Baidu says it's developed China's first fully autonomous self-driving car | Hacker News Reader

Baidu says it's developed China's first fully autonomous self-driving car

venturebeat.com

91 pointsdoppp68 comments

And they'll actually be able to go full-speed on the highway, kill people (less than real drivers), but still continue on in development without massive public outcry!

I wonder, if someone could make a self-driving car that could pass a road test (including taking verbal instructions from the tester), would people be ok with that one being on the highways?

I wonder if people would even show confidence knowing the difficulty of other drivers' road tests, given how easily they can vary based on instructor and region.

> kill people (less than real drivers)

What does this mean?

in china, you can override the law without major consequences...

some day, at full velocity a car will kill someone, and they will discover a bug, patch it , and keep life... in america that would mean a thousand million dollar lawsuit.

The moment you start mass-producing self-driving cars, people will start dying in accidents that were caused by them. Even if they are vastly safer than human drivers^+ , the sheer amount of cars on the highway and the varied situations they get in will mean that there will be accidents caused directly by bad decisions done by the computer. The legal system we have is a lot more lenient to human error than it is to other causes, and this is a major problem, because it means that even if going all AI cars would mean that we lower the fatality rate by 97%, that still means a thousand deaths a year that can be blamed on one company. That company could not possibly continue operating, even if they would be saving 29000 lives every year.

I am generally not a fan of the Chinese government, but this is one way in which their system is better than ours. They have people in positions where they can do the math and decide that if some decision greatly saves lives on the net, they can enact it even if it directly causes (less) lives to be lost.

^+ And really, that isn't a hard goal to reach. People largely don't cause fatalities because they are bad at driving, they cause them because they are impaired in one way or another and because the reaction times at high speeds are too hard for humans. Most early AI cars will probably be shitty, but they will largely be shitty in ways that don't endanger the occupants.

> but this is one way in which their system is better than ours.

That is ludicrous.

> They have people in positions where they can do the math and decide that if some decision greatly saves lives on the net, they can enact it even if it directly causes (less) lives to be lost.

What you mean to say is they have people in positions where they can enact it, for any number of reasons . . . Let's hope the decider guy isn't the air quality monitor for Beijing.

I understand your argument, but I see two other possibilities:

1) Drivers remain liable for actions taken by their car's AI. This would encourage people to only use truly driverless AI in the safest possible conditions, ie not on crowded streets or when its raining. This will let market forces balance the technological development of the cars with public adoption

2) After initial legal precedents for damages are set, some companies will make enough money to insure a large fleet of autonomous vehicles. And again, market forces will be able to pace the tradeoff between how "safe" the AI technology is vs how quickly its adopted.

tldr; an authoritarian government isn't necessary to push this particular technology forward. As much as I hate to say it, the Free Market will do just fine in this case

In China, I'd assume that they would just go fleet and skip autonomous personal vehicles. Basically just taxis without drivers.

If the AI is as good or better than a human driver your first point wont matter.

You can't avoid accidents anyone who is driving a car on a daily basis will be involved in an accident involving another human being in one point or another (and I don't even count rush hour fender benders) that's just statistics.

If driverless cars can improve those statistics they'll be used if only than to get lower insurance premiums in the end a company that can reduce the chance of you getting into an accident by even 10% would have every insurance company in the world behind them.

I for one do not see a major legal battle or precedents especially not from places like China, smart cruise control and various other drive assist technologies (lane keeping, auto park etc.) have been around for about a decade and we still haven't had major cases around them and there is no way that say Toyota's auto park feature did not cause some accidents.

Driverless or not as long as there is a driver in the car they are responsible for any liabilities as they have full control over the car even in drive assist mode (steering and breaking is always accessible to the driver) so unless there will be some major issue and that issue has to be that the AI actually prevented the driver from using the car controls I don't see any major legal issues arising.

