> when musk says this he gives data to back it up - that humans manage just fine with visual systems
What humans do 'just fine' isn't very relevant to self-driving tech. Humans balance just fine as well and yet the state of the art for robots is quite primitive.
Airplanes also don't fly like birds.
> lidar is really expensive
Which, again, is somewhat separate to whether it is effective.
>Humans balance just fine as well and yet the state of the art for robots is quite primitive.
That's the whole point. Full autonomy will never truly happen until computers can do what human brains can with regards to driving. No amount of sensor technology will make up for that fact. It's a software problem, not a hardware problem.
Sure. But that doesn't mean they have to do it the same way humans do, (same as airplanes don't emulate birds) - that is you can probably arrive at a similar outcome via multiple approaches, some quite different to how the human brain works.
Last I checked it cost more than a $100,000 (on average, in the US) to raise a human to the point where they have sufficient control over their behavior and body that they could be trusted to safely drive a several ton vehicle.
Humans have the biology for visual systems, and constant training, and it still takes us almost a decade to fully process distance properly. (This is why children will run in front of moving vehicles--they can't actually tell how far away they vehicles are.)