Google video showing a new level of robocar operations


Today Google released a new 3 minute video highlighting advanced self-driving car use. Here I embed the video, discussion below includes some minor spoilers on surprises in the video. I'm pleased to see this released as I had a minor & peripheral role in the planning of it, but the team has done a great job on this project.

This video includes active operation of the vehicle on not just ordinary streets, by private parking lots for door to door transportation. You can click on it to see it in HD directly on Youtube.

While much of the focus of my robocar talk has been on the benefits for mainstream society, there is a segment of society that is already chomping at the bit for robocar technology, namely those who can't drive because of a disability. The numbers of such people are quite large. They include not just those disabled by illness or accident, but also a fair percentage of the aging population. 21 states now have rules under which they strip people of licences if they fail to past vision or driving tests beyond a certain age.

It doesn't happen a lot right now because the lobbying power of seniors is quite strong and set against it. But many seniors voluntarily stop driving when they notice their ability fading. Very common is "night blindness" which causes many seniors to drive durign the day but avoid it at night. Hard statistics on this are difficult to find, but even the figures for more serious blindness are quite large, and 21% of seniors no longer drive. Seniors in general cut their total miles in half from when they were middle aged, and women cut them even more. The National Eye Institute says that over 3 million in the US are blind or vision impaired, but only about 300,000 of those are under 60.

Vision impairment is just one type of disability of course, and a wide variety of other factors limit driving, including being in a wheelchair, parkinson's, epilepsy and many more. As robocars improve these folks may well be first in line (if their disability hasn't left them with no income.)


Is this legal? I had the impression that the law required a licensed driver in the driver's seat.

This was shot in California, but as you will note at the end of the video, police supervised the exercise to assure it was safe. That doesn't mean it's fake -- the car does what you see -- just that it's a prototype.

This is exciting, but I'm almost more interested in what Continental (the tire people) are doing, which you also posted about. They have a 90% solution which is much cheaper than Google's 100% solution. As we know from the computer industry (including free software), 90% but much cheaper can win out over the 100% solution.

Cool video, though.

At least based on the video, the Continental system (as well as all the other announced car company systems from Mercedes etc.) is a wheel-and-pedals cruise control. You must keep your eyes on the road. That's not a 90% solution, though it is still a nice product. I'm not saying the Google car is ready to hand over to a blind person -- it is still a prototype -- but the difference between a system where you can ignore the road and do other things and a self-steering cruise control is much, much greater than the difference between 90% and 100% in my view. Admittedly I have a bias, but I felt this way long before getting involved in the Google project.

This is not to decry efforts to do a vehicle where you can ignore the road using cheaper sensors. They just are not apparent yet.

I think the main issue here is that if a car handles 90% of what we do, it may even lead to more distracted driving.

I think it really has to be a 100% system to work.

I'm wondering if the Google car's lidars will get much cheaper than they are now, as from what I understand they cost $80k.

I have good faith in Moore's Law and volume pricing to solve this problem. Everything electronic gets cheap when there is lots of demand. The lidar is the gold standard because it sees in 3D. It doesn't have to try to split the foreground from the background with heuristics. If there is something in front of the laser it always sees it. 100%. (Well lots of 9s, but if it fails you know it.)

I have read a while ago that there have been advances in stereoscopic robot vision gaining a depth map from the displacement of recognizable feature and patterns. Presumably it wont be quite as accurate as high end lidar but having cameras on each side, + nose and tail of the car would still give the computer much more better depth perception and 360 information than a human with a measly 2 eyes 3 inches apart. Such image processing is quite a parrelizable task for small specialized processors which is good as Moore's law is currently going the direction of more cores rather than higher hertz. I notice the price on small hd cameras is only going down also so all up the components for 3d vision would amount to the equivalent of about 8 flip video cameras and a few graphics cards. Which in mass production should not cost more than a few thousand dollars and you can return you roof rack to its usual purpose.

I am not even close to blind, but I will bet that someday my vision will make me a markedly worse driver. I feel confident that this will be a viable option long before then for me. Computer-driven cars have always been an obsession of mine. Thank you so much for updating this blog. It is always a joy to see the (1) next to your feed at the top of my reading list.


Add new comment