And by the time driverless cars become truly driverless as in Total Recall (the Schwarzenegger one) driveless insurance companies would find a solution for that just like they found a solution for human drivers. As long as it's an accident and the AI and all car systems were in working order the insurance will cover it if the car was not maintained properly then the owner of the service will be liable. This is pretty much how human driver coverage works, if you did nothing wrong, the car was maintained, and you weren't negligible (e.g. drunk driving, speeding, etc.) you are covered by your insurance in any other case you are on your own.

> This would encourage people to only use truly driverless AI in the safest possible conditions, ie not on crowded streets or when its raining.

You just cancelled any driverless cars in NL or the UK.

Driverless cars are only useful if they have a 'backup' driving mode by a person in the most extreme of circumstances. Under all normal conditions the car should drive itself. Otherwise it is not a 'driverless car' but merely a car with a software augmentation for the driver and they should be lumped in with the regular cars.

Self driving cars should be able to move from one place to another without a licensed driver present or required in all circumstances normally encountered.

> Driverless cars are only useful if they have a 'backup' driving mode by a person in the most extreme of circumstances.

This sounds terrifying. There you are, in your self-driving car, doing some light text editing, when BAMN the alarms beep and you're handed back control of your car. Hey, try not to kill the pedestrians.

No, obviously not in the middle of driving.

I meant that as in after the car has been sideswiped by another vehicle or if the batteries have been drained to the point that the on-board electronics need work, in a garage and so on. In times when the cars self-driving capabilities can not be relied on.

Not in normal traffic and certainly not as a 'ok, I give up, you try something' mode.

Hence the 'Under all normal conditions the car should drive itself' bit above. That includes trying not to kill the pedestrians.

Is that really necessary? With a fleet car, a replacement comes soon enough, and the towing vehicle can pick up the broken one. With personal SDCs, it's probably the same, with the possible exception of the replacement car being the same premium brand (and it won't be able to arrive as soon as a generic fleet one, I guess).

Yes. After all a driverless car could be stuck in such a way that it is blocking traffic and a 'jog mode' would allow you to get it out of the way, just like you would with a normal car. 'Soon enough' may cause a 20 km traffic jam or even an accident.

FWIW Google already has a patent[1] on remote assistance that would cover that scenario

[1] https://www.washingtonpost.com/news/innovations/wp/2015/09/0...

I'm sure they do. Wonder how they plan on using that trick on a car that has it's radio and/or sensors disabled and where the 'remote operator' has no clue what is going on to bystanders.

> a car with a software augmentation for the driver and they should be lumped in with the regular cars

Which have already existed for some years now (see Mercedes Benz' DISTRONIC PLUS: https://www.youtube.com/watch?v=zXB82PfNucg).

>Chinese government... can do the math

I read 1 in 5 deaths in China are caused by air pollution. Not sure they are doing a great job there.

Actually, it could still be worth it. Lives are, by and large, prolonged and bettered with wealth. It could very well be that they did the math and 1 in 5 for 30 years due to rapid industrialization is better than the alternative.

It was OK that children were put to work, maimed and died in droves during the industrial revolution. We wouldn't be where we are today without their sacrifice!

Lots of what we're doing right now is luxury earned by our current level of development. It's nice to find an emotional example like maimed children, but if you tried to run 19 century industry with all of today's regulations and benefits, you's simply fail.

And from that we go to "was _this_ particular thing really necessary?", which doesn't really change the main argument. Sure, they could've done without maimed children. Add 8 hour workday, weekends, maternity leave, minimum wages, social security, pensions, health care and pretty much all the things we take so much for granted I can't even think of them. Like I said, it's about wealth - some things you can afford as a country when you're US and some you don't when you're Moldova or Albania, not to mention Congo.

Or maybe they decided the math works out?

You're essentially saying that autoritarianism is better because it can at times disregard individual rights in the name of efficiency.

He proposed this is one advantage that system has in this particular matter.

You didn't disprove that point, just reword it into the ominous general case (totalitarianism == good) which is a statement he never made.

Yes, he's saying that it is a good feature to be able.to disregard individual rights in the name of overall efficiency. To say that that is controversial is an understatement.

