It’s funny, I see people online in random comments saying “AI” hasn’t materialized in to anything useful. But if you realize that most talk of “AI” is talk of ML, and if you pay attention to robotics, you’ll find it’s making a huge difference. The fact that anyone can credibly say perception is not the major hurdle anymore is a testament to what the current generation of machine learning algorithms have done to advance the state of the art. I’m excited for the time when even prediction feels like a solved problem.
I would think that perception in poor weather remains a real challenge, but certainly we’re leaps and bounds ahead of where we were even 5 years ago.
It's categorically untrue that perception is solved, given that things like this happen: https://twitter.com/greentheonly/status/1228067903666348045. And that's considering that Tesla's perception team is run by one of the world's most respected experts in the field, and that their access to data is essentially unlimited.
Now what bothers me is that leaders in the industry say without batting an eyelid that perception is solved, given that they witness this type of events on a regular basis.
Elon Musk recently claimed „correct vector-space representation“ is the hardest task of autonomous driving. If true or not for Tesla i dont know but i guess its a yet to be solved problem for everyone in the field. Would you agree?
We have cars that can drive themselves at speed under normal conditions. That's pretty huge re: "useful". Extending satisfactory performance under fringe conditions is just a matter of improvement.
I would think that perception in poor weather remains a real challenge, but certainly we’re leaps and bounds ahead of where we were even 5 years ago.