Tesla announced the latest software update for its self-driving cars on Sunday, revealing that the vehicles will be relying more heavily on radar to make decisions. The announcement comes several months after it was revealed that a driver using autopilot was killed when his Tesla failed to notice a truck stretching across the road in front of him.
All Tesla cars manufactured after October 2014 had already been using radar for autonomous driving, but relied more heavily on the car's optical camera. The update to Version 8.0 significantly increases the role radar will play.
Tesla's announcement came with some interesting insight into how its vehicles will avoid accidents in the future--and apparently make the driving experience a smoother one. It's all about machine learning: Teslas have the advantage of collecting data from the entire fleet and how drivers behave at certain GPS locations, then using that information to make future decisions.
As Recode points out, Tesla's blog post on the update explains that radar can be tricked into thinking there's an obstacle up ahead, when in reality it's a low bridge with a dip under it or an overhead sign at a point where the road inclines. To solve that, Tesla cars that are not on autopilot note whether or not drivers brake at that specific location, then upload that information to the company's database. If several cars drive past something without braking, it's added to what Tesla calls a "geocoded whitelist" of objects that don't require braking.
On the other hand, if the Teslas notice some people are slowing at a certain location, the fleet will add mild braking, even if its optical camera doesn't detect any object. If Tesla's fleet becomes 99.99 percent sure of an obstruction, it will add a full brake for that location. In this way, Tesla creates a map of the world, noting obstructions that it should always break for, plus areas where its radar is and isn't likely to produce false positives.
Tesla seems confident in its fleet's learning ability: "The net effect of this, combined with the fact that radar sees through most visual obscuration," the company wrote in the post, "is that the car should almost always hit the brakes correctly even if a UFO were to land on the freeway in zero visibility conditions."
A big advantage of using radar: It can see past vehicles in its path by sending radar pulsations around them, then detecting the echo that returns. So even if a car in front of a Tesla doesn't notice an obstruction or brakes late or drives right into it, the trailing Tesla should be able to brake ahead of time and avoid both.
Like any form of artificial intelligence, Tesla's fleet will only grow smarter as more people use it. These are relatively early days in Tesla autonomous driving--there are only about 100,000 Teslas currently on the road using this radar technology. The cars' AI it will be at its safest when there are more vehicles to help establish Tesla's maps and its fleet has learned with almost near certainty what does and doesn't constitute an obstruction in the road.
That said, a big part of what makes Tesla a hit with consumers, and what perhaps gives it the best chance to remain relevant over time, is that it's a software company at its core. Having a product that can be updated continually is a compelling sales pitch. Very few vehicles grow smarter as they get older the way Teslas do.
Tesla's Version 8.0 update should be available within two weeks.