Who could've ever predicted this scenario? A great example of something that's trivial for human beings to understand but where ML training sets will come up short.
We discount how much the concept of “understanding” is required in visual perception. We don’t just see shapes we have a complete understanding of what we’re seeing.
I might see a big rectangle flopping in the lane in front of me. I can immediately ascertain whether it’s a piece of tumbling plywood, or foam based on movement characteristics, color, apparently size, etc. I can then use that understanding to decide what evasive actions are required.
A Tesla it seems has absolutely zero of this capability.
It was a tow truck towing a car backwards. It was just enough in my half-asleep state to scare the shit out of me.
Humans have millions of years of evolution behind our visual processing systems. We have developed hacks that prevent our brain from getting tricked by unusual situations like traffic lights on trucks or backwards cars being towed. We only developed the first computers a hundred years ago, and only in the last 40 years have a small subset of people started learning about visual processing systems.
It's easy to look at this video and scoff because of how trivial it seems. But it's instead a marvel of our minds that we can pick up on context clues so quickly and accurately that such oddities basically never puzzle us. Given the pace of our innovation, it wouldn't surprise me if our computer systems match ours within a few human generations at latest.
And you're sure that it was a bump, and not the driver pulling up rather close behind it and then giving a sharp tap of the brakes to wake you up? ;)
There are some funny(?) youtube videos of people in the passenger seat waking up to this with tractor trailers being towed in reverse.
This is not a FLOPS problem. Moore's law can't save you here.