Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance • Hacker News Board

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#261

Post by dogma1138 » Wed, Sep 21, 2016, 6:35 AM UTC

Spooky23 wrote: post #241

harryh wrote: post #166

Earlier quoted context omitted.

In practical terms I don't agree that the threat of someone doing all of the following things is worth worrying about (in comparison to many other more likely failures): 1) determine what physical hardware in aws the target is running code on 2) somehow get the aws virtual machine manager to let the attacker run their malicious code on the same hardware 3) somehow pierce the protections of the virtual machine to read…

People said the same thing about cold boot attavks against encryption keys. Yet today the police and others are using that and other NAND attacks regularly. HIPPA is a very easy compliance standard to meet. If it seems difficult to meet those requirements with your standard tool configurations, you should think about what that means with respect to the integrity of your data.

I would like to see a case of a cold boot attack by the police.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#262

Post by detaro » Wed, Sep 21, 2016, 6:41 AM UTC

LeChuck wrote: post #260

seanmcdirmid wrote: post #244

Earlier quoted context omitted.

> You could expect them to not call it autopilot, and instead call it lane keeping, like it really is. That is all that most auto pilots in planes do, however. I don't get where "autopilot" somehow came to mean full autonomy in cars but not in planes (and other vehicles like boats) where the term was used previously. Ridiculous.

Yes, but with planes and ships (on ships it's almost always called track pilot and speed pilot, by the way) you can let go of the controls for extended periods of time. In cars, right now, that's a recipe for disaster. Functionally they may be the same, but practically speaking it's very different.

This is the key point. People don't care about what it does in detail, they think about the user experience.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#263

Post by s_kilk » Wed, Sep 21, 2016, 7:31 AM UTC

Animats wrote: post #249

ekianjo wrote: post #216

Earlier quoted context omitted.

Why was this not widely reported?

The dash cam video was only released last week, in conjunction with a lawsuit. Now it's on all the mainstream news outlets, from the Wall Street Journal to the New York Times to Fox News. This is yet another "Tesla hit slow/stopped vehicle on left of expressway" accident. There are now three of those known, two with video, one fatal. Watch the video. The vehicle is tracking the lane very accurately. Either the driver…

Jeez, that's pretty bad. This seems like the most basic case that autopilot is supposed to solve, and yet it still crashes?

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#264

Post by seren » Wed, Sep 21, 2016, 7:51 AM UTC

yoav_hollander wrote: post #87

seren wrote: post #33

Earlier quoted context omitted.

I have a hard time understanding the current AV SW stack. On one hand, at the low level, sensor, motor control, etc you likely have traditional hard real time/MISRA C code, but on the higher layers you probably things like DNN, image recognition, which are much less deterministic. So I am not sure how do you reconcile these two worlds, and prove it is safe and always work in timely manner. It seems the only sound app…

A few comments on this: First, as etendue says, it is not easy. The problem of mixing “Boolean” verification with probabilistic, less-deterministic verification is especially hard. I discussed this a bit in [1], if you care to take a look. Also, I think most current AVs are not driven by DNNs at the top level (comma.ai [2] is one exception). See [3] for some discussion of that, and of verifying machine-learning-based…

Thanks for your input, really interesting topics on your blog as well.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#265

Post by mjevans » Wed, Sep 21, 2016, 8:29 AM UTC

Animats wrote: post #131

From the regulations: "Fall back strategies should take into account that—despite laws and regulations to the contrary—human drivers may be inattentive, under the influence of alcohol or other substances, drowsy, or physically impaired in some other manner." NHTSA, which, after all, studies crashes, is being very realistic. Here's the "we're looking at you, Tesla" moment: "Guidance for Lower Levels of Automated Vehic…

They would be better off making SAE 1-2 features 'game' like.

You get positive points for avoiding situations that are noticed (but not within the audible warning threshold) or correctly reacting to input (warned but not yet in automated 'fail safe').

Now if only they could automatically make cars exiting a rolling slowdown on the freeway actually get back up to the indicated speed of travel in an expedient manor.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#266

Post by lisivka » Wed, Sep 21, 2016, 8:37 AM UTC

s_kilk wrote: post #263

Animats wrote: post #249

Earlier quoted context omitted.

The dash cam video was only released last week, in conjunction with a lawsuit. Now it's on all the mainstream news outlets, from the Wall Street Journal to the New York Times to Fox News. This is yet another "Tesla hit slow/stopped vehicle on left of expressway" accident. There are now three of those known, two with video, one fatal. Watch the video. The vehicle is tracking the lane very accurately. Either the driver…

Jeez, that's pretty bad. This seems like the most basic case that autopilot is supposed to solve, and yet it still crashes?

