Live data from Hacker News

Tesla Live Stream – Autonomy Day [video]

livestream.tesla.com

271–280 of 550 posts

Re: Tesla Live Stream – Autonomy Day [video]

#271

Earlier quoted context omitted.

Elon Musk is noted for his premature optimization. Case in point: both the Gigafactory (vastly overbuilt for the quantity of batteries actually produced) and the Alien Dreadnought (vastly overbuilt for the number of cars Tesla current produces...assuming that Tesla is ever able to get the fancy automation working). Boring Co digging a two-mile tunnel in West LA without bothering to learn how to pour concrete smoothly…

It's easier to optimize a working system than it is to get an optimized system working.

So when are they going to get them working?

Re: Tesla Live Stream – Autonomy Day [video]

#272

Earlier quoted context omitted.

The presentation makes it clear your claim is entirely false, you should watch it.

Which presentation did you watch? Karpathy said specifically "it's not a massive amount of data, it's just very well picked data" when talking about how the cars only send data when one of the configured triggers fires.

There’s a large gap between ‘a few frames’ and a massive amount of data, and the amount sent lies somewhere in the middle. Clearly they can’t send all data (nor would they want to) but it seems it is sufficient for significant learning to take place and the examples shown were good quality over at least a few seconds, so hundreds of frames for each example.

Re: Tesla Live Stream – Autonomy Day [video]

#273
post #186

Earlier quoted context omitted.

Great-grandparent comment suggests that they did not, in fact, design for it. Assuming the best is nice when you have no information, but we have information.

I'm not assuming the best, I'm assuming basics. We've just watched a presentation where Tesla put immense resources and talent into designing a state of the art vision processing chip. Do you think they're going to let it be rendered useless by not implementing simple physical measures to keep vision coming in from the cameras? (What measures are already there plus easy opportunity for improvement). Of course not. Su…

I have seen recent Teslas in person in the showrooms, and there doesn't appear to be any wiper for the cameras. Especially with a firsthand owner account just a few comments up I think it's actually only logical at this point to assume they didn't put wipers on the cameras.

Re: Tesla Live Stream – Autonomy Day [video]

#274

Earlier quoted context omitted.

It's easier to optimize a working system than it is to get an optimized system working.

So when are they going to get them working?

Beats me. It could be a while before we see anything that can compete with conventional rideshare at scale.

Re: Tesla Live Stream – Autonomy Day [video]

#275
Seems to me that relying too much on Machine Learning for this is actually the real risk, not LIDAR. Tesla's competitors are using ML, but combining it with sensor fused data from LIDAR, Radar, accurate maps, and visual data. In other words, they're starting from a view of the world with super-human sensing.

Starting from a purely visual domain likely dooms your system to making the same types of mistakes that humans make visually in judgement. At least with accurate maps and LIDAR as backup, you can sanity check output of your visual processing against a map and LIDAR. If your claim is that the map might be out of date, or the LIDAR too low rez, it still helps to err on the size of caution.

In the worst case, the map tells you you can't go somewhere that you're allowed to go. In the case where it says you are allowed to go somewhere were you shouldn't, well your visual system should be telling you not to go there. If it isn't, you're in a lot worse trouble than having a map with flaws.

Likewise, the claims that "LIDAR is expensive" is like claiming EVs are expensive because "batteries are expensive!" If AVs become common, then they'll be a huge demand for LIDAR and the costs will decline. LIDAR costs are already declining. Maybe if Elon took some of Tesla's miracle engineers and had them make a LIDAR, they could not only defeat Nvidia at chip making, but defeat all LIDAR makers as well. (sarcasm) My guess is the real problem with LIDAR is drag and vehicle trim/styling.

At this point, Tesla is basically stuck. They bet big early on a shitty sensor suite before AV technology had been worked out, and wave their hands about how using consumers as guinea pigs to feed them fleet data will magically fix deficiencies in sensors. Well, what if this is a false hope and it doesn't? It would mean they'd get sued by everyone who bought the AV suite as an option and have to issue refunds or recalls to upgrade.

I've said it before and I'll say it again, you don't ship AV as an MVP on a $100k vehicle and promise magic upgrades and fixes later before the technology is even close. It's risking people's lives and it's already killed people.

Re: Tesla Live Stream – Autonomy Day [video]

#276
post #247

Earlier quoted context omitted.

If the current state of autopilot is even tangentially related to their progress on full self-driving, then it's a lot more than a few years away.

How recent is your experience? I’ve owned a Model 3 for over a year and the pace at which autopilot has been improving is staggering.

