Whether they call their system "Autopilot" is immaterial. You are the responsibly party as the driver, and this is made clear both at purchase time and during vehicle use.
Whether they call their system "Autopilot" is immaterial. You are the responsibly party as the driver, and this is made clear both at purchase time and during vehicle use.
If you don’t agree, do not engage the safety system. As the operator, you have final authority.
And faults in safety systems can leave their manufacturers liable, just like faults in any other system. The fact that it is a safety system doesn't mean anything changes.
> If you don’t agree, do not engage the safety system. As the operator, you have final authority.
"Don't like, don't use" has never been an acceptable answer in discussions of product safety-- in neither the court of law nor the court of public opinion-- and that's not going to change now.
A recent example is the ignition switch recall that old GM had. It took operator action to trigger the fault, but GM still repaired the faulty cars and paid liability claims.
https://en.wikipedia.org/wiki/General_Motors_ignition_switch...
The fault existing is why they paid claims to the people that the fault killed, and why they did a recall.
Expect Tesla to pay a massive fine as a result of this. Assuming, of course that they survive the lawsuit and the loss of buyers...
People suck at remaining alert and attentive when doing boring shit. So when autopilot was all over the road and screwing up every 30 seconds, it wasn't much of a problem. Now that it can navigate highways for an hour or more without needing a safety critical intervention, we have a problem.
Additionally, I'm not a professional driver, I don't have any autopilot on my car, but I've definitely briefly fell asleep while driving late at night. What do you think rumble strips are for?
Such an extremely low number of (but extremely highly publicized) fatality is not reason enough to say autopilot is a net negative. In fact, the extreme media focus that every fatality during autopilot causes should goad improvement of autopilot safety until it's far better than human drivers.
(And let's not pretend that this autopilot error doesn't also happen with human drivers... The only reason this failure was fatal is because less than two weeks earlier, a human driver crashed into the same barrier--likely due to the same confusion--and destroyed its effectiveness since the highway department was slow to reset it.)
There is no path from Autopilot as it is to full autonomy without doing a major shift in strategy. Google figured out very early on that the incremental approach is unfeasible. Most of the other big car companies at the outset were intending to develop autonomy incrementally as well, but then they thought about it for a few minutes.
As the technology improves, conditions can be relaxed for full self-driving.
The very fact that in this case, the driver had multiple problems with this same spot, and that other drivers also noticed problems in this area, points to an obvious mitigation strategy: flag problem areas geographically for autopilot, develop specific strategies that address the problem, and verify it has been fixed by examining how other Teslas on the road in "ghost" mode (i.e. with the human driving but the software pretending to self-drive) respond to the changes.
Tesla, due to its massive and well-connected fleet, has a lot of tools in the toolbox to address these problems, in some cases more than Google (although I agree Waymo's self-driving capability is currently more impressive).
This is what I have been thinking as well, except it appears that Tesla doesn't currently have a way to handle such scenarios.
Ideally, they should be recording when a driver makes a sharp swerve to correct autopilot's naïve lane marker following algorithm, then use that information to inform future drives using autopilot on that stretch by disabling autopilot as that patch of road is approached, while they work on a reliable fix.
Considering they have enough data to do realtime traffic analysis, they easily should have enough data to determine problem areas for the autopilot.