I don't see great advantages in Lidar and Radar, because obviously most cars function without them and need only two eyes. But I could imagine that HD mapping has certain advantages. However, Tesla could create more precise maps or just some other local hints for the few difficult crossings while leaving all other areas to normal FSD.
I have a similar train of thought. I'm very impressed by what Tesla has with vision only. It's a low resolution approximation of what you'd get with LIDAR, but it comes with the added benefits of not having to do sensor fusion, and it comes out of embarrassingly cheap consumer hardware. It's not _good_, but that wasn't ever the question... the only question is "is it good _enough_?"
Tesla understands a very important point that I feel Waymo never grasped: you can do a _lot_ with low resolution sensor data. Waymo has made fantastic strides with their sensors, and the high resolution point clouds are really impressive and powerful for offline analysis... but they just aren't necessary to drive a car.
I think Waymo lost their technological lead due to spending too much time on sensors and not enough time on prediction and planning, and now multiple robo-taxi companies are at its heels.
Regarding maps; early on while I started really digging into this technology, I didn't think maps were too important... but over time, and especially after seeing the common failures of FSD, my mind has been changed. Even FSD now requires the presence of map data that is far more detailed than typical 2d maps.
The cars should be able to handle entirely new situations that do not align with the maps - and I'm impressed with how FSD does this OK a lot of the time - but there isn't any doubt left in my mind that HD or near-HD maps are a requirement for a smooth typical experience. They help so much, and are able to be maintained pretty easily.
So yeah, driverless might not require LIDAR, but I really think it requires remote rescue and maps. Those two last points are why I don't think Tesla will get to driverless with their current approach. I've said many times, but I'm pretty sure we'll get a good consumer system with HW4 and maaaaaaaaybe HW3 (or some sort of HW3.5 retrofit), but we just aren't going to see driverless from Tesla unless they start doing the things that the other robo-taxi companies do.
I just want to drive home the point on how different driverless and non-driverless are. True robo-taxis have some embarrassing moments where they get stuck behind cones, or road closed signs, and they also appear to be a little over timid compared to FSD at moments... but even with all of those problems they are still competing at a level above FSD.
If FSD Beta can get to be good enough for general release it would be a confirmation that Elon's idea works. Otherwise, it would confirm Waymo was right.
Depends. FSD will almost surely be released as a level 2 ADAS, not a driverless robo-taxi. Having a backup human driver gives them a *lot* of leeway.
Unrelated, I want to see Tesla admit this and start putting in more controls for the driver. I think it would be really cool if you could tell the car when it should or should not try to pass other cars, for instance. A lot of the poor behavior of FSD could be mitigated by this.
I doubt Elon is going to allow this though. He's very adamant that such controls are unnecessary, and including them would admit "defeat".
But will it happen sooner in top 100 cities of US / the world ? Personally I've no great interest in systems that I can't buy for my car - I see very little benefit of Robotaxi over Uber for customers. Yes - the robotaxi company may make more money compared to Uber - but having more rich companies in the world is not an aspirational goal for me.
I don't really use taxies much either, but I do believe robo-taxies are going to be huge. They could be cheaper than owning a car in a city, and will be a great public service for the elderly and intoxicated. They could also be made smaller, and take up less space on the road. I think they're the future for "public transit" with larger autonomous vehicles doing routes, maybe even cross city, and smaller vehicles doing more last mile sort of things.
cnevpost.com
