Autonomous Driving — with Only Cameras and a GPS

U.K. startup Wayve says its machine learning capabilities will dramatically reduce the cost of self-diving systems.

When you’re driving your car, you’re generally making decisions about where you’re going using your eyes — potentially with some assistance from the navigation app on your phone.

An autonomous driving startup from the U.K. this week released footage of a self-driving car that maneuvers essentially the same way: using visual cues — via cameras — and a basic satellite-navigation system.

And they suggest that the method could be the key to finally moving autonomous vehicles from the testing phase to the broader car market.

Wayve, a software firm founded by Cambridge University engineers in 2017, has been testing its technology on nearby roads for more than a year, but this week showed an autonomous Jaguar equipped with end-to-end machine learning capabilities.

Wayve’s system, in effect, learns by imitating human drivers, then improves through interactions with safety drivers that can intervene to avoid violations or accidents.

Although the safety drivers look to be keeping a pretty tight leash on the vehicle, the result will end up being an algorithm that can drive on roads it’s never seen before without HD maps, extensive coding, sophisticated sensors and other expensive gear.

Wayve engineers, in fact, said everything runs on the equivalent of a laptop.

Although the system means Wayve’s testing phase will take longer than competing self-driving projects, engineers argued it would quickly pass its rivals once it is ready to deploy — and predicted it would be the first to put autonomous vehicles in 100 cities.

More in Automotive