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

Text Messages Between Travis Kalanick and Anthony Levandowski

spectrum.ieee.org

Post reply on HN
Print view View original

71–80 of 124 posts

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#71

Post by totalZero »

furioussloth wrote: post #41
paul6987 wrote: post #35

Earlier quoted context omitted.

Don't forget illegally running autonomous tractor trailers in Palo Alto and on the roads of Nevada. These guys are criminals... total disregard for human life (but their own) in every shape and way there is. From potentially killing people, too treating everyone is who not their BRO like trash; drivers, employees, customers, business partners, etc, etc, etc... GROSS!

Your comments come off as if you are more bitter about some company you can just choose to ignore than benchamark is about Travis.

If you ever walk through a crosswalk, ride a bicycle (or motorcycle), or drive a car, you may pay with your life for ignoring self-driving cars that violate traffic laws and/or fail to react intelligently to unusual situations.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#72

Post by stouset »

ariwilson wrote: post #61
tim333 wrote: post #47

Earlier quoted context omitted.

Incidentally, Musk's take: "The whole road system is meant to be navigated with passive optical or cameras and so once you solve camera vision then autonomy is solved if you don't solve vision it's not solved so that that's why our focus is so heavily on having a vision neural net that's very effective for road conditions." https://www.youtube.com/watch?v=gv7qL1mcxcw&feature=youtu.be...

Why would you want to limit yourself to passive cameras and make your life harder? This is like limiting yourself to flapping bird wings to make airplanes.

His thesis is that relying on cameras makes it easier, since the entire preexisting road network is literally designed around optical navigability.

Adding other sensors isn't free. Every minute you spend on developing techniques to process inputs from other sensors, not to mention integrating their conclusions with that of other sensors, is time, money, and energy you could have used to improve your optical system.

I'm not saying I necessarily agree (though I find his position intuitively compelling), but he clearly thinks that it's easier, faster, and cheaper to bring an optical-only system to a point of reliability than it is to bring a mixed-sensor system to the same point.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#73

Post by rasz »

gruturo wrote: post #31
b4ux1t3 wrote: post #25

Earlier quoted context omitted.

I know that you were just doing a tl;dr and not necessarily trying to start a discussion about it, but wow, that seems kinda stupid to be coming out of someone who is ostensibly fairly intelligent. A 3D point cloud is great, but with the raw computing power we have these days, image processing should be pretty reliable, at least until LIDAR becomes more. . .well, reliable and affordable. Doesn't the HoloLens just use…

> A 3D point cloud is great, but with the raw computing power we have these days, image processing should be pretty reliable In theory? Yes. In practice? That's how the first autopilot fatality happened. The image processing confused the white side of a semi truck with the sky.

True, but it was Mobileye _one_ quite low res BW camera system that decapitated the driver.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#74

Post by babesh »

josh2600 wrote: post #45
Afforess wrote: post #42

Earlier quoted context omitted.

Normal humans also run red lights when the yellow-to-red light time is too short. Transit guidelines generally state that yellow lights should stay lit 1 second for every 10mph of the speed limit. When the yellow light is shorter than this, it's easy to get caught in the middle of a red.

Have you seen the video? It's way more jacked up than what you're implying: https://m.youtube.com/watch?v=pzzQ42D9Srw Watch that and tell me that it's reasonable or caused by a short yellow...

I've run that light myself by accident. It's a hard light for humans too. This light is in the middle of a block rather than at the end. The area is full of activity including people walking but also popular sights. That brick building to the right is SFMOMA. Its hard to notice the traffic lights amongst all this activity not to mention that you don't look for it. Also that middle lane can easily have its views obstructed on both sides.

That area is also messed up because you need to be in the correct lane else you will be forced up different streets so you have many cars trying to change lanes. Lastly, that area has its share of asshole drivers that cut you off, speed, etc....

From my perspective, that car is behaving like someone not aware of the light and not sensitive to context (should have driven more slowly). You can call that asshole behavior if you wish.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#75

Post by aianus »

ariwilson wrote: post #61
tim333 wrote: post #47

Earlier quoted context omitted.

Incidentally, Musk's take: "The whole road system is meant to be navigated with passive optical or cameras and so once you solve camera vision then autonomy is solved if you don't solve vision it's not solved so that that's why our focus is so heavily on having a vision neural net that's very effective for road conditions." https://www.youtube.com/watch?v=gv7qL1mcxcw&feature=youtu.be...

Why would you want to limit yourself to passive cameras and make your life harder? This is like limiting yourself to flapping bird wings to make airplanes.

Because not everyone can afford to spend $20k on extra sensors that make the car 1% safer. And holding back autonomous cars until they're perfect can kill more people than near-perfect autonomous cars. It's an economic tradeoff like any other.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#76

Post by rasz »

jfim wrote: post #70
sanguy wrote: post #24

Earlier quoted context omitted.

