Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

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

bloomberg.com

Post reply on HN
Print view View original

41–50 of 300 posts

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

#41

Post by euroclydon »

I'm surprised that self-driving technology is focusing on replacing the driver as an autonomous actor, processing visual and radar/lidar signals in order to know about its surroundings. I've always thought we'd get further faster by having automobiles also talk to other vehicles nearby, and design roads to support the computer driven vehicles.

Two examples are:

1) If the vehicle is talking to the cars in front of it, it can know they are braking before it senses that visually. Also, the vehicles can speed up in a gridlock scenario more in unison, like a train.

2) On the interstate, markers in the pavement can be specifically designed for computer sensors rather than human eyeballs. Also, cars can draft together to save fuel.

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

#42

Post by DougWebb »

rubidium wrote: post #40
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…

Now you understand the job of systems engineering :) Verify components, validate the entire system is the typical approach.

That sounds pretty much just like web application development, or any other front-end user-facing development. You can verify internal components through testing, but once you introduce non-deterministic random variables like browser software your user is using, and your user's themselves, all you can do is validate the entire system through real-world testing and hope you've covered the edge cases you need to handle and will fail gracefully for the ones you missed.

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

#43

Post by seren »

rubidium wrote: post #40
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…

Now you understand the job of systems engineering :) Verify components, validate the entire system is the typical approach.

The point I was trying to make is that if you have actuators running MISRA C that are going to be driven by something written in Tensorflow, does it still makes sense to have a requirement to use MISRA C in the first place for the low level part ?

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

#44

Post by etendue »

seren wrote: post #33
etendue wrote: post #16

Earlier quoted context omitted.

The report makes reference to "Assessment of Safety Standards for Automotive Electronic Control Systems" by NHTSA, which itself reviews ISO 26262, MIL-STD-882E, DO-178C, the FMVSS, AUTOSAR, and MISRA C. In this context, they mean verification and validation in the systems engineering sense. Software would be included in that it is a part of the whole system.

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…

I think the simple answer is that it is not easy. To start, rigorous design processes with risk analysis upfront are certainly necessary, as are well-defined operational contexts for the autonomous functionality, and a very disciplined approach to clearly defining safety-critical subsystems and minimizing their surface area.

There's a surprising amount of work in the literature that serves as a guide for using neural networks in safety-critical contexts, e.g., http://dl.acm.org/citation.cfm?id=2156661 and http://dl.acm.org/citation.cfm?id=582141.

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

#45

Post by nommm-nommm »

travoc wrote: post #27
revscat wrote: post #20

Earlier quoted context omitted.

And is the fundamental belief of libertarians: that governments can do nothing right, either morally or pragmatically.

That's not accurate. They seem more concerned that every bit of power given to government to do something right will eventually be used to do something wrong.

That's exactly what the libertarians on HN post. I seriously hope it's a very vocal minority though.

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

#46

Post by Conlectus »

euroclydon wrote: post #41

I'm surprised that self-driving technology is focusing on replacing the driver as an autonomous actor, processing visual and radar/lidar signals in order to know about its surroundings. I've always thought we'd get further faster by having automobiles also talk to other vehicles nearby, and design roads to support the computer driven vehicles. Two examples are: 1) If the vehicle is talking to the cars in front of it,…

I think that concern over malicious communicators will at least slow this down. If not implemented correctly, it could give hackers a dangerous amount of control over traffic.

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

#47

Post by Angostura »

ams6110 wrote: post #28
e40 wrote: post #18

Earlier quoted context omitted.

Many people automatically assume a government can't make up sensible regulations. There are a lot of them in the US. It's a meme you hear all the time, especially in a POTUS election year.

My biggest gripe is overreach. You start with sensible building codes, and eventually the city council is telling you what color bricks you have to use before they'll approve your plan. Yes this happens.

In, say a relatively historic area where all the buildings are the same, what is wrong with mandating brick colour?

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

#48

Post by leereeves »

luma wrote: post #39
e40 wrote: post #18

Earlier quoted context omitted.

Many people automatically assume a government can't make up sensible regulations. There are a lot of them in the US. It's a meme you hear all the time, especially in a POTUS election year.

I think it might be deeper than that. I don't feel that the US government, on it's own, is incapable of drafting up reasonable legislation. The problem is that the US government is 100% for sale to the highest bidder, and corruption runs deep (we just call it "campaign contributions" as if that makes it better). If sensible regulation is proposed, it'll last 30 seconds before the good senator from [some self-driving…

And on top of that you have obstructionism (from both sides) and deliberate efforts to sabotage the other side.

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

#49

Post by leereeves »

euroclydon wrote: post #41

I'm surprised that self-driving technology is focusing on replacing the driver as an autonomous actor, processing visual and radar/lidar signals in order to know about its surroundings. I've always thought we'd get further faster by having automobiles also talk to other vehicles nearby, and design roads to support the computer driven vehicles. Two examples are: 1) If the vehicle is talking to the cars in front of it,…

The first self-driving cars will have to coexist with humans driving old cars without such communications.

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

#50

Post by etendue »

seren wrote: post #43
rubidium wrote: post #40

Earlier quoted context omitted.

Now you understand the job of systems engineering :) Verify components, validate the entire system is the typical approach.

The point I was trying to make is that if you have actuators running MISRA C that are going to be driven by something written in Tensorflow, does it still makes sense to have a requirement to use MISRA C in the first place for the low level part ?

I'd be very wary of using complex SOUP like TensorFlow, even if brought under my quality system. I think a good answer here is that once one goes under design control the subset of functionality needed should be implemented in-house under the organization's SDLC.
Post reply on HN