It's a classic moral dilemma (https://en.wikipedia.org/wiki/Trolley_problem). If you knew 100 people would die this year but you could do something that would save them all at the expense of killing 50 other people, what would you do? The coldly logical choice is to "disregard individual rights in the name of overall efficiency" and prefer 50 deaths to 100, particularly if you're a government - isn't it?

In the US, 32,719 died in 2013 from automobile crashes. Any wide-scale deployment of an automobile technology on public roads will mean that people die in crashes with car which have that technology.

That number alone proves nothing. You could try to say that "even if a self-driving car is a thousand times safer than the average human, that still leaves 32 deaths per year". But I don't buy that logic. You need a breakdown by cause. For instance: how many of those involved at least one person being drunk? Sleepy? Going above the speed limit? There are large classes of failure modes that self-driving cars are not prone to. Is it really too much to believe that some day we will have a civilization that can move people from place to place without killing anyone? I remain optimistic. But even if not, we can at least get this thing off the list of top k causes of death, for some integer k. Already it's responsible for less than two percent of US deaths. (source: wolfram alpha) A bit more work and we can move it onto lists like this instead: https://xkcd.com/369/

> some day

Yes, but not tomorrow. Tomorrow, we can deploy a technology that will decrease deaths but not to zero.

Fair enough. I'll settle for decreasing to non-zero for tomorrow :)

Why do you think that?

The general belief that public outcries are not allowed in China.

Sure that is the general belief.

Free speech in China is limited compared to the west. Even though you can see public outcry comment on the factory explosion in Tianjin and the recent smog in Beijing.

But what is mostly missing from the picture are the collosal improvements in public safety, pollution, health that has happened in the country over the past decades. These improvements are far larger than anything we have done in the west. And are a result of directed government policy.

Why would the Chinese government allow self-driving cars that kills people just because it can censor Twitter?

My point was that they would be saving lives by allowing driverless cars, since humans are more dangerous behind the wheel. And, because they can censor twitter (as an abbreviation for controlling propaganda) they can get away with it in a way the US could not.

What if careful drivers are less likely to be injured in a crash if self driving cars are not introduced, despite there being less overall road injuries? Would not these drivers' outcry be justified then?

Less injuries > more injuries. Less deaths > more deaths.

Generally i think everyone would agree with your statement. However, the trolley problem is a though-experiment which shows that the argument is more nuanced (at-least psychologically).

https://en.wikipedia.org/wiki/Trolley_problem

A TED talk on this as well: https://www.ted.com/talks/michael_sandel_what_s_the_right_th...

Yes, like pretty much everything, this rule gets hairy when you zoom in on it enough. That's why, e.g., medicine has the concept of Quality-adjusted life years (QALY), to capture the intuition that it's better to live 10 years in perfect happiness than 15 years in agonizing pain, etc. However, I think the principle is generally sound, and if you oppose the option that would lead to less deaths, you have to show what society loses that's more important than the lives saved (and yes, it happens; for instance, car traffic in general is society trading some loss of lives for huge boost in convenience and economic prosperity).

What society loses is an inefficient but largely accepted legal framework for compensation in the event of injury or death. This framework right now is balanced by individuals accepting culpability in the event of an accident. An individual has certain limitations to obtain and use legal leverage to evade financial responsibility. A massive lucrative corporation has far more leverage and interest in avoiding financial responsibility and you have shifted the obvious blame away from the individual. You've disrupted a finely honed set of checks and balances that enable global highway transit.

What's needed is a big pot of money for the greater good, and that means government. Government means regulation. Regulation stifles innovation and adds overhead that scares away competition.

And now you've got yourself a catch-22 that requires a company with a very specific set of parameters to achieve any progress. Which is what we've got.

I would love to see it actually try to drive in a city like Beijing or Shanghai. I've been to both cities several times for work, and I can say that if it can drive in conditions like that, then it can drive anywhere.

