[Fricken]

>No consensus on the technology approach

Waymo, Cruise, Zoox, Pony.ai, Baidu's Apollo, Argo.ai and Aurora all have/had very similar approaches to the technology. Tesla is the major outlier and they haven't accomplished much in spite of the hype.

[Zigurd]

I don't want to split hairs but I think the focus on sensors misses the point. There's a lot of diversity in terms of other on-board hardware, software architecture, and the role of geospatial data in the AV system.

It's a product area that is very far from being able to horizontalize. Waymo Driver is going to run on Waymo hardware for a long time to come. Toyota is supposedly trying to use Waymo technology for personal vehicles. I expect adapting it will take years. The software is nothing like an app running on an operating system. All of these systems probably require years of effort to move them to a different hardware platform.

[Fricken]

>There's a lot of diversity in terms of other on-board hardware, software architecture, and the role of geospatial data in the AV system

I'm curious to know where you get information on stuff like this. The Google self-driving car project was fairly transparent in the early days but since things have gotten competitive everybody is pretty tight-lipped about the particulars of what they're doing.

[AlotOfReading]

It's a small industry and people talk. You can also infer more about the systems from the bits that are public and/or reasonable.

Unfortunately, information ages quickly. Stuff that Waymo published about their architecture only a few years ago is now wildly out of date.

That said, diversity is decreasing. Most players are standardizing on relatively similar hardware platforms using nvidia compute, with connectivity heavily focused around ethernet as opposed to older buses.

[Zigurd]

There's a lot of information about these systems out there. Waymo has has described in detail how their remote monitoring works. It doesn't rely on perfect connectivity. It doesn't rely on low latency. It can't even steer the vehicle remotely. It can tell the vehicle what it's next move should be. But all of the processing to get it to where the remote monitor thinks it should go is in the vehicle. It's details like this that distinguish between a demo and a product.

I make TikTok's about technology and project management. Elon's management style has been, some might say, a running gag in my videos, so I am more tuned into these topics than your average bear.

[imtringued]

Tesla employees are using LiDAR in secret to gather training data, because they will get fired by Elon if he finds out. Tesla is in the business of blatant self sabotage.