Oh yeah, the current HW is far from ready for primetime. I do appreciate the foresight of Tesla leaving space for computer upgrades but the camera position is almost certainly not up for it (note how terrible it is at sensing rain on the windsheild, the cameras being blinded by the winter sun, back up camera being covered by rain, etc). This isn't to mention the inherent limits of small digital cameras (CMOS jello effect, low light wash out, dynamic range of single frame video, CMOS sun blindness, etc.) and the complete lack of time-of-flight sensors (why are we limiting the car to only what humans can do? Isn't the whole point to do better than humans...). These are just some of the hardware related challenges that Tesla faces with regard to true robotaxis.
Also, what is Tesla doing with all these Model 3 leasse turn ins from 2019/2020? You know, cause they wouldn't let owners buy out leases since by 2023 Tesla would use them as robotaxis. They have to be piling up somewhere...
I was always skeptical of the robotaxi claims. Even if Tesla was able to get the software to a point where they thought the car could drive without a human at the controls, regulators likely would disagree. I've worked on embedded software, though nothing quite like an autonomous driving system. I've mapped out what it would take to do a real full self driving system and I don't think a system that would work under all conditions can be put on a car.
Aircraft can be made to fly themselves because there are vastly fewer edge conditions and where most of the edge conditions exist (around airports), there are people on the ground watching everything. People do fly without filing a flight plan, but almost all aircraft flying any distance do file one so people on the ground know where the plane is supposed to be. There are few enough aircraft in the air that they can be monitored by people on the ground.
There are vastly most cars traveling vastly more complex routes with no active outside monitoring. Then add in all the edge conditions. Other cars doing crazy things, pedestrians, animals, objects like loose balls from a playground, etc.
Then there are all sorts of weather conditions. Heavy rain, mud, snow, fog, etc. Every winter my backup camera gets pretty dirty just driving around town. It makes backing out of parking spaces more difficult, but I have an advantage over a camera on the car when I'm driving. I can move my sensor suite (eyes) around to get a better view of my surroundings. If things are really bad I can get out of the car and see what is going on. The cameras can't move with respect to the car.
One time many years ago I was stuck in the thickest fog I've ever encountered. It was so bad I mistook a mailbox for a stop sign. I knew the stop sign was close so crawled forward until I found it. Then I had to make a right turn onto a busier street. The way I made sure no cars were coming was by rolling down the window and listening for other cars. Something a camera system can't do.
Then there are many more things a nefarious actor could do. People could blind cameras with laser pointers or balloons filled with paint. Various types of sensors could be jammed electronically.
If all cars have a self driving system that can talk to all other cars and there is a central monitoring system, then driving without humans may be possible, but trying to train a computer to do the job that humans do and work in an environment with humans doing human things with no human in the loop is an almost impossible task. It will make decisions as well or better than other humans 99% of the time, but it will also get confused sometimes.
Humans get confused and do the wrong things too. If they are too bad we take away their license to drive. But humans have creativity and computers don't actually have it. AIs have been trained to emulate creativity, but they are still following a script of some kind. Present humans with a situation they didn't anticipate and a fair number will do something stupid, but quite a few will come up with something on the fly and may end up doing something brilliant nobody ever thought of. Present an AI with something it isn't programmed to handle and it will lock up.
Cases where pilots saved their passengers and sometimes their aircraft in a crisis were situations where someone did something brilliant at just the right moment that a computer was incapable of doing. For example the case of Flight 1549 which crashed on the Hudson in 2009.
US Airways Flight 1549 - Wikipedia
The pilot made a brilliant split second decision that saved everyone on board.
But then there was the Air France flight that crashed into the mid-Atlantic because a junior pilot didn't realize that their airspeed was off because the pitot tube had iced up. Humans can do the wrong thing too. A computer may not have done better though. It took a senior pilot to realize that the instruments were giving them bad data, but he realized too late.
Cars will get to a point where they can drive themselves most of the time, but I don't think we're going to get rid of the driver in most conditions any time soon. They may be able to take the driver out of the loop on open highways with limited traffic entry and exit in good weather, but even there when weather gets bad, it will be back to manual control.
flybitlux.com
www.teslarati.com
www.10news.com
www.notateslaapp.com