I did a road trip this weekend and used autopilot for more than 400 miles on a brand new P3D (with the FSD option, limited though it is right now). Mostly freeway, some rural roads. Here are my observations, I assume it's pretty typical but obviously all I have is anecdata.

Lane following works effectively, but it wanders a bit. Mostly not enough to be bothersome, though sometimes it will hug one side or the other close enough to make me think about the big truck next to me. Sometimes a little late reacting to a curve, though not to the point of going across the line. Planned lane changes work pretty nicely. Identification of vehicles around the car is good towards the front and noticeably spotty to the rear quadrants. It reliably misidentifies the merging traffic lane as one big lane as soon as the line ends, and promptly drives to the right to straddle the 'middle' and then corrects as the line converges to the normal width.

Adaptive cruise control is flat out disappointing. Granted, I've never driven anyone else's automation, but it actually made me motion sick at times. I thought it did a really admirable job of handling people merging right in front of the nose of my car without jamming the brakes. Just smoothly slipped back until the right following distance was achieved. But in open traffic, when I came up on another car, it would abruptly slow down a bit too aggressively, and then fiddle with the speed a bit until it matched the leading car. If the lead car switched lanes, it waited until several seconds after the car was in the next lane and then abruptly sped up to the programmed max speed. There were times I could tell how frustrating this style was to people behind us who were not using a similar system.

Twice it braked suddenly, though not any kind of panic stop, and then immediately sped back up. Like it saw something concerning and then changed it's mind a split second later. Made my wife a little nervous after the second time, but it didn't happen again the rest of our trip.

Overall I'd classify it as a moderately drunk driver :). I let it drive us into a couple moderately uncomfortable merge situations just to see how it would deal with traffic entering the freeway, and while it didn't get us in a wreck or anything, it didn't make me feel like AP is on the verge of becoming viable FSD.

For what it's worth, this was on I-5 in between Portland and Seattle, in mostly light traffic. We have what are, in my experience, pretty easy relaxed freeway configurations. Long ramps, no ramp lights, etc.

I borrowed the P3D so I could have myself an extended test drive to decide if I really wanted a Model 3 or not. Wicked fast (which is still an understatement), I love the power but would probably choose to steer it myself for the forseable future. I wish they'd put in nicer seats (like, my legs were getting numb at one point!), just standard Recaros like everyone else would be perfectly nice. I think I've decided that I really like it but I'm going to wait for Model Y before pulling the trigger. I will appreciate the additional head space and seat height, and hoping that in the meantime they do v3 or v4 of the seats.

Re: Tesla Live Stream – Autonomy Day [video]

#277

Musk: "You're only going real fast in the forward direction." Dozens of times in my life, I have been driving down the freeway at 70 MPH in a 60 MPH zone, and I've noticed someone weaving through traffic behind me, going closer to 140 MPH. I need to know whether to change lanes, stay in my lane, stop changing lanes, pump my breaks to indicate there is slowdown ahead of me that car might not see, etc. Just food for th…

I’ve been driving for nearly 30 years and have never even seen one person going 140mph, yet alone dozens

Re: Tesla Live Stream – Autonomy Day [video]

#279

Earlier quoted context omitted.

lidar is not required and is expensive. Given that the only the company claiming that is the one whose cars sometimes hit (or even swerve into) clearly visible stationary concrete barriers in broad daylight [1, 2, 3, 4], I consider that claim far from proven. [1] 2019 https://arstechnica.com/cars/2019/03/dashcam-video-shows-tes... [2] 2018 https://news.ycombinator.com/item?id=16772748 [3] 2017 https://bgr.com/2017/03…

This argument misses the forest for the trees. Even level 5 autopilot will sometimes crash into objects that are clearly visible to most humans, as will lidar-based solutions. But if the robot fleet is an order of magnitude safer than the human fleet, then it makes rational sense to prefer a robot driver to a human.

I think you might have replied to the wrong post here?

I didn't say anything about humans vs self-driving cars. I'm arguing about self-driving cars with LIDAR vs those without.

Re: Tesla Live Stream – Autonomy Day [video]

#280
post #139

Earlier quoted context omitted.

Does anyone doing lidar not also do visual recognition?

I don't think any one is doing this. This is the point Karpathy is making. Vision is essential and needs to be perfect and also has all the information to recreate the 3d environment. That's why lidar is not required and is expensive.

Vision is not perfect though. It is impaired during night time and difficult conditions and can be obscured if dust/dirt gets on the camera.

And not sure who you are but I trust all of the other companies who are doing this a lot more. And they are all using LIDAR.

Post reply on HN