He also claimed he wanted to make self driving safer than humans. You can't do that well if the car has the same visual limitations a human has.

Musk maintains that camera-only technology is the most “human” way to approach self-driving, since people use their eyes to navigate the road.

A few years ago I was driving on the motorway, came up on a bend in the road and was greeted by a dense freak fog bank out of nowhere. I immediately let go of the accelerator to reduce speed, at the same time my dashboard lit up like a Christmas tree and a collision alert started blaring. That gave me enough time to apply the brakes and prevent a collision with the first and only traffic jam I've ever seen there.

I see no reason to not augment our own capabilities with radar and lidar, to see what we as humans can't.

This was him justifying what was a cost saving decision that became a face saving battle for him personally as everyone told him he was wrong.

If there is one thing Elon cannot stand above all others is admitting he was wrong, especially when he has spent years promising this and now he would have to retrofit at his cost Lidar to all those cars he sold as self driving ready with an expensive optional extra.

He might be able to avoid any sort of punishment from the US government as long as he stays in Trumps good books, but he will not be able to do so in Europe or similar.

waymo is already superior to tesla in self driving,waymo has been around for a few years already. teslas already too late into the game.

The only viable competition to LIDAR is structured light (see Leap Motion, there's equivalent sensors for cars), which uses an IR source with patterned light and multiple high frame rate cameras to calculate depth from the reflections. In theory light field photography with special lenses is possible too, but far more computationally heavy for real-time use IIRC

There's some safety issues with LIDAR at close range (it's a laser! it can damage cameras, etc), which is basically the main reason to not use it. But Tesla are dumb enough to try to replace them with cameras alone, and not even using proper multi-camera techniques to calculate depth

I bet it has to do with the costs to implement.

There is a segment of highway near Budapest where cops often hang out on an overpass with a speed camera. Teslas do a light phantom brake under that overpass, like going from speed limit to 80% speed limit, because apparently the car learned that that's what it should do there.

I wouldn't have believed it's that stupid, but there is a whole discussion on local owner groups about that specific spot, and I've also seen it first hand from the passenger seat.

Relevant:

I mean the data IS sanitized, but not to the level that would have required certain human things to not happen.

Part of what's led to its improvement over the years is better going through the data and removing bad things or properly labeling them.

That left turn that was cut to short makes it into the first set of training as a cursory look at it seemed okay, and then they see that cars are cutting turns to short. So they go through the data again and try to find examples of it and then label them properly so it doesn't think it's okay.

But that's not a simple process, and then trying to only have certain good behaviors gets really hard because they're actually very uncommon in normal driving because the bad behavior is socially acceptable.

You can establish rules in top of the neural net output. No acceleration higher than 0.3g. No speed higher than the limit + 5 (if you know what the limit is, which teslas struggle with to this day). No running of red lights.

The thing is that you have to break the law to be an effective and even safe driver. Going way under the limit or refusing to go into the opposite lane for a moment means that you cause traffic congestion and piss people off. Waymos definitely break laws at times, I've seen it personally. And other times waymos get too "safe" and end up locked in place for 30 min at a time. The real world is a chaotic place and there's always been a discrepancy between what the laws are and how people actually drive. Lidar helps see things humans cant, but the main problem is the intelligence required, which may improve over time

The worst Telsa driver I know. Hum. That’s a tough competition.

Yeah that's easily $100 her could be pocketing himself.

Someone who thinks laser is lame is a huge moron. Case in point.

Compared to the Cybertruck's design, I don't think having a LIDAR antenna on top of the car is so bad.

Looking at their recent facelifts, I don't think they have many good designers left

This has nothing to do with aesthetics and everything to do with reducing production costs. The sensors themselves add to costs, but then the data processing for all of this data adds to the computational needs of the system, sensor installs requiring calibration and flexibility for changes in sensors add costs. The list goes on.

Musk and Tesla are standing firm on their position that LiDAR is needless and that others will follow their lead here, but so far in practice the lack of LiDAR in Tesla vehicles has shown to be a serious disadvantage compared to other systems.

Yeah, the "appeal to nature" fallacy that Musk is using is really dangerous. I want the best tools that are practical for the job, since this has actual life and death consequences. A mix of sensors that work together and covers each others limitations.

Not sure what you mean. He's always been pretty involved in Tesla designs.

His particular degree of involvement in this decision has been well documented for years, long before the consequences of those decisions.

https://www.forbes.com/sites/bradtempleton/2019/05/06/elon-musks-war-on-lidar-who-is-right-and-why-do-they-think-that/#19deb44f2a3b

https://insideevs.com/news/658439/elon-musk-overruled-tesla-autopilot-engineers-radar-removal/