But level 3 *is* eliminating a driver that needs to pay attention... Level 3 requires the driver to be awake and conscious, yes, but not that he's paying any attention. He can read, work, socialize, whatever he wants. Except go to sleep. He's only required to take over if the car prompts him to, and the car must give him a few seconds warning first. I.e. a controlled handover.
I will be *extremely* surprised, to say the least, if Tesla ever releases a true level 3 system OTA with their current sensor- and driver monitor system, at least on anything else than certain US highways ...
But who knows. I like a surprise
Before writing I referenced the Wikipedia article and the definition of 3, 4, and 5 was a bit vague. I do think Level 4 will be achievable on restricted access roads like highways, but I just don't see it on city streets, there are too many things that can go wrong.
The incremental changes in Air Craft automation are largely slowed by legal forces and cost constraints not technical. The plane I fly now has 1990's technology. Some new stuff is coming with ADSB with in/out tech but hell I flew a AC with auto-land 15 years ago. The tech is there to have a plane do everything and if we look at the accidents in the last 15 years they are all caused by the humans at the controls. (Of course we don't hear about the accidents that did not happen because of the humans that intervened) I fly in KLGA (NY Laguardia on a regular basis and the plane will turn the wrong was while on AP when joining the localizer to runway 22 about 20% of the time. Now it is not really going the wrong way but the algorithm to determine turn rate is kinda wonky so it is trying to bracket the course...the Air traffic controllers don't like this so we take over and once locked on course it is fine. I am sure that new tech is probably better at this..but it cost a lot of money to change existing planes.
I think we will probably see complete autonomous planes carrying people before I start to push up daisy's.
I do agree the edge cases for car's are dramatically tougher. So maybe level 3 in 90 percent of road areas that are Geo fenced off?
I was at Boeing in a group that did engineering testing on the LRUs for commercial aircraft from 1987 to 1994. I worked on the ARINC interface hardware. I remember when TCAS came along and I was there through most of the 777 testing. The first few planes were in flight test when I left. My last day I snuck onto airframe 1 when it was parked across the street from my lab and looked around when everyone was at lunch. The only time I've been on a 777.
We had flight deck simulators used for laying out the instruments and to get user feedback on the designs. There was a joke that the next gen flight deck would have one pilot and a dog. The pilot's job was to feed the dog and the dog's job was to bite the pilot if they touched anything. I had a feel for the tech available in 1990 and that design was actually technically possible, but the FAA would have gone into fits if anyone suggested it.
The regulatory agencies are understandably very conservative. I was one of the lucky engineers at Boeing who was actually designing and making something. Most sit around doing probability studies to meet FAA requirements. I had a friend in the engine group on the 777, he said the FAA was requiring they include armor plate on the inside of the engine nacelles because they couldn't prove that the likelihood of a fan burst going through the entire plane and taking out the engine on the other side was improbable enough. It has never happened in the history of commercial aviation (though it did happen to a small twin engine Cessna once and it was a prop blade from the #2 engine taking out the #1 engine).
The FAA also has fits if AP gets something wrong. The entire 747-400 fleet was grounded when an early 747-400 tried to fly over the north pole (Scandinavian Air) and the plane did an S maneuver around the pole. The computer couldn't handle 0,0 coordinates. Nothing bad happened, but the fleet got grounded anyway. I had to go to Everett during that time and they had brand new 747s crammed everywhere. I have never seen so much aluminum jammed together.
But it does make sense to be conservative with commercial aircraft, the results of something going wrong are catastrophic.
Disclaimer - not a pilot but have been an air medical crewmember for about 10 years so I'm an informed neophyte when it comes to aviation things but I will defer to pilots on this stuff
Building on what
@Unpilot said above, it seems like the aircraft autonomous function problem while not easy is still inherently easier than the car one. Yes, there are some very real challenges you only get with aircraft - 3 dimensions, weather much more of an issue, you can't just tell the plane to pull over if system can't figure out what to do, the risk of screwing up is potentially quite a bit more loss of life and/or property, etc. But it seems like the two biggest complicating factors for fully autonomous car operations are the other drivers and the lack of control over the roads. (By that I mean that changes can happen to a road at any point and without needing approval from the feds, also a big pothole can appear, road debris can appear, etc.) Those factors are far more controlled in the aviation environment.
