Hacker News Reader

On a related note, here's Chris Urmson's long talk at SXSW on how Google does automatic driving.[1] This is worth watching. He shows what Google cars sense and gives an overview of how they analyze the data. This is the first time Google has released this level of detail. If you're going to comment on automatic driving, you need to see this.

Urmson goes into great detail on Google's 2mph collision with a bus. The sensor data from the Google car is shown. The video from the bus (it had cameras, too) is shown. Exactly what the software was trying to do is discussed. The assumptions the software made about what the bus driver would do are discussed. What they did to prevent this from happening again is mentioned.

Most of the talk is about the hard cases. In the beginning, Google developed a highway driving system, but that was years ago. Now they're working hard on dealing with everything that can happen on a road, including someone in an electric wheelchair chasing ducks with a broom.

[1] https://www.youtube.com/watch?v=Uj-rK8V-rik

Cool! Thank you for sharing this.

Here's the part where he starts talking about the 2mph bus accident [1]. He says the self-driving car should not have moved. He does not say why. He says they ran a lot of tests to make sure it will not happen again.

In plain English, I'd say, when the SDC is stopped on the side of the road, give sufficient time for traffic coming from behind to clear out before the SDC moves forward. Also, if an accident is imminent, being in a non-moving position is probably safer and less likely to cause liability for the SDC than moving. Plus, rolling over some sandbags is probably preferable to being hit by a bus.

[1] https://youtu.be/Uj-rK8V-rik?t=21m45s

> Volvo is very close to shipping their autopilot. They're testing now. In 2017, they will release 100 self-driving cars to customers. They do not require the driver to pay attention. Volvo's CEO says that if a Volvo self-driving car gets into an accident, it's Volvo's fault, and they will take full responsibility.

Wow really? That's either crazier than Elon or they have a really good product. Either way, it should be an interesting couple of years.

> That's either crazier than Elon or they have a really good product.

Or they are just familiar with product liability law and are engaging in the time honored corporate PR tactic of spinning the consequences of the mandates that the law puts on the company as if they were a civic-minded free choice of the company.

Could be. I personally doubt that policy ever sees the light of day. Companies are legally obligated to be financially responsible on behalf of their shareholders. A few executives holding the company over the edge of a cliff named "class action lawsuit" doesn't spell responsible to me. However IANAL

Volvo can buy auto insurance for their self-driving cars, and probably will. As they get more driving experience, their premiums should go way down.

Is that insurance affordable for Volvo in America? Liability payouts here can be huge.

Volvo putting 100 cars on the road in Sweden in 2017 doesn't necessarily mean they're close to shipping autopilot in the US. Insurers may request more testing to give an affordable price.

From what I've read, Tesla isn't insured. My guess is that has either to do with cost or insurers' reluctance to put a value on a brand new feature

From a Fortune article [1],

> Tesla noted that it does not carry insurance against such events. Were the company to be sued because of a death related to, say, its autopilot feature, it could result in a “substantial monetary award” that would have to be paid from “company funds, not by insurance.”

[1] http://fortune.com/2016/07/06/tesla-autopilot-crash-material...

Some of the textbook cases in the US (which, as I understand, actually has weaker consumer protection than a lot of developed nations) on general product liability concern automobile manufacturers liability for harms resulting from defects; its not like if Volvo kept quiet, attorneys who might otherwise recognize the opportunity for a lawsuit (class action or otherwise) wouldn't notice the opportunity if there were harms resulting from Volvo's self-driving system.

It really seems to me that the sales benefit of reassuring customers about Volvo's confidence in their platform is likely to far outweigh any marginal increase in lawsuit risk to Volvo in this kind of statement.

> It really seems to me that the sales benefit of reassuring customers about Volvo's confidence in their platform is likely to far outweigh any marginal increase in lawsuit risk to Volvo in this kind of statement.

Oh, don't get me wrong, I think insuring liability at the manufacturer level is the right thing to do. I just have doubts that it will be affordable any time soon. Animats was saying Volvo is close to shipping their autopilot. I doubt it for the reason of the cost of insurance. I could be wrong.

You can charge customers a subscription for "Volvo self driving insurance", or charge a lot for the car to cover Volvo's costs.

Accidents and lawsuits are more likely to occur at the beginning of self-driving cars' introduction before many people have bought them.

Likely, the first companies to do this will either (a) wildly succeed or (b) fail dismally and teach a lesson to the others. I don't see much middle ground. Again, I could be wrong.

I love the work that Volvo is doing, but I am highly skeptical they will release a stage 4 autonomous vehicle in 1 year's time. You might be buying into their PR the same way everybody else bought into Tesla's PR.

Yes, some of that is PR and yet to be proven. But an appropriate sensor suite for the task of autonomous driving is not really a question. There's a right way to do it, and a wrong way to do it. For robotics researchers, there's really two ways of looking at things. Some say "look, we as humans are proof that the only sensors you need to do autonomous driving are two cameras (out eyes). So the real research question is, how do we make a system that works with just these sensors?" The other camp says "In order to get away with just two sensors, you need human-level AI. We are not there and we're not going to be for a long time, so let's work within our current capabilities and engineer a proper system."

As a community, robotics researchers now know what kind of engineered system works. In the 2007 DARPA Urban Challenge, EVERY car that finished the race had a 360 degree velodyne laser. It was very much a game-changing sensor for autonomous cars.

For Tesla to now say "Oh, we'll do the same thing but without all the fancy sensors" is just incredible. Speaking in terms of the information that the cars acquire, Tesla is driving blind compared to its rivals.

I agree that Volvo has a better approach and from what I can tell the necessary hardware to have a much more robust system. What I'm arguing is that there are too many problems that still need to be solved for them to have a stage 4 autonomous vehicle.

