Min Li Chan, Jaime Waydo, Kimberly Toth, Wan-Yen Lo
Despite the fact that this talk was at the tag end of a long day, it was one of the most heavily-attended. Having learned through experience this morning that people start queuing up for sessions really quickly, I armed myself with a laptop and one of the ice-creams that were being handed out, and prepared to stand in line 45 minutes ahead of time. There were tons of other people who had had the same forethought, and already the line snaked half-way round the building.
Jaime Waydo kicked off the talk with a simple illustration to show that the concept of self-driving cars isn't new - there were ads, even way back in the 1950s, that showed a car driving itself while the passengers sat gathered round a table, playing board games. Its only recently though that things have progressed enough to start putting the idea into execution. One reason X is so passionate about the project is because the stats showing that humans are accident-prone are overwhelming. Over 1.2 million people die every year in road accidents, and more than 90% of those are due to human error. Solution: take humans out of the equation. There's an additional, equally compelling reason: self-driving cars would mean that thousands of people over the world who have lost their mobility, due to various reasons ranging from failing eyesight to old age, would be free to move around again.
X already has a couple of prototype cars ready that are being tested out in different places like Seattle, Austin and Mountain View. (Fun fact: the team has different nicknames for the cars, Marshmallow Car and Bubble Car being the current favorites.)
One of the first and most complex problems the team had to solve was how to pinpoint exactly where the car was at any point in time. Using GPS wasn't accurate enough, so the team decided to use a system of prior-made maps and sensors. Next, they needed to figure out how to detect both dynamic and stationary objects encountered on the road. For this, the car is equipped with several sensors that give it a 360-degree view up to a distance of two football fields. The car also needs to be able to predict what objects around it will do next - for example, if it detects a construction zone ahead, with some lanes closed off, it should be able to figure out that cars ahead of it will soon make lane changes. This is where machine-learning algorithms come into play. The project also needs several mechanical engineers to work on a myriad of things, from back-up systems for steering and braking, to computer systems specifically for self-driving, to the various sensors used.
The team then showed videos of scenarios they encountered while out on the road on test drives - in one instance, the car stopped for a group of people jumping across the road, while in another, a cyclist suddenly became unnerved and made an abrupt U-turn (a human driver narrowly missed him, but the self-driving car was able to come to a halt at a safe distance).
The session then became a little more interactive, with audience members invited to put themselves into the shoes of the self-driving car team to solve different problems the car might encounter - snowstorms, crowded pedestrian crossings, seasonal changes, and so on.
More information on the self-driving car project is available here: https://www.google.com/selfdrivingcar/
(This post was syndicated from http://kodbalekaapi.blogspot.com/)