Teslavision uses confocal diffuse tomography. Humans don't.There’s a fundamental mistake being made by Tesla with this vision only BS - one minute they keep saying humans are rubbish at driving and we need FSD to save mankind, then the next minute they say oh but we need to make FSD work in the same way that humans do - using only eyes.
Just duh.
The very low-res radar had it's own problems, such as getting reflections from bridges, losing track of objects etc. Throw in the fact that is was a 3rd party sensor and therefore outwith Tesla's control and you have a pretty poor input source. Sure there are some things that it is better at than simple vision, but if you have two sensors and they disagree, which one do you believe?