Hacker News Reader

Comment by arijun | Hacker News Reader

One thing that gets mentioned is the many thousands of miles that have been driven by Tesla vehicles, giving them sensor data in so many situations that may not have been seen yet by other companies. We may have to wait and see how much of an advantage that actually gives them (if any).

View on HN

All the people I've talked to who have worked on self-driving cars seem to think that Tesla is working blindfolded by not having LIDAR. Time will prove if they are right, but certainly the biggest jump in self-driving car ability was due to adding LIDAR rather than any software or data trick.

Musk's opinion is that passive image recognition (ie just cameras) is a requirement for full L5 autonomy. If you have to build that, then LIDAR isn't necessary.

You may get a significant boost in ability with LIDAR, but it's possible that over-reliance on it will hamper those companies' ability to develop passive image recognition and they will end up stuck.

That opinion certainly seems possible, and Musk does seem to have a track record of being very intelligent and perceptive, and is known to develop his opinions from first principles. So... I'm generally inclined to believe he's correct.

The problem with Musk's argument is that there isn't any indication so far that the machine learning has tricks up its sleeve that are sufficient resolve all of the ambiguous situations drivers come across. Having direct, accurate measurements of the environment serves as a ground truth that you don't have with just cameras. Humans don't have LIDAR, but we do have a different ground truth: we have a mental model of the world, and can reason about it. It's not just that we know things like 'fast cars take time to stop' or 'he looked at me so is probably aware of me', but rather being able to synthesize based on really complex behaviors we've never seen before, and in many cases, direct communication. All that is to say, there is a lot more to human drivers than their eyes, and for the moment there is nothing showing that we're on the cusp of developing real AI that can address those concerns. I don't think it's just a matter of labeling more data and GPUing bigger models.

So while Musk might be right if we get such AI, it's not clear that it's about to arrive, and even if it does, we can still use LIDAR. In fact chances are that if you have such great AI, additional sensor inputs will turn it from 'Level 5 Autonomous Pilot' to 'Ultra-super human pilot who can preemtively react to situations you didn't even realize were possible in the moment.'.

Of course there are questions at hand:

1. Is Elon right and I'm wrong? Maybe strong(er) AI is right around the corner. 2. Is current LIDAR tech good enough to reliably provide the ground-truth I spoke about?

Elon Musk could easily be wrong. Has been wrong multiple times in the past.

But one of Elon Musk's most underrated traits is his adaptability: if a genuinely better/cheaper technology becomes available or if the current picked solution just isn't working, he'll switch to the better option quickly.

For instance: SpaceX originally scoffed at the idea of vertical landing rockets, thinking that parachutes were just fine for recovery. They switched after parachutes failed and then Elon sent out a company-wide email of Masten Space Systems (at the time, not much more than some guys in a garage) doing this: https://www.youtube.com/watch?v=H2Yt5L5TlGM

It's not even that hard to change your mind like that. It's just that usually people don't do it because they invested so much emotionally/verbally/etc in telling people they had the "right" solution and don't want to lose face by changing course.

The issue is that Musk committed to the "LIDAR is unnecessary" path when he told users that the 6 figure cars he was selling had the necessary hardware to support L5 automation. There is going to be blowback if he suddenly says "Whoops, you gotta buy another or drop 20k on upgrades if you want SDC capability".

I don't think Musk would spring for LIDAR until it's under $1000. If necessary, they can then find a way to retrofit the hardware onto those who already paid for the full autonomy option (maybe tens of thousands?), and for others, they can wrap the (relatively modest) cost into any future autonomy upgrades. "Enhanced autonomy" is a $5000 option for the Model S, with "Full autonomy" being another $3000. A $1000 LIDAR system definitely fits.

The thing is if you have LIDAR and video you can create video only models using the LIDAR data to use to teach the video models.

I have read about the sensor suite on Teslas on the road being different from the suite on the Waymo cars, for example. So, the sensor data from Tesla may be more or less valuable, regardless of the amount of it. Wait and see is certainly the game.

Has there been any report on how much data is actually uploaded back to Tesla? I would be pretty pissed as a owner for uploading all of my rides back for privacy as well as bandwidth reasons.

Reply on news.ycombinator.com