I used to work on self-driving cars, specifically this one - https://www.washingtonpost.com/local/trafficandcommuting/dri... - and while that was a few years ago I know just how many things the technology is fundamentally not ready for, such as adverse weather.

I agree that Tesla took the wrong approach, I was just commenting that I would be amazed if Volvo was that many years ahead of everybody else.

> What I'm arguing is that there are too many problems that still need to be solved for them to have a stage 4 autonomous vehicle.

Oh, I would agree with you there as well.

> I agree that Tesla took the wrong approach, I was just commenting that I would be amazed if Volvo was that many years ahead of everybody else.

I think that's where we were differing. I think Volvo is many years ahead of Tesla only because Volvo is using the state of the art in driverless tech. Other rivals like Google, BMW, and VW are doing the same. On the other hand, Tesla is choosing to use technology the robotics field had proven to be ineffective over 10 years ago.

Agreed, but that's because they're actually selling cars. Most people don't realize a single sensor can cost $20,000 - now imagine you have 10 of them. That's an order of magnitude more expensive than the car.

I'm not trying to play devil's advocate, just pointing out that it's a fiendishly difficult problem. I like what Tesla did, I just dislike how they handled the PR.

Likewise I like what Volvo is doing, and I'm curious how they're going to price it.

> Agreed, but that's because they're actually selling cars.

Actually selling cars to the public is an exceptionally poor excuse for -- as it has been described upthread -- using technology that is known to be inadequate to safe implementation of the feature set.

If a feature cannot be safely implemented at a price point suitable for the target market of the vehicle, then it shouldn't be in the vehicle. Deliberately choosing otherwise goes beyond mere negligence into actual recklessness, IMO. (Note, I'm not saying that Tesla is doing this, I'm saying that the use of the excuse "they're actually selling cars" as a response to the claim of proven inadequacy of the sensor choice made upthread is an inadequate excuse, since if that was the reason and if the sensor choice was proven to be as inadequate as described, then that would be reckless.)

The real issue with Tesla is that they marketed it as a self-driving feature when really they had developed cruise control plus. Lots of manufacturers have had cruise control and lane following with automatic braking for some time, but they didn't hype it the same way Tesla did. I don't think Tesla built a bad system, they just made it out to be more than it was, and given the circumstances I agree it was arguably reckless (it's hard to say how much they directly marketed it and how much was just hype from the public in general, as they obviously can't control users uploading videos of hands free driving on YouTube).

I do think the way they've handled the crashes lately though has been pretty poor - especially spinning bad statistics to justify what happened rather than admitting fault.

> The real issue with Tesla is that they marketed it as a self-driving feature when really they had developed cruise control plus.

Yes, I think that this is part of the real issue, but there is also a real engineering issue. Other vehicles with cruise-control-plus require hands on wheel at all time. Tesla seems to alert you when your hands are off, but the autopilot will continue on merrily. The Tesla that drove under the semi had its roof torn off, the driver decapitated, and it continued driving only until it was stopped by a pole. Also the crash happened in the first place because the AP was essentially blind (it couldn't even see the semi). You and I both know there are ways to tell when you're blind (usually involving orthogonal sensors suites, something the Tesla lacks).

Yes. Your very logical argument ought to be shared more.

Most people are unaware of Tesla's second-rate sensor system. They think it is grade-A because it is called autopilot.

That may not constitute false advertising. It is, at least, highly suspect.

> Most people don't realize a single sensor can cost $20,000 - now imagine you have 10 of them.

True, but part of that cost is an economies of scale issue, and we are finally seeing lasers like the velodyne come down in price. The 64 laser version used in the Urban Challenge cost over $75,000. Today, you can get a smaller version for $7k. Still a lot, but the price is coming down and it's coming down fast. Even putting one of these guys on board would be a huge improvement of sensing capabilities over what they have.

Remember, people are paying $5500 for the autopilot alone and as much as $130k for the whole car. If anyone would be willing to pay a premium for some better sensors, it's Tesla's customers.

Sure, the cost will come down. But comparing the cheaper $7k sensor to the original $90k sensor is disingenuous, as the original is still several times better than the cheaper one in terms of FOV.

I was referring to IBEO sensors which are already used in automotive manufacturing. When these cars are actually commercialized, the sensor suite is still going to cost tens of thousands of dollars.

Anyway, I'm sick of coming across as so negative. I think self-driving cars are awesome (it's pretty surreal when you actually sit in one driving around a city) and I look forward to them becoming more common. I just think it will take longer, and cost more, than most people like to admit.

> Anyway, I'm sick of coming across as so negative.

Oh quite contrary, I believe your input has added greatly to this discussion. Thank you.

> I just think it will take longer, and cost more, than most people like to admit.

Agreed. Even Google now seems to be tempering their predictions, saying anything from 3 to 30 years. I say, they will be here when then are ready and no sooner. Don't rush it.

> I like what Tesla did, I just dislike how they handled the PR.

Are they not being a bit irresponsible unleashing a beta feature on the public and using them as guinea pigs? Do you feel they should require hands on the wheel more than once every few minutes?

These are design decisions to (a) collect more user data and (b) enable the easiest way for users to abuse the system.

> For Tesla to now say "Oh, we'll do the same thing but without all the fancy sensors" is just incredible. Speaking in terms of the information that the cars acquire, Tesla is driving blind compared to its rivals.

Wow, there's a headline. You sound pretty knowledgeable about the history. Why not write a blog post? Constructive criticism is a positive thing.

>>They do not require the driver to pay attention.

Completely besides the point. A true self driving car won't have a steering wheel or a break pedal.

Or will disable steering and brake controls when self driving feature is active.