Multiple images can be used to compute a 3D point cloud. This is computer vision stuff around for many years. The challenge is this is a passive sensor in that the cameras count on light illuminating the scene. So at night; in bright light (that causes images to blow out); shadows; etc; you can have voids. If a person is in that void bad things can happen. But cameras now cost under a $1 each in volume (thanks smartp…

> Multiple images can be used to compute a 3D point cloud. [...] Add a GPU enabled system to process (it's quite compute heavy) and you are set. It requires enough information ("features") in the images to compute a stereo pair. Flat patches of color have no features and as such, cannot be correlated between cameras. In such a case, you have a spot with no depth information. This is exactly why you want to have other…

Everything has texture, even things that appear solid, this is how optical mice can work on glass. Throw in NIR/UV camera, or maybe thermal for good measure, slap LSTM on top and you are covered for everything human would spot.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#77

Post by aeontech »

boomboomsubban wrote: post #58
aeontech wrote: post #51

Earlier quoted context omitted.

No, if their car is not running any lights under normal circumstances, but has ran it even once at a particular intersection, you bet I will send my engineers there with some cameras to tape the intersection so we can analyze it and see what is happening. If the intersection is poorly designed/timed, why would it be implausible to happen to catch human drivers make the same mistake?

>If the intersection is poorly designed/timed, why would it be implausible to happen to catch human drivers make the same mistake? Unless they had already pulled the self driving cars, which the timeline presented suggests they haven't, why would that data be important enough to text back without the corresponding number of Uber violations?

I'm saying there's not enough context in the released messages to declare this unambiguously bullshit, and provided a reason why they would record the intersection, as parent comment asked "why in the world would they be taping it".

The number of Uber's violations could have been communicated previously, or could have been implicit in the context of the conversation (at least 1 violation, causing the investigation) - either of these 3 explanations seems equally likely.

Considering how dangerous this is, I really doubt they would run a red light SIX times through the SAME intersection purposefully, risking the life of REAL PEOPLE around them each and every time.

Uber's management and engineers may be irresponsible, but I don't think they're that evil, and I would hope no self-respecting engineer or human being would go along with that. "Oh, that's funny, let's try that again" works for software development, it's emphatically not how real world testing involving danger to life and limb works.

As to "why would this be important enough to text back", that's pure speculation, but if I was tasked with investigating this, I would be pretty relieved to see that this is a location that's confusing human drivers as well, not just my software, and would be pretty likely to communicate that back to my team.

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#78

Post by aeontech »

revelation wrote: post #63
aeontech wrote: post #51

Earlier quoted context omitted.

No, if their car is not running any lights under normal circumstances, but has ran it even once at a particular intersection, you bet I will send my engineers there with some cameras to tape the intersection so we can analyze it and see what is happening. If the intersection is poorly designed/timed, why would it be implausible to happen to catch human drivers make the same mistake?

Because it obviously isn't either of that. Did you watch the video?

I watched the video. You asked "why would they be taping the intersection", I gave you a pretty obvious explanation, not sure why you are so vehement about it.

In turn, why would any sane person run a two ton piece of gear through a moving intersection, endangering pedestrians and cross traffic through a red light where it is known to have failed once already, not once, not twice more, but six times in a day?

Why would you, as a responsible person, not immediately tell your test drivers to avoid this intersection until the issue is root caused, or, more likely, suspend testing for that day entirely?

Or do you just assume every engineer at Uber is a cackling mad scientist blinded by their quest to create our new machine god?

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#79

Post by studentrob »

ramdomize wrote: post #68
studentrob wrote: post #67

Earlier quoted context omitted.

> I thought that that was (also?) a problem with the radar, where it bounced under the truck and thought the road was clear. Ultimately this was driver error, since he wasn't paying attention and didn't break. As for why autopilot did not engage the breaks, you could blame any of the forward facing sensors since they all failed to see the truck. If the vehicle had windshield-height radar, or a better vision system, p…

Part of the fix was to use radar: https://www.tesla.com/blog/upgrading-autopilot-seeing-world-...

Oh interesting. That's a pretty detailed report.

One thing jumps out at me,

> This is where fleet learning comes in handy. Initially, the vehicle fleet will take no action except to note the position of road signs, bridges and other stationary objects, mapping the world according to radar. The car computer will then silently compare when it would have braked to the driver action and upload that to the Tesla database. If several cars drive safely past a given radar object, whether Autopilot is turned on or off, then that object is added to the geocoded whitelist.

Relying on whitelists seems like a hack. Then again, I'm not building it =)

Re: Text Messages Between Travis Kalanick and Anthony Levandowski

#80

Post by valuearb »

x0x0 wrote: post #52
josh2600 wrote: post #45

Earlier quoted context omitted.

Have you seen the video? It's way more jacked up than what you're implying: https://m.youtube.com/watch?v=pzzQ42D9Srw Watch that and tell me that it's reasonable or caused by a short yellow...

Holy cow. With a pedestrian in / on the edge of the crosswalk. And there was even another car that had been moving just a few seconds earlier stopped there too...

You don't think it was going to stop if the pedestrian stepped out in front? If autonomous cars can't do that, they can't do anything.
Post reply on HN