I’ve been a bit disappointed for some time now with the progress of self-driving cars and I’m only now getting around to writing about this. There are a small handful of non-vaporware projects and companies still working on self-driving vehicles in a way that you can contact them through a sales channel and actually get a response.
The rest of these businesses claiming to be working on self-driving vehicles aren’t really in any meaningful way. You cannot contact them and get a response, or they’re looking to work with manufacturers directly and it’s never going to happen. I really don’t know how they stay in business other than selling solutions to other people who want to be working on self-driving cars and selling those small businesses hardware and software, but of course, you can’t actually contact them and get a response. So as far as I’m concerned, they’re vaporware, or lying. The latter of which case does actually seem to be the reality, as some apps that are supposed to control these systems are not actually published on the App Store, Google Play, or some alternative app store.
As of July 2024, Waymo, Cruise, Baidu, Tesla, and Comma appear to be the only real companies working on real, tangible, visible, self-driving car technologies. There are a plethora of others, some are derivative companies, some are too small of businesses to mention, or they’re working on trucking solutions. These are the largest players.
Waymo, Cruise, and Baidu use lidar, and seem to have no weird ideological hangups on solutions to achieve self-driving cars, and are shipping commercial and private systems that aim at achieving SAE Level 4 self-driving. They use maps, GPS, lidar, cameras, and compute. Everything they can at their disposal without discrimination.
Tesla and Comma are the only businesses shipping SAE Level 2 self-driving systems worth talking about from the perspective that they want to be shipping “self-driving” systems. Toyota Safety Sense is also an SAE Level 2 self-driving system. It disengages often and is advertised as a “safety and convenience” package. Cadillac ships Super Cruise. It’s described as “hands free driving.” Consumer reports has tested many other manufacturers SAE Level 2 self-driving features that don’t market themselves as “self-driving.”
Unfortunately, Tesla and Comma are not making meaningful progress anymore. Lidar is expensive, and Tesla, or specifically Elon Musk, uses the argument that humans can drive with only vision, and so should computers, as an ideological limitation that is preventing their engineering teams from making any more meaningful progress on Tesla Full Self-Driving. They also don’t use any sort of 3D mapping, because… they can’t.
Comma is also ideologically stuck. Openpilot can’t turn left or right. The comma 3x device that they ship essentially now only provides you the best lane keeping in the industry, but doesn’t do much more than that. It changes lanes too aggressively, and they’ve removed navigation features. Their engineering team claims they will bring them back eventually, but who knows when that’ll happen. Their hardware is also limited in that while Tesla vehicles have sufficient cameras and compute, for some reason Comma thinks the comma 3x is also sufficient, even though it can’t see completely around the vehicle. It has two cameras facing forward, and one into the cabin for driver monitoring.
You can’t see behind your vehicle. Comma claims they are “solving self-driving,” and they can’t turn right. They can’t even make a device that drives the speed limit based on reading speed limit signs, the latter of which any Toyota Safety Sense system can do. You need to set the speed yourself.
There’s apparently no one else working in this space to deliver consumer-owned self-driving solutions, and it’s driving me crazy.
I am working on a hardware array to ship 8 5-8MP cameras that you can attach to any vehicle, a GPS unit, and a 2-in-1 laptop that serves as your dash touchscreen and compute unit for machine learning and as a self-driving development kit.
I would love to sell a Velodyne VLP-16 along with the kit as an optional add-on, but understandably, I think many developers will not want to shell out for the $4,000 option. That would make the total development kit somewhere in the ballpark of $8,000. But if we could allow drivers to submit 3D mapping data, it would be a better value than Tesla Full Self-Driving. The current state of the art is still to use common machine learning techniques, just specifically trained on traffic signs and driving data.
I have to think someone else is out there doing this right now and I just haven’t read of it. But progress doesn’t just happen; you have to get it done yourself if no one else is going to do it.
Thanks to Jacob Robbins for reviewing drafts of this post.
Leave a Reply