KLam
Member
You see, there are infinite number of users. Each have a different preferences on Speed on certain roads, different acceleration preferences, different slow down profile preferences, different road conditions, position of stop signs relative to the stop lines, different weather conditions, etc. It is quite difficult to settle on a one size fit all parameters.I’d be fine with this if it allowed use of the original method of speed control. But the old mode no longer exists.
But seems it is a limitation of the NNs and they CAN’T have a system that just ramps up to the set speed reliably, as it did before, when it has to take into account many other factors as input to the neural network. But right now the NN seems to completely ignore any inputs from driver requested speed above a certain level when using the old mode (which is now a new mode because it is not the same as the old mode).
That being said it seems to be consistently below the limit (perhaps a safety guardrail as you say), so surely they can fix that with training.
But it may be that this borking is going to be with us for the long term. We’ll see if they can figure it out. It could be a more difficult problem than autowipers!
It seems tricky to train a neural network on what speed it should travel. And apparently they can’t just tell it the speed (otherwise the old mode would work). So the days of the car traveling at the requested speed may be behind us.
I have no idea what Tesla is doing or going to do. But here with one input of just the accelerator, Tesla can collect a lot of information on the general public preferences. Also, the users are very familiar to setting the cruise speed with the accelerator.