Pretty interesting change
going from 3 to 2 front cameras. Before they had long range narrow camera for detecting vehicles far away, the wide camera to detect vehicles near with a wide angle and the main forward camera for middle range improvements. My guess is that Tesla decided that with their new autolabeled 4D dataset the forward wide camera is good enough at the middle range also.
It might be hard to get good labels for the wide forward camera in the image space, but in 4D space it will be a lot easier to generate the labels. Then the question is if the neural network can see good enough in the images to predict the correct labels. With photon count super human detection will be possible, this likely extends the range of the camera somewhat.
Also by increasing the resolution(which has been rumoured) they will effectively increase the range.
Basically the main forward and the radar is elimated and the wide forward will have some extended range thanks to software. The loss will be the blue area outside of the grey area.
Or in image space, the "in between" image quality gain will be sacrificed and it will look the the image quality around it:
(for more examples and source, see
this thread)
I am not sure, but I think the narrow long range camera will also get a wider field of view. This will negatively impact the long range performance, but increasing the resolution will improve it. Thus the image in the center will expand somewhat. I think Tesla has figured out a good tradeoff here, they know how many hundred meters forward they need to be very accurate and would not do this change if it risked introducing significant risks.
Also with memory the neural network will have remembered what it saw in the high zoom middle image as it moves forward, I guess that will help a lot. So basically using memory to zoom out from the narrow forward camera.
What they are left with is one less sensor that can fail, less power consumption, less data to process and one less part. The best part is no part.
In hindsight this change makes a lot of sense.