Hands off and eyes off autonomous car operations seem likely to happen in the near future only in controlled settings - special lanes that are sectioned off from normal travel lanes and include only vehicles that have similar autonomous features. This infrastructure opens up lots of possibilities like communications between cars to coordinate maneuvers and that could allow for more efficient traffic movements. I can see congested areas attracted to invest to implement these features on highways because it can create highways that move more vehicles more safely and that creates savings in a variety of ways.
While level 4 and 5 automation would be great - what I want most is to have the car be able to make reasonable interventions to mitigate or even prevent a serious incident if I'm distracted or fall asleep at the wheel or make a really bad decision. I'd also want the car to monitor my attention level to try to mitigate things before it had to intervene. I'm fine with the assumption for now that I'm operating the vehicle. This is like an intelligent risk averse copilot. That is basically level 3 (with some fine points of difference). It puts the emphasis on safety and not as much on convenience feature.
Regardless of what Tesla has said they can do, that is what I want my car to be able to do.
With aircraft there are basically three failure modes:
1) Mechanical failure
2) Pilot error
3) Something colliding with you or act or war (terrorism also being an act of war)
Planes rarely are in any situation to collide except around airfields. Reliability has become good enough that most accidents today are from pilot error, though probably half or more of those are pilot error that turns a relatively minor problem into something catastrophic. The Air France plane that went down over the Atlantic was a mystery until they found the flight recorders over a year later. It turned out the plane was being flown by the junior most member of the crew while the two more senior pilots slept. The pitot tube iced up and the plane couldn't determine airspeed. The junior co-pilot did all the wrong things and the senior pilot didn't figure out what he was doing until they had run out of altitude and slammed into the Atlantic.
The results of a problem in the air can be much more newsworthy than a car accident, but they are rarer and there are fewer problems to solve. With air transport there are some general aviation people who are not the best trained people in the world, but the worst trained general aviation pilot has more training before getting their license than 90% of the people driving cars. The majority of people in the air are professionals who are highly trained especially when you get above lower altitudes.
In aviation you do need to be concerned with thunderstorms, wind shear, ingesting rain and hail at low altitudes, and one 747 was once blinded by ash from a volcano. But all commercial airliners have weather radar now and they have had regular radar for some time. They generally know what's around them at all times and with TCAS, the planes can communicate with one another and steer away from each other if both planes have the system and are in danger of collision. I think all airliners have TCAS at this point. It's been around 25 years at least.
On the ground there are all sorts of objects that can confuse an auto driving system. There are other drivers who are poorly trained, could be high on something, or just aren't paying attention. Then there are various animals that can "appear" out of nowhere, including humans. Weather at ground level can have much worse visibility and can confuse sensors much more easily than aircraft sensors. The radar suite on an airliner is probably never going to get caked with snow, but Bjorn Nyland found that snow buildup on the nose of a Tesla was common and it tended to blind the radar.
Ground transport are also more prone to bad actors messing with them either for pranks or for more nefarious purposes. Imagine crowded streets like Manhattan with most cars automated. A few people with balloons filled with paint could paralyze traffic for many blocks in all directions. Less obvious tricks would be to set up transmitters that jam radar/lidar.
Another question I don't think anybody has answered is how will cars manage in an electronic soup of other radar or lidar units operating around them, especially in a place like Manhattan where stray signals will be bouncing off the buildings.
These are just the things that occurred to me sitting here writing this. I've thought of many more that I have now forgotten. When I'm out driving I go into software design mode and think about how I would tackle this or that problem. If I had written them down, I would have a pretty staggering list at this point. I have worked on some relatively complex embedded systems, but the complexity of automated driving is staggering to contemplate.