Live data from Hacker News

The View from the Front Seat of the Google Self-Driving Car

medium.com

281–286 of 286 posts

Re: The View from the Front Seat of the Google Self-Driving Car

#281
post #183

Earlier quoted context omitted.

I think it is extremely likely that hailing and renting a self driving will be much, much cheaper than owning, maintaining and insuring your own vehicle.

I have many friends who buy cars from the $500-$1000 range and then drive them for a few years, fixing everything themselves using replacement parts. The cars are usually 15-20 years old. The insurance is $100-$200 a year. If they drive daily,do you think anyone will rent them a brand new, technology packed car for what is effectively $20 a month? HN lives in this bubble when everyone can afford a new car, or at leas…

>do you think anyone will rent them a brand new, technology packed car for what is effectively $20 a month?

Self driving cars won't always be new though. The price of the technology will drop rapidly as they become mass produced. And there will always an economic advantage to renting vs owning. The cost of the car is effectively divided over several people instead of one.

Re: The View from the Front Seat of the Google Self-Driving Car

#282

Earlier quoted context omitted.

What completely infuriates me is pedestrian crossings that don't automatically allow pedestrians to cross when the lights are in a state that would allow it.

Are you asking why peds are required to press a button at some intersections for a walk signal? If so, consider that some intersections may require a long time (~45 seconds) for a ped to cross. When no one is crossing, this may be sub-optimal and a shorter time for that specific light combination should be used.

I'm asking why, if the lights will cycle in such a way that pedestrians COULD cross, why they should not automatically allow pedestrians to cross. It's infuriating to miss a cycle as a pedestrian, and whom does it benefit? I'm not suggesting that there not be buttons which can force a cycle or accelerate a cycle. How would motorists feel if traffic lights defaulted to red?

Re: The View from the Front Seat of the Google Self-Driving Car

#283

Earlier quoted context omitted.

I think the existing auto fleet can be overcome fairly easily, albeit by a company that enters the market in a position that's initially losing. The model I see that will remove individually-owned, occupant driven vehicles will some variation of: My company will purchase your auto and credit you some amount of haulage on my autocar network, and perhaps some cash. For the low income people using sub-$1,000 cars a toke…

Where do parents fit in this utopia of yours? I've got 2 car seats and a pram and some toys in my car in the driveway. Personally owned transport makes more sense for people who have to move things other than themselves and a laptop (tools, babies, etc)

Utopia? Did you read the last sentence of my comment? I think a self-driving car future is a future with disempowered individuals in a neo-feudal transportation world. It sounds terrible to me, but I think it's likely to happen (at least in the US). There are practical and economic advantages in that future and some large existing economic players who can profit from it.

I'd tend agree with your comment in the short term. Moving my daughter's child safety car seat would be a pain, for sure. Longer-term I would expect to see modular standards emerge for child safety car seats. You lock your child's seat into the standardized base that's hiding under a seat cushion in the back of the car. There are already stroller / pram systems that incorporate the child safety car seat, so I could expect that market to boom, too.

Moving things besides yourself is something a lot of people in urban areas already do with public transit. With a private autocar service it's more likely that you'd get more space than on a subway or bus, so I'd think it would be even easier to move more than just yourself. I would expect a rental option to let you keep the car with you, as opposed to it returning to the fleet when you reach your destination.

Re: The View from the Front Seat of the Google Self-Driving Car

#284
post #66

Earlier quoted context omitted.

In all seriousness though, what would happen in this case? Do self-driving cars have emergency stop procedures? And how safe are they? This is still a serious concern about self-driving cars for me. Especially since digital image recognition still isn't that great. What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode?

> Especially since digital image recognition still isn't that great. I believe digital image recognition is better than human, and a lot faster to boot (hah!).

You should take a look at this: http://fastml.com/what-you-wanted-to-know-about-ai/. Digital image recognition isn't that good

Re: The View from the Front Seat of the Google Self-Driving Car

#285
post #66

Earlier quoted context omitted.

In all seriousness though, what would happen in this case? Do self-driving cars have emergency stop procedures? And how safe are they? This is still a serious concern about self-driving cars for me. Especially since digital image recognition still isn't that great. What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode?

> Do self-driving cars have emergency stop procedures? Yes, of course! Just staying on a road is a piece of cake. Most of the research for self-driving cars is dealing with unexpected cars and pedestrians. > And how safe are they? According to the article, 1.7m miles driven without at at-fault accident. Self-driving cars drive very conservatively. > What if the car sees a tumbleweed that it mistakes for a boulder and…

I guess what I'm asking is are the emergency stop procedures safe? If you're going 75mph, how much distance does it need to stop without rolling or harming the passengers?

Re: The View from the Front Seat of the Google Self-Driving Car

#286
post #284

Earlier quoted context omitted.

> Especially since digital image recognition still isn't that great. I believe digital image recognition is better than human, and a lot faster to boot (hah!).

You should take a look at this: http://fastml.com/what-you-wanted-to-know-about-ai/ . Digital image recognition isn't that good

I read over it, but there are a couple of problems with that. Firstly, I never said that computers understand the images, just that the recognition is better than human. When they get it wrong, they do tend to get it wrong in ways that look ludicrous to people. Secondly, Stanford didn't win the 2014 ImageNet challenge: GoogLeNet did.

However, I was still wrong. Have a look at the summary paper from ImageNet: http://arxiv.org/abs/1409.0575

Page 31 states: "Annotator A1 evaluated a total of 1500 test set images. The GoogLeNet classification error on this sample was estimated to be 6.8% (recall that the error on full test set of 100,000 images is 6.7%, as shown in Table 7). The human error was estimated to be 5.1%."

There are some bits there where GoogLeNet did better than human, but it was usually on specific classifications (what breed of dog is that, for example). Generally, a human is slightly better.

It's worth noting, however, that the error rate was halved from 2013's winners. I reckon the 2015 error rate will beat human, and in following years there will just be no contest. Just like chess programs have gone far beyond human capabilities.

And, of course, with cars we're talking about having multiple images plus depth perception, which makes it a lot easier for the machine.

Post reply on HN