Microsoft Uses Cabbies and Fancy Math to Provide Better Driving Directions Microsoft, Cabbies Team Up to Improve GPS Directions
How good is your GPS in city traffic? Certainly not better than a cabbie 20 years on the job. That's the premise of ingenious project called T-Drive.
Does your GPS know how to navigate city traffic? Hardly. And certainly not better than a cabbie 20 years on the job. That's the premise of an ingenious project at Microsoft Research called T-Drive.
Anyone who has ever driven in a city knows there's often a faster route—down a side street, or a bit longer on the odometer but far speedier, maybe the lights are longer. Anyway, GPS isn't so good at making the call to get off the highway and hit the back streets. So, researchers Xing Xie and Yu Zheng turned to the experts: over 33,000 cabbies in China. They monitored the GPS locations and times of taxis over three months to determine which routes were the fastest, essentially mining their collective intelligence and human knowledge to defeat the machine algorithm using something called Variance-Entropy-Based Clustering. The data revealed which stretches of which roads were consistently chosen, and which were avoided and when.
Right now GPS, Mapquest, and the rest of them use distance and posted speed limits to choose your route. That's about as useful in dense cities as asking a tourist for directions. So there's clearly room for improvement. In the end, Xing and Yu were able to build a model that consistently beat traditional GPS directions by about five minutes for a 30-minute ride. Not bad.
They've made a prototype available for members of Microsoft Corp to test out. To have this work in U.S. cities, they'd have to repeat the massive data gathering undertaking in each city. That's only possible if the expert drivers are GPS equipped—and don't follow the suggested directions. It's probably worth the effort, though. After all, time is money, and five minutes per ride adds up.
Via MIT Technology Review. Image: Microsoft Research