Dutch regulator approves Tesla’s FSD driver-assistance system but won’t release safety-test details
Tesla has pressured officials to withhold safety-testing results from the public
Dutch regulator approves Tesla’s FSD driver-assistance system but won’t release safety-test details
Tesla has pressured officials to withhold safety-testing results from the public
So if the self-driving machine learning algorithm learns by collecting data from human drivers, how could it ever be better than the human drivers it learns from? Especially knowing that these models are nondeterministic and often generate unpredictable and wildly incorrect output.
Not that I trust these systems too much yet, I do see why they can be better:
But as you say, the AI is still putting out weird things once in a while, and maybe the next big painting on the side of a truck makes the AI drive over into incoming traffic…
Also it could monitor the exact speed, acceleration and trajectory of all the other vehicles around it, and (ideally) know exactly how to handle the car to react accordingly.
Yes, this is the only way I see self driving working out personally. It needs to be a general, brand agnostic rollout and all vehicles on the road can talk to each other and function in unison. They would still need all the radar/camera tech as well obviously to account for pedestrians, animals, stuff on the road etc but with a global network I can see it be better than humans.
It’s not purely machine learning. Much of it is manually programmed in, like an adaptive cruise control.