Autonomous automobiles have been a hot topic at SXSW, and Google added to the excitement with its session on the subject, led by Chris Urmson, Director of Self-Driving Cars at Google[x]. Urmson discussed the future and past evolutions of the driverless car, with a focus on data collection and the learned behaviors these vehicles are capable of.
The core of the technology developed by engineers for a car to operate autonomously involves an impressive system of sensors, tracking a variety of functions. For example, the location of the car is pinpointed with lasers, radar and cameras, and driverless cars can sense other cars and their respective actions. Data is constantly being collected about the car and the roads and neighborhoods it explores. While driverless cars have covered over 1.4 million miles to-date, and collected data from all of these trips, there’s still ground left to cover, and lessons left to learn.
Inputting data like maps into self-driving cars may help them with geography, but won’t help them understand an obstruction, such as a cyclist or a person in the road. This is one of many roadblocks — no pun intended — to putting autonomous automobiles into mass production. Learned behaviors are much harder to “teach” to an autonomous car. Although when the car sees an action or an obstruction once, it “remembers” how to react to it. Urmson shared some quirky situations the car has witnessed, including a woman in a motorized wheelchair chasing ducks in the road with a broom. The likelihood of that exact scenario occurring ever again is very small, but unlike a human, the car will remember that forever, as well as how it stopped and waited for her to move.
Since the car has to constantly learn and collect data from ever-evolving environments, Urmson said that he believes the technological updates will also come out incrementally. Urmson made one thing abundantly clear in his session: the benefits of self-driving cars outweigh the pitfalls, with a huge net benefit to humanity by cutting down on the amount of driving accidents and deaths.
Written by IEEE on March 15, 2016