Musk's live broadcast started from Tesla's headquarters.
To let everyone know that this is not a "pre-planned route", Musk randomly selected a destination on the map - Stanford.
Soon, the car drove into a construction area, and the vehicle passed smoothly without panic
After driving smoothly for a while, the vehicle hits a speed bump, and even if a bicycle is speeding on the right side, Model S will not be affected in the slightest, and will still slow down.
Musk said: "It may not be clear from the live broadcast, but the car actually drives very smoothly. ”
Subsequently, the vehicle carried Musk to the first destination, Stanford.
As you can see from the picture, Model S knows how to give way to pedestrians when there are many people.
He further stated, "Similarly, there is not a single code that says to stop or wait for other vehicles in front of a stop sign, or wait X seconds or something like that." ”
As Musk said before: "The new FSD V12 version is an end-to-end output!" In other words, FSD V12 no longer relies on traditional HD maps and navigation data, abandons the traditional "if else" rule-driven triggering method, and uses AI algorithms to identify road and traffic situations through on-board cameras and neural networks.
To put it bluntly, FSD just watched a lot of videos, then completed the training and got a neural network.
But the problem is that putting AI in charge of driving may have weaknesses in some scenarios. Case in point: Self-driving cars must stop at an uncontrolled railroad crossing and wait for an incoming train. If FSD V12 does not correctly and promptly determine what happened, it could theoretically expose passengers to serious accidents.
This shortcoming was also exposed in this test drive of FSD V12, that is, the "running a red light" mistake. However, Musk himself reflected, saying that "to solve this mistake, you need to feed more traffic light videos to let it learn." ”