IMHO, autonomous car should use at least two autopilot systems from independent vendors, to avoid single point of failure.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#267

Post by lisivka » Wed, Sep 21, 2016, 8:39 AM UTC

jaxbot wrote: post #219

ekianjo wrote: post #216

Earlier quoted context omitted.

Why was this not widely reported?

Last I heard, nobody has been able to verify if the car was actually in autopilot mode. However, the emergency braking also clearly failed, if the police report that no attempt to stop was made is true.

It's classical Single Point of Failure: single fault of just one subsystem leads to crash.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#268

Post by pmyjavec » Wed, Sep 21, 2016, 9:45 AM UTC

joemi wrote: post #191

pmyjavec wrote: post #167

What is it about self-driving cars that has HN readers so incredibly excited? Sorry if it's a little off-topic. The reason I ask is there are plenty of other countries in the world where cars just aren't that important, let's take Netherlands for example. If you have automatic cars, society here is not just going to be that excited AFAIK. Public transport here is great and most people cycle everywhere, because it's f…

Good questions and it's great to hear this take on things. Cars play a large role in America. I don't know all the history about how it came to be this way, but I can make some guesses... * America is very big, and a lot of the settlements are spread out by a ways. Cars make those communities less insular because they provide a way to get from one town to another, where biking would take a very long time and signific…

What happened was; an auto-industry was born and lobbied the government to tear down a lot of the tram, bike and rail infrastructure you guys had. Especially in LA.

A similar thing happened in Sydney, Australia, the city had one of the largest tram systems in the world, the director of the motor trade authority was elected into government and made sure that all the tracks were literally tared over, they are still under the roads. The excuse used was that the network was over congested / too popular, and cars would solve the problem, guess what? Sydney is in the middle of putting the tram network back in to solve the car problem :) Melbourne, Australia was spared because it built it's network in the 40's and 50's and it was difficult to persuade the working class that putting in such a newly built system was a good idea. People literally move to Melbourne because of the convenience they provide.

If you're interested Bikes vs Cars (http://www.bikes-vs-cars.com/) is an awesome documentary, it shows that LA even had things called "Bike Highways" at one stage. You will see how people were so negatively influenced by big business lobbying, and from memory outlines why New York was somewhat spared from the fate.

Sounds like you're being told the same stories again either way.

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#269

Post by pmyjavec » Wed, Sep 21, 2016, 10:06 AM UTC

tstrimple wrote: post #228

pmyjavec wrote: post #194

Earlier quoted context omitted.

Thanks for the response and totally agree with the sitting in traffic, wasted time thing ,but with all due respect it still seems inefficient, wouldn't public transport and telecommuting be better options in a connected world? How will there still be no traffic jams, or the car will be like an office? In that case why not just work from home and come to work for meetings here and there? Might flexible / less work hou…

The more human drivers that are replaced by software, the less congestion there will be on the roads. It's delays in human reaction times which cause cars to back up on roads and take forever to get through intersections. https://www.youtube.com/watch?v=iHzzSao6ypE

I still don't see how the cars are going to entirely solve the problem of everyone needing to be in the same place at the same time, they still have to cope with each others presence, unless they can fly.

In my opinion a good public transport system does. I've spent time in Tokyo, the rail network is second to none, it's congested sure, but it works.

The problem is that people still need to get in and out of the cars for example, which means there is still wait time. Breakdowns will happen. All humans drivers will have to be banned etc.

It sounds to me like the auto-industry is still pushing a selfish agenda and the valley is buying it, because it's an opportunity to take more money.

For me, I'm happy because I'll still ride my bike and it will be a much safer place for me to exist because self-driving cars won't try and kill me, hopefully :)

Re: Self-Driving Cars Must Meet 15 Benchmarks in U.S. Guidance

#270

Post by ocdtrekkie » Wed, Sep 21, 2016, 12:17 PM UTC

studentrob wrote: post #258

aaronblohowiak wrote: post #224

Earlier quoted context omitted.

Can you back that up?

Google is definitely level 3. Their presentation at SXSW this year shows it, https://youtu.be/Uj-rK8V-rik?t=17m54s

Nobody's saying their cool little 3D renders aren't awesome looking, but that doesn't really mean much. Google has PR down to an art form, but if you asked Google to drive one of their cars to Chicago, they couldn't do it, no matter what the weather. They only work in a small nearly closed course environment.