Comment by systemvoltage | Hacker News Reader

Comment by systemvoltage | Hacker News Reader

Parent Full thread

Mimicking nature, the argument that if our eyes can do and therefore, it must be sufficient is... true - we have the evidence in front of our eyes (sorry for the pun). However, evolutionary processes have no hindsight and they often get stuck in local minima in terms of the optimization space. Of course, I am making the argument that having LIDAR only helps to augment the sensory input from cameras. It doesn't take anything away. I believe Tesla wanted to save a buck in its initial days and banked on cameras-only solution to FSD, continued to double down on it instead of re-evaluating the cost benefits. I am sure there are heated debates internally at Tesla team. Whether they'll use LIDAR or not seems to be a shut case for now.

Tesla's main goal is to do a lot while keeping the costs low. The fact the the entire interior is devoid of critical things like buttons is a testament to not how they're good at UX/UI, but how they've shaved off $$$. Pretty impressive. A potpourri of features in a Tesla car are all software-based (fixed NRE cost, essentially zero bloat on the COGS). Spending extra on a set of buttons, let alone a LIDAR system, would piss off the bean counters. If anything else, their marketing is great and people love the brand (and the stock).

I believe they said that LIDAR was so low resolution and was giving so many false readings that they had to give it up when using it in the real world (non pre-mapped cities like some other companies).

Yes, they have been repeating this argument but intuitively it doesn't doesn't make sense to me. How is it more difficult than stiching a dozen cameras together? Even a single point in the lidar map can provide distance information to giant median columns Tesla cars keep missing. Any computer-vision experts can weigh in on this? Isn't Waymo doing this already?

Another argument is to not fuse the data. Just keep it as a backup decision maker akin to how Aviation systems are designed. If both systems agree, then you have additional reliability. Would love to hear from someone who's worked on these kinds of FSD systems.

>If both systems agree, then you have additional reliability.

Andrej covered this in a previous talk. If both systems agree, then nothing would be different if you did not have the 2nd system. If the systems do not agree then you ultimately still have to pick one in most scenarios.

IE if you are travelling down the highway at 80mph with a car close behind and your lidar+radar system says there's an immediate obstacle but your vision system says it's an overpass, do you emergency brake?

I see, I was thinking about instances where it crashes into pillars which would be pretty obvious case for disgareement between vision and LIDAR. It would alert the driver to take control instead to crashing into it. Tesla autopilot has some scary corner cases where LIDAR can be a backup system for human intervention. But, for FSD level 5 - it's a different story.

You're making the word "just" do a lot of work here.

I disagree, I think its much more fundamental then simply saving a few $.

This goes for lidar and the interior, is simply what the believe the right solution is.

If they could save 100$ per car and have a traditional interior, I don't think they would do it.

The same for lidar, even if it was free, they wouldn't use it. They simply believe the additional complexity of sensor fusion is not worth it.

Every project Elon does I hear him talk about cost efficiency. For the second richest man on the planet he's very concerned about money and cost. Even with his rockets, the conversation always revolves around cost per kilogram (or whatever) to get stuff to space, and how his main objective is to lower that cost as much as possible. Today he talked about the importance of keeping the Tesla bot inexpensive. The importance of making compute power cheaper per metric of computing.

In regards to LIDAR - I think it's possible to drive safely without depth perception. One eyed humans do it just fine. It may not make sense to have LIDAR on every single machine you want to have vision processing with, including the Tesla bot. So it makes sense to solve problems without using it.

Stereovision is basically monovision at a distance of 6 feet iirc, the depth is inferred from cues like light, size, shape.

Reply on news.ycombinator.com