Uber simulated a city to teach drivers how to optimize their earnings

Alternative-cab service Uber has been busy doing artificial-intelligence research. The startup has apparently figured out exactly how to position its drivers to maximize their earnings.

In a blog post today, Uber data scientist Bradley Voytek explains how Uber’s “science team” simulated a city and learned that taxi drivers can just stay parked between trips and make twice as much as those who drive around in search of passengers.

It’s fascinating to see this kind of valuable information coming out of Uber — even if it is sort of common sense.

Uber has been hiring data scientists. Meanwhile, competitor Lyft could be looking to boost its own operations through data analysis, given that it hired a vice president of data science from Netflix last December.

To come to the conclusion above, Uber created an artificial city called Uberg. It spans 100-by-100 blocks, existing in the artificial world of Python. Uber drops 250 passengers and 500 drivers in Uberg. Each passenger has a random destination in Uberg.

In Uberg, drivers are simulated in three ways. Type-one drivers remain stationary between trips. Type-two drivers go back to the high-demand hotspots after each trip. And type-three drivers motor around randomly between trips.

Uber also manipulates the dispatch distance, which is “the farthest distance between a passenger and driver where we’ll allow a request to go through,” Voytek writes.

Uber found out through the simulation that when dispatch distance is one block, which is equivalent to a street hail-only system, those who drive randomly get more trips and have fewer lost trips.

However, when the dispatch distance increases to five blocks and drivers receive passenger information in that perimeter, it turns out all the three types of drivers complete the same number of trips.

And we all know that driving around costs money, because gas isn’t free.

Uber discovered that “drivers who are constantly, randomly moving around a simulated city travel 10-20 times the distance compared to drivers who remain stationary or gravitate back toward a demand density between trips,” Voytek writes.

As a result, drivers who get information from Uber on where passengers are end up earning more money than their peers who drive around at random.

This research also teaches drivers when to be stationary and when to go back to a hotspot. “When dispatch distances are very short [which means busy hours] drivers should navigate back toward demand density,” Voytek writes. “However when dispatch distances are relatively longer [which means a less busy time], drivers maximize their earnings by using less gas by remaining stationary between trips.”