... And some have argued that L5 is an asymptote, something that AV's can get really close to without ever actually reaching. Perhaps replacing L3-L5 with a more gradual L4, would be more useful.
This is a great topic. In my view, L4 with pre-drive calculated restrictions would be a major asset to those who cannot drive for reaons of disability, age, sobriety state, past-record legal restrictions etc.
Of course it is necessary that the projected probability of an accident should be extremely low. However, it's
not necessary that the projected probability of
controlled disengagement, autonomously stopping to park or at least to pull safely out of the roadway, must also be incredibly low. This is an important distinction, and it mitigates some of the "march of nines" performance goals that people bring up when discussing L4/L5 autonomy.
The idea is that the requested L4 drive is
quite likely though maybe not
extremely likely to be fully successful. Say 99%, 99.9%, we can debate the threshold and it could even be user-specified*. This pre-estimate would be based on knowledge of the available routes, traffic conditions, construction projects, weather forecast etc.
Being L4 with an unlicensed or currently driving-incapable passenger, the manual controls (if even present) would be offline. Again, the expectation is reasonably high that the drive will be autonomous and is unlikely to encounter a disengagement. But in the event of one, there is a set of recovery options:
- Whatever situation that forced the disengagement is likely temporary and L4 re-engagement can be requested after a time.
- If the car does have controls, a human driver can be called upon to take over:
- Any licensed passenger if present,
- A field servce agent,
- Law enforcement/responder personnel (but perhaps only to move the vehicle from a "relatively safe" position to a "very safe" position well away from traffic),
- Any trusted Good Samaritan (by judgment of the adult passengers and/or a contacted remote guardian).
- If no controls, a remote-assistance operator can take over:
- if necessary to the completion of the drive,
- or more typically just long enough to clear the scene of the disengagement, enabling L4 re-engagement.
These points apply both to personally-owned L4 vehicles and to fleet RoboTaxis.
*The probability-of-success threshold can depend to a large degree on the tolerance of the user to a potential disengagement. A RoboTaxi fleet company and its customers presumably have a low tolerance for delays or problems. OTOH a blind or otherwise disabled adult, riding in their own AV, may be quite willing to accept an occasional delay event in exchange for the independence of life enabled by the AV technology.
I think these considerations should serve to broaden the definition of what L4 needs to achieve, and to the quoted point, allows for gradated and more quantifiable rating levels within L4.