This spring, a city in the Netherlands will become the first to allow fully autonomous shuttles regularly on its public roads — in the form of a small bus carting people between two towns.
They're called WEpods, and they're only large enough to fit six people comfortably. It's a project of the town of Wageningen, which is in the central part of the Netherlands where farming is big business. The community is using the buses to shuttle visitors in between the towns of Ede and Wageningen (about a 17-minute drive) as well as around its university, a centre for agriculture research.
Autonomous buses will lend it an air of "new, flexible, sustainable and social mobility" for visiting businesspeople and tourists, the project's website explains.
The buses — which are an altered version of those made by Swiss robotics company EasyMile and have been tested in several private projects — won't go terribly fast: They will peter along at roughly 24km per hour. They also won't go very far. And a human will always be watching remotely via camera to make sure nothing goes awry. But it's still a big deal, since it will be the first regular use of totally autonomous shuttle on a public road.
While Google and others have been testing their driverless cars in public for a while now, but they have humans inside in case of emergencies — meanwhile, smaller autonomous prototypes have seen short tests in public, but nothing permanent.
Seemingly anticipating public anxiety, the project's creators launched an online forum where people can ask questions prior to the 30 November launch date. Some of these comments are fairly nuts ("I would feel in such a car as a cookie in the cookie jar, which are short lived."). But another discussion on the forum is actually pretty informative — a postdoc researcher named Joris Ijsselmuiden, who studies robotics and agriculture and works on the project, posted a gif that shows how the pods identify street signs and objects using computer vision.
While of course the buses use GPS data, they also use computer vision to glean information about where the bus is heading independently. Ijsselmuiden explains:
If the accuracy of the GPS system decreases, for instance by trees along the road, it must be helped by landmark detection. Here the cameras detect objects along the way and compare them with known objects from earlier recordings. The position of these objects is known and so the vehicle can calculate where it is located.
It's pretty cool to see this kind of machine learning literally in motion — even if it's only going 24km/h.