I've been working with pretty much top tier vision tech, and it's still so far off from being viable enough. It's insane how well it works, but to use it in a dynamic environment driving around on roads... Pure madness to believe it will be enough with only cameras in like the next 10-15 years at least (imo).

I bought a Tesla despite my hatred for Musk because it was the only reasonable alternative when it came to electric cars where I live at the time, and got one of the old ones with lidar and cameras. I have this super weak shimmer of hope that they will go back to using both again, since the warning systems and their "autopilot" feel way shittier than it used to since they've allegedly patched away the lidar.

fortunately LIDAR unit costs are going down, so multiple units, fusing their data with regular camera arrays should resolve a very good view, and be good at error-correcting for each other's shortcomings.