Comment by n0tth3dro1ds
3 years ago
I told you so.
https://news.ycombinator.com/item?id=33212850#33214018
> This is also how Zoox, Tesla, and Cruise do their demo videos to scam more money out of investors: they collect ultra-HD maps in a very narrow area or a very specific route. Then they drive the route/area about a thousand times, recording each drive. Then they upload the drive with the fewest mistakes to YouTube. Just like me taking a thousand half-court shots with a basketball, hitting one, and then claiming I can do it every try. There, I just gave you the formula to raise $100mm from FOMOing VCs.
You describe mapping and cherrypicking as intertwined, but they're pretty orthogonal to each other.
You can cherrypick without mapping (Nikola rolling a truck down a hill).
You can map without cherrypicking (Waymo/Cruise allow you to record your ride in fully driverless vehicles in SF right now).
It’s about scale. When you show a video of the car driving itself, you’re implying it can do it in similar arbitrary scenarios. When the car can only drive itself on a route with an ultra-HD map, yet your business model doesn’t allow you to collect ultra-HD maps across the country, and you don’t mention this in your video, you’re committing fraud.
Waymo and Cruise can drive SF because they limit the number of scenarios and they collect ultra-HD maps for that tiny area. Throw a Waymo van in Knoxville, TN and it fails left and right, even in similar scenarios due to the lack of the quality map.
Ok, and when they want to expand to Knoxville, TN, they'll have a couple people take X months to map it (in parallel with all the other cities). Mapping is probably the closest thing to a solved problem in autonomous driving.
> you’re committing fraud.
Where is the fraud? Waymo and Cruise are not selling cars. They are selling only the rides which they have the ability to provide. Not only that, they pretty prominently display mapping as a part of their stack, so it's not like investors are deceived.
4 replies →
This person just has no idea how AVs work and thinks an openly shared approach to AVs is a conspiracy
https://medium.com/cruise/hd-maps-self-driving-cars-b6444720...
https://blog.waymo.com/2019/09/building-maps-for-self-drivin...
https://zoox.com/journal/putting-our-robots-on-the-map/
When you're ignorant on a topic, it's not surprising that every fact you come across might end up being misconstrued as discovering some well-hidden secret.
I literally work for an automotive OEM and have worked in computer vision for over a decade. Nice links to engineering blogs written by my colleagues.
You are a moron and I seem to have struck a nerve. Hello, salty Tesla owner or engineer.
6 replies →
Ah, the old binary tree con: https://math.temple.edu/~paulos/oldsite/scam.html
I agree with you that these videos are extremely cherry picked, if not outright faked.
In response to the link you provided though, it's fair to add that with current FSD beta, a trip such as the one Tesla faked in 2016, is now almost trivial. So there's definitely some progress in the space. Of course the timeline Elon (and others) claimed were absolutely bullshit.
Where to start...
First off, you're saying AV companies share examples from repeated runs? What a truly shocking insight to absolutely no one. That's how you track improvement in AVs, repeating runs and comparing. There's no trickery involved in the fact you see runs that were repeated when almost all runs are repeated by design.
Second, what Tesla did is nothing like you described. Tesla lied about who was running the drive they showed. A human was driving when they claimed the human was only there for legal reasons. No one else has done that.
Maybe hold the "I told you so" for something you understand enough to actually tell anyone anything about.
> for something you understand enough to actually tell anyone anything about.
I work in computer vision and work at an automotive OEM. Left autonomous field specifically because it was fraudulent. Have many friends that helped stage fraudulent videos like the ones that Tesla/Zoox/Waymo put out regularly.
When you post a video of a “successful” drive, have the CEO tweet about it and say the car is driving itself, yet the same software version of your autonomous stack literally crashed on an earlier run, you’re committing fraud. If you imply via video that your AV can handle a certain type of driving scenario without issues, yet in the actual video your AV is relying on an ultra-HD map of that specific route that your company cannot conceivably collect for the majority of roads, you’re also committing fraud. If you think otherwise you’re drinking somebody’s Koolaid.
> When you post a video of a “successful” drive, have the CEO tweet about it and say the car is driving itself, yet the same software version of your autonomous stack literally crashed on an earlier run, you’re committing fraud
Right. Just like Steve Jobs holding an iPhone that crashed before the keynote is fraud.
Fraud is saying it's driving itself when a human is driving.
> If you imply via video that your AV can handle a certain type of driving scenario without issues, yet in the actual video your AV is relying on an ultra-HD map of that specific route that your company cannot conceivably collect for the majority of roads, you’re also committing fraud.
This is so utterly nonsensical I don't even know where to start. Driving scenarios are not fungible. Navigating around a stopped car in downtown SF is not the same as being able to navigate around a stopped car in Mumbai rush hour, or navigating around a stopped car in the middle of a California wildfire with 0 visibility and a fire roaring 10 feet away.
So when a video shows that the vehicle can handle a given situation and the vehicle handles it on video, the implication is not "we've now solved every variation of this situation to ever exist", the implication is "we were able to handle the situation you just witnessed".
I thought it ended at you not understanding EVs, but it's even worse: you don't even know what "fraud" means, and you clearly don't understand how bleeding edge technology is presented. That mentality is definitely more at home in the fossilized environment of an automotive OEM, but it must certainly be interesting to watch all that "fraud" translate into hundreds of thousands of miles on public roads and open betas with real riders... sour grapes much?
2 replies →
One difference is Tesla runs on all roads basically (subject to weather view on cameras). It's gm super cruise and google that only work on road that have been imaged. It does seem that super cruise and google waymo is ahead of tesla. Tesla is trying to solve a harder problem, random roads and conditions. I think that tesla won't succeed very soon, and as the "pre-scan the road" groups get better and better tesla's approach will be seen as unacceptable risky. I have only driven the new versions a few time, it was interesting to see all the things they could see but it just wasn't safe. Waymo has their weird stuff but did seem much safer in theory. But who knows if there is a hole, like happened in Arizona to that poor woman who got run over in the night while crossing the dark road.