, and need to have all the requests submitted before they can create a route.

In contrast, the new system allows requests to be rematched to different vehicles. It can also analyze a range of different types of vehicles to determine, say, where or when a 10-person van would be of the greatest benefit.

The system works by first creating a graph of all of the requests and all of the vehicles. It then creates a second graph of all possible trip combinations, and uses a method called “integer linear programming” to compute the best assignment of vehicles to trips.

After cars are assigned, the algorithm can then rebalance the remaining idle vehicles by sending them to higher-demand areas.

A key challenge was to develop a real-time solution that considers the thousands of vehicles and requests at once. We can do this in our method because that first step enables us to understand and abstract the road network at a fine level of detail.

—Daniela Rus

The final product is what Rus calls an “anytime optimal algorithm,” which means that it gets better the more times you run it.

Resources

• Alonso-Mora, Javier and Samaranayake, Samitha and Wallar, Alex and Frazzoli, Emilio and Rus, Daniela (2016) “On-demand high-capacity ride-sharing via dynamic trip-vehicle assignment,” Proceedings of the National Academy of Sciences