211–220 of 495 posts
can we just make a law that the driver is responsible/liable for the car they drive, regardless how they drive it? I mean breaks failing while you are actively driving is one thing, but to completely surrender control, that is a choice.
Earlier quoted context omitted.
> it's a common luxury car feature Almost all luxury cars implement this feature the same way that Tesla does, with radar that ignores stationary objects. I'm not aware of another company that implements the steer towards obstacles part of the problem, though.
This is simply not correct. In fact, the link in the comment you just responded to contradicts you: https://www.caranddriver.com/features/a24511826/safety-featu... It's actually pretty frustrating to have written a little summary of what likely happened WITH SOURCES, that a lead car changed lanes and confused the AES system, only to have people ignore it. The only thing worse then a pedant is an incorrect pedant.
Super interesting to see that Subaru is doing emergency breaking with cameras, though. Seems I'm a couple of years out of the loop.
The problem with Tesla's "Autopilot" is that it's marketed heavily as just that: autopilot. It's good enough to work that way 99% of the time, too, which is a recipe for disaster: things that work 99% of the time but have the potential to fail catastrophically often get glossed over by humans because it's hard to keep paying attention to something that rarely fails.
That's all very true, but one interesting thing is that the driver in this case knew that at this location the autopilot was 100% likely to fail. I believe I read he had even reported it several times to Tesla Service and his family. The lane markers in this area, based on video I watched of it years ago, are pretty bad, IIRC. Like, a human could have problems following the markers bad. As evidence to support this I'…
Tesla made some very big claims in their recent autonomy day event- basically, they claimed that they are years ahead of competitors, while operating on "hard" mode (no lidar). And yet, a number of participants in the short autonomy day demo rides claimed that the support driver had to disengage the autopilot. Has Tesla provided any evidence which shows evidence that they are in fact far beyond competitors?
Evidence showing they are ahead in terms of customers? For now, maybe. In terms of safety? No.
According to some fans, they sell a product you control. You're allowed to use it anywhere, therefore, they're ahead.
According to Tesla, they designed a chip that's best-in-class and they share a small amount of safety data quarterly [1]. Musk will not share any more data [2] because that would allow people to "turn a positive into a negative".
Earlier quoted context omitted.
That's all very true, but one interesting thing is that the driver in this case knew that at this location the autopilot was 100% likely to fail. I believe I read he had even reported it several times to Tesla Service and his family. The lane markers in this area, based on video I watched of it years ago, are pretty bad, IIRC. Like, a human could have problems following the markers bad. As evidence to support this I'…
Except he won’t argue that because he is dead. Which, in itself, is the best counter argument to your conclusion that he knew what he was doing and allowed it to happen intentionally. Unless you’re also theorizing that he chose to commit suicide using one of the most creative methods he could, while making Tesla take the blame. It’s not impossible I suppose, but it doesn’t seem like the most reasonable conclusion.
I know the argument is always that autopilot is not intended by Tesla to be abused (never mind the CEO on national television abusing it). However, there's what you tell humans, and then there's pragmatism regarding human nature. We should consider the latter. It's possible to be both academically correct on the warnings/instructions given and also practically wrong on human psychology.
Needless to say, people who think like this make terrible dictators when given the chance.
The problem with Tesla's "Autopilot" is that it's marketed heavily as just that: autopilot. It's good enough to work that way 99% of the time, too, which is a recipe for disaster: things that work 99% of the time but have the potential to fail catastrophically often get glossed over by humans because it's hard to keep paying attention to something that rarely fails.
This is exactly why Waymo and Cruise and such believe autonomy Level 3 or so is utterly dangerous and have intentionally skipped it entirely.
Earlier quoted context omitted.
On the flip side, Tesla fixed this specific issue (and apparently unfixed it again) with over the air updates. If this was another manufacturer the cars would need to be recalled and it would likely be years before a fix was rolled out to all vehicles and therefore lives would continue to be in danger. So it isn't like over the air updates are inherently bad. The reality of the situation is that there almost certainl…
I won't argue against over the air for everything but when your 'autopilot' system can be updated any day for any reason, I would argue that it failing and killing you shouldn't be as excusable.
Earlier quoted context omitted.
Aren't the algorithms for extracting depth from stereo imagery already solved and thus also essentially "free"?
No, it's like this: Radar, LIDAR, ultrasound: returning packets have X value, where X is run through a single algorithm to return depth. In the case of ultrasound and radar, the strength of the return packet is generally directly proportional to depth (algorithmically). LIDAR systems might require an additional processing step depending on the type of LIDAR system used, since many systems use shifted-arrays or rotati…
Err, no it isn't. You send a tx pulse and correlate it with a rx pulse and use the time of flight and the wave speed to get a range. You need precise information about where the tx pulse was sent to turn the range into 3d point relative to the sensor. The correlation and orientation steps are algorithmic and fundamenally statistical. For ultrasound and radar you can put a code in the pulse to make matching them up easier (the pulse lengths are typically bigger than the resolution you require). All algorithms. All can and do go wrong.
Old school stereo matching looks at cross correlation (or cross entropy) of image patches. If the cameras are orthorectified (translated with no rotation) then you just scan across the row of pixels to find the patch with the best match and how far across it is is proportional to the depth. Feature correspondence (with sift like features, very low level) might be used for getting relative rotation and translation info between the cameras for calibration and as a depth prior. People are doing the same sort of thing with NNs to get the depth map.
They should focus on making quality, well-performing electric cars. Stop using us all as beta testers for an autonomous future most of us never asked for.