I'm sorry you feel that way. I indeed watched the entire stream and not once do I recall Musk mentioning sensor fusion.
Instead, he just focussed on the drawbacks that LIDAR had, which can be corrected if feeds from LIDAR and camera were to be combined. This is why every other AV company uses sensor fusion. No one uses LIDAR just by itself and Tesla should not use vision just by itself.
i guess the main question is why humans can do it without lidar ? or the other way put:
why would a lidar bridge the gap between a current vision-based system and human level performance ?
Humans are able to perceive the world due to a bunch of monocular and binocular vision cues (https://en.wikipedia.org/wiki/Depth_perception). Out of these cues, cameras are not able to offer a few. For example, cameras can't offer accommodation as they can't mimic changing focal distances like a human does. Tesla also does not seem to have cameras that focus on different focal lengths as the human eye can.
In addition, even if we can get all the cues, current computer vision techniques are nowhere near as smart to figure out a lot of stuff that the brain can. There is no general AI that can learn from one instance of a test case as the human brain can. Even stuff that is simple to the human brain can trick computer vision (https://globalnews.ca/news/3654164/altered-stop-signs-fool-s...). There is a lot of catching up to do in terms of computer vision.
A LIDAR can help with these cases a lot. It's an extra input that generates a 3D point cloud of the surroundings around it. LIDARs have their own failures but can be combined with computer vision to compensate. This is the central idea behind sensor fusion. Not able to detect if an object is indeed a few feet in front of you due to confusing parallax? LIDAR can help. Looking at a new object that you have no idea what it is cause you have never seen it before? LIDAR can at least tell you how far away it is and it's shape in addition to how fast it is moving.
Now, LIDAR is not a panacea that comes to the rescue of computer vision's failings and solves all of AV's problems but it definitely furthers what a self-driving car can perceive and deliver a higher amount of safety to the end user.
Instead, he just focussed on the drawbacks that LIDAR had, which can be corrected if feeds from LIDAR and camera were to be combined. This is why every other AV company uses sensor fusion. No one uses LIDAR just by itself and Tesla should not use vision just by itself.