Friday, July 12, 2019

Prototypes

I've said this before but if the light at the end of the tunnel was just around the corner for self-driving vehicles, there'd be a working too-costly-for-consumer-sales prototype. People don't want to think about it this way because they just want to think about it as a "software problem" which is in a way true, but it's just not a problem that's going to be solved with a few more lines of code. If expensive sensors+advanced mapping+current state of software+state of the art computing power haven't yet solved the problem, I really don't know what's supposed to solve the problem in another year...or another year... or another year. A few more people clicking captcha stop signs?

Making cheaper LIDAR will make them cheaper to produce, when they work. which they don't.

The vast majority of self-driving developers, however, consider the laser sensor not vestigial but a crucial element of a safe, capable system. That’s why many of those outfits have developed their own systems (like Waymo) or acquired lidar makers (like Cruise, Aurora, and Argo). It’s also why everyone who hasn’t taken such steps should welcome Luminar’s announcement Thursday that it has developed a production-ready lidar that will cost as little as $500—cheap enough to make it work not just on robotaxis, but on consumer vehicles.

If they can't make it work with the expensive LIDAR, they can't make it work with the cheap one, as cost isn't a barrier to "making it work," it's just a barrier to making it a commercially viable product.

This was my favorite recent Musk (who thinks even LIDAR is unnecessary) tweet.



It's especially funny because he's been touting his will be released any day now, 3 months, 6 months tops, "advanced summons" feature which supposedly lets you press a button and have your car come find you in the parking lot. Suddenly he realizes this is actually a hard problem, not an easy one (it is! parking lots are a nightmare! how did he not realize this before...)