2daMoon
Mostly Harmless
The "current rate of progress" doesn't tell us exactly when it will exceed human performance. If you relied on the rate of progress from 2020-2023 then you would conclude it wouldn't reach robotaxi level for decades. Sounds silly right? The fact that miles / critical disengagements went from 100 to 400 in the last 6 months doesn't mean it's going to hit 30,000 miles per disengagement in 1 or 2 years. That slope will change unpredictably. I'd bet it gets steeper first, then slows down. But when, who knows?
How do they determine that the human didn't make a mistake by disengaging? What if the vehicle would have handled the situation if the supervisor didn't intervene?
That uncertainty of when is why TSLA is becoming a risky investment. If robotaxis are planned for production in 2 years, we are basically requiring progress to reach a high standard in a short window. Otherwise the produced cars are much less valuable.
Since Tesla doesn't provide disengagement data, the FSD Tracker critical disengagements is the best we have.
Because of the human element in disengagement any measurement of this will not be black and white.
The goal of Robotaxis is to eliminate the human element from the equation in order to improve safety. It would be unproductive to use data based upon frequency of human intervention as a measure of progress.
Unfortunately, it does seem that the best way to measure for whether or not a Robotaxi is "safe" will be from the same statistical methods that have traditionally been used. Actual cases which result in measurable damage to person or property. Which is the standard currently used to evaluate humans, is it not?
The uncertainty is not reduced simply because a human is involved.
The folks solving this at Tesla know this and certainly do rely upon the feedback from the supervisors of FSD in order to make improvements. However, critical disengagements are not a factor that can be used to determine whether or not FSD can operate a vehicle safely.
If that were the case, how many of our spouses should lose their driving privileges as a result of human intervention being voiced in any situation where they got through it fine by ignoring the intervening spouse entirely?