George Hotz cancels his Tesla Autopilot-like ‘comma one’

Post by jhayward » Fri, Oct 28, 2016, 7:20 PM UTC

wyager wrote: post #188

smallnamespace wrote: post #174

Earlier quoted context omitted.

The timeline is not unreasonable at all. They were planning to ship by the end of the year -- obviously the regulator should get involved before it's in the hands of users. I've seen requests from regulators before. If you don't have or can't produce exactly what they asked, you call them up and tell them some parts will take longer. The point is to have a dialogue; the threat of sanctions is just to make sure you wo…

Two weeks is not reasonable for the demands made in the letter. That, plus a $21,000/day penalty, is a clear "fuck you".

Post by agumonkey » Fri, Oct 28, 2016, 7:24 PM UTC

sedachv wrote: post #296

argonaut wrote: post #245

Earlier quoted context omitted.

It is not "similar tech" at all. Yes, both approaches involve some machine learning (which is a huge field). But Cruise's approach is similar to Google/Uber's approach in that it relies on LIDAR. I'd assume (but might be wrong) in that it also relies heavily on premapping, as does the Google/Uber approach. Comma's approach was complete end-to-end learning with just a single camera (to give you some context, research…

There are more details in this lecture Hotz gave at Berkeley this September: https://www.youtube.com/watch?v=Hxoke1lDJ9w One of the things that seemed missing from the lecture when I watched it is how data was going to be collected for handling exceptional situations/collision avoidance. Hotz also glossed over the question in this interview from July: https://www.youtube.com/watch?v=2zy_07g2IrM#t=35m I suspect they r…

Post by lutorm » Fri, Oct 28, 2016, 7:25 PM UTC

TeMPOraL wrote: post #187

erikpukinskis wrote: post #180

Earlier quoted context omitted.

Proof that so-called autopilot systems are a danger to the public would be evidence that they perform worse than the average driver. I haven't seen such evidence, what are you referencing?

It's the other way around - you need a proof that they perform comparable to an average driver during expected use, and they do not have absurdly stupid corner cases during actually expected use.

Post by jacquesm » Fri, Oct 28, 2016, 7:26 PM UTC

tptacek wrote: post #263

patio11 wrote: post #257

Earlier quoted context omitted.

This letter is designed to kill a company operating with complete disregard for proper operations in a regulated industry. We are in entire agreement, but people cannot believe in the superposition of this and "Oh that's just some simple questions from the regulator." Edit to add: Quoted bit appears to have changed after my quote. I think the old version remains accurate.

They're asking for documentation that any serious vendor would already have. That we believe this would kill comma.ai is really more a testament to our priors that George doesn't keep anything resembling rigorous test documentation.

Post by jacquesm » Fri, Oct 28, 2016, 7:27 PM UTC

snotrockets wrote: post #292

jacquesm wrote: post #4

Earlier quoted context omitted.

That may be just a fig-leaf to plaster over 'we couldn't do it'. It's one thing to release software, it's quite another thing to release this software and have it tested, audited and certified safe and that's definitely part of the process of delivering software that is critical to the well being of the users and the other people you're sharing the road with. And that second part is part-and-parcel of wanting to oper…

An ex-colleague of mine used to say the difference between a software engineer and a real engineer is that the latter's signature is criminally binding. We can refuse to be real engineers (it is a great responsibility), but then we should drop the title.