AI links taxi drivers in Japan to potential customers

Taxi drivers in Japan cannot tell where potential passengers may be as ride-sharing services, a common way to connect drivers for hire with passengers, are not allowed in Japan, but no longer.

Shin Ishiguro, Data Scientist, NTT DOCOMO, explained how their solution works at the recent NVIDIA GPU Technology Conference.

Source: NVIDIA blog. Picture of a taxi against a blurred background.

Ishiguro and his team take anonymised cell phone signals to infer where people are congregating. Deep learning algorithms running on an NVIDIA DGX-1 supercomputer then map that data to information collected by taxi drivers to forecast demand for taxi services. NTT DoCoMo used stacked denoising autoencoder algorithms, a relatively new deep learning technique, to do the job. 

The result is an app for taxi drivers showing where passengers are likely to be found. Rolled out February 15 this year to 1,350 taxis in Tokyo and another 1,150 taxis in Nagoya, Japan, NTT DoCoMo’s app has steered taxis toward passengers with an accuracy rate of 92.9%. The achievement means drivers who use the app have pocketed 1,409 Japanese yen more for every day of driving compared to taxi drivers who do not.