Level 2 means that the car automates several different parts of the driving task, but that you must be ready to take over at a moments notice - you still have to keep your hands on the wheel and your attention focused on the driving task in case you need to intervene. This is what Autopilot, and the FSD beta is. You can turn them on but you have to keep your hands on the wheel and your attention focused. This also means you're legally liable for any accidents, since the onus is still on the driver to intervene.
Level 3 basically means that, in certain situations, you can take your attention off driving completely and play on your phone, read a book, etc. The car might still need you to take over in certain situations, but it will warn you in advance and give you enough time to transition back to driving. Companies have largely shied away from pursuing level 3 system because it puts the legal burden on them - if the car gets in a crash when it's operating at 'level 3', the onus is on the company, since the driver was not told to be ready to intervene.
WRT your point about how they could 'surpass' Tesla, two things:
Level 3 can be pretty restrictive - I'm not sure on specifics, but they could offer this on, say, only the ten largest interstates in Japan, and it would still qualify as level 3 self driving. The 'levels' aren't really a strict progression in capabilities. A level 2, 3, and 4 solution is really three separate problems to solve, not succesively better versions of the same system. It's conceivable to me that a legacy automaker could focus on getting this working on a few major interstates and have a system they're confident in after a few years, since they don't have to worry about nearly as many edge cases as a general highway ADAS solution.
Tesla's data/fleet advantage is overstated in their messaging. Most of the stuff presented in their autonomy day presentation doesn't actually exist (follow some Tesla reverse engineers for more info). Self-driving/ADAS is not, by and large, a data problem.
It's widely accepted that the disengagement stats companies are required to report to government agencies are essentially meaningless. There's no real standardization in how any of this stuff is reported. There's no way to do independent verifiable data driven analysis of the systems.
Waymo's the only company that's put out a really robust deep dive into their safety stats - the white paper they out out was really pretty comprehensive.
Legal liability tells you how confident the manufacturer really is in its technology.
can you explain more?
level 1 = expect car to go into ditch if you're not careful
level 2 = hands on steering wheel but it should be ok (Tesla)
level 3 = hands off steering wheel, looking around, expect to have to take over control (Honda's beta test)
level 4 = either taking over control rarely needed, or vehicle has limited ability (e.g. can't drive in snow)
level 5 = you can sleep
The confusion lies around whether we're allowed to watch TV - however I'm fairly sure no country or state has passed legislation allowing us to take our eyes off the road while driving.
Level 3: driver is expected to take control, with brief notice, when vehicle cannot determine correct action.
https://www.nhtsa.gov/technology-innovation/automated-vehicl...
Level 2 requires instantaneous takeover should the system fail, Level 3 system must tolerate ~30s delay, etc. in feature checkbox style.
Automakers are fully committed to the Level system, but it does not agree with the reality that algorithms don’t fail gradually like a turbine engine would, so the reality do not agree with the Level system and Level 3 is just going nowhere.
Your confusion around Level 3 comes from the fact that Level 3 do not exist. So no worries.
I'm not confident L3 actually exists. The only real levels are 2 and 4.
It's pretty "easy", actually, by making the take-over optional. I.e. you need the driver to take over within a few seconds if you can't navigate, but you'll fall back to a safe stop if the driver can't take over. So your auto-pilot doesn't need to handle all situations (well), but you can safely be inattentive; it might just cause a sub-optimal response.
This changes the legal situation for car manufacturers dramatically and has technical implications. If the power in your car for the controlling device fails, normal cars have the driver as fallback to break. Honda needs to have some technical redundancy somehow.
Tesla acts as a third party. They leverage a tort system not designed to accommodate the type of telemetry they have to aggressively find fault with the drivers action.
https://www.sae.org/news/press-room/2018/12/sae-internationa...
TL;DR: In levels 0-2 you are driving the car and the electronics are helping to manipulate the controls. In levels 3-5, you are handing over the responsibility to drive to the car. You do not need to babysit the car to make sure it's not trying to do something stupid.