You haven't been to India I suppose. But more seriously I agree. Having self-driving cars negotiating road traffic in places where the traffic rules are 'blurry' and rarely (if ever) followed to the dot; is going to be a big challenge.

Road conditions and general behaviour in traffic is much better in China than in South-East Asia or India.

I would not be surprised if China ends up being the first place with self-driving cars at a large scale. Because the road instructure is in general good and new - and with government support (which a company like Baidu has) legislation and infrastructure will be changed to accommodate self-driving vehicles much faster than we would be willing to do it in the west.

According to other sources, they did drive it in Beijing - and in bad conditions [1]

"We’ve tested the car at night and day, as well as in medium rain, and – since this is Beijing – in smog of course."

[1] https://www.techinasia.com/baidu-autonomous-car-hits-roads-c...

Beijing covers a very large area, including isolated (if not deserted) country roads and massive amounts of highway. It's much more likely that they drove in a rural area (or in a deserted corner of Zhongguancun) than on any road inside the fourth ring road.

Saigon where I am would be an interesting test case too:

http://www.saigon-gpdaily.com.vn/dataimages/original/2011/12...

It takes a bit of getting used to as a human driver too but in some ways isn't so bad. The rules work more like walking through a crowd of pedestrians going in random directions than regular driving.

Baidu also claims innovation is their source of power (or some crap like this).

Even though it's most famous for showing fake hospitals' ads for 1000 Yuan per click.

I wonder what is in common between a search engine and a driverless car. Apart from creating a Google copycat. I'm persuaded they did reinvent all the internals of the car, so when you're at that point of innovation, why copy Google's list of departments?

Top software engineers, especially in the field of machine learning and AI in general.

Yeah, I think it's more accurate to say that both Baidu and Google got their lists of departments from the same source: academic machine learning.

Andrew Ng works for Baidu.

Their face recognition software is currently much better than second best (Google) 0.37% vs 0.15% error, Their voice recognition seems to be the very best. What they are famous for now is not what they are famous for in the next few years.

Do you have a citation for that? Those error rates seem extremely low for facial recognition. They look closer to MNIST than anything.

I'd much rather be driving next to a car that can see 360 degrees and respond in milliseconds than next to a human. The sooner this happens the better.

In this scenario, you as a human driver is still a fundamental flaw.

Herd immunity. If 90% of cars are super predictable autonomous vehicles, the 10% remaining will be very safe.

Unless they're teenagers. God help us.

Same for me, with a few small caveats. And, I would add: next to a human that might have had a few too many drinks just before getting into the car.

Same Baidu that just cheated on machine learning contest? K I'll pass..

That's the kind of claim that is difficult to believe unless we have a footage of the driving from inside the car or something like that.

There must be some subtle differences in the definition of China's first fully autonomous self-driving car:

https://translate.google.com/translate?hl=en&sl=zh-CN&u=http...

A fully autonomous self driving car is pretty easy. Making one that doesn't crash is the hard part.

You can wedge a brick on the accelerator pedal and call that autonomous. Its autonomous for the whole 10 seconds before it hits a wall.

BIDU launches one or two vaporwares every year and this is one of them.

okay, since someone is probably pissed at this comment, to clarify, this sentence is a quote from @Fenng here http://weibo.com/1577826897/Czz1zj82y

Let's see how many vaporware were released by BDIU in 2014:

http://tech.sina.com.cn/i/2015-01-08/16219941986.shtml

Andrew Ng works for Baidu now, so I'm not surprised.

Keep in mind this is Baidu.

This baidu: http://dswalter.github.io/blog/machine-learnings-first-cheat...

And yet still their search engine does not support indexing HTTPS content.

hahaha

I find it comical that they used a BMW to do this rather than a Chinese car.

And Google using Toyota instead of an American car... It's almost as if mega corporations didn't care about this kind of stuff.

Toyota's sold in America are largely American. Currently the Toyota Camry is the 'Most American Car'.

Though, the prius is made in Japan currently with plans to move production to the US.

Reply on news.ycombinator.com