Of course they did. There's only so much you can do with "feed more data in the machine learning pot, stir it, and see what comes out." We're doing a good job of proving what the limits of ML based self driving are...
Computers aren't human, and as it turns out, human brains and visual processing systems are more than the "couple crappy cameras and a bit of a neural network" thing that a lot of SV types seem to believe it is.
We're firmly into the "fusion" realm of self driving cars. A lot of money and a lot of time, effort, developers, etc, have been fed into the maw, and we've tried all the easy options - which it's clear don't work. I'll take the pessimal position of saying that the current approach Tesla is using (vision only) can't work, though I'm sure many will disagree with me. It comes down to a difference of opinion about the complexity and capabilities of the human brain.
A sensor-heavy setup like Waymo is using has the greatest chance of working, though probably only in heavily-mapped areas where it can use the sort of peripheral data humans use to identify the lanes by building locations and such when the lanes are unclear. As long as not much changes, which probably requires very regular remapping of regions (or an awful lot of data upload from cars), that's the most feasible approach, though it's not a general purpose solution given the amount of data required to do high precision mapping of the entire country, much less the world.
We'll see, but "a couple cameras and a car-scale computer" just doesn't seem a solution that's going to work.
Computers aren't human, and as it turns out, human brains and visual processing systems are more than the "couple crappy cameras and a bit of a neural network" thing that a lot of SV types seem to believe it is.
We're firmly into the "fusion" realm of self driving cars. A lot of money and a lot of time, effort, developers, etc, have been fed into the maw, and we've tried all the easy options - which it's clear don't work. I'll take the pessimal position of saying that the current approach Tesla is using (vision only) can't work, though I'm sure many will disagree with me. It comes down to a difference of opinion about the complexity and capabilities of the human brain.
A sensor-heavy setup like Waymo is using has the greatest chance of working, though probably only in heavily-mapped areas where it can use the sort of peripheral data humans use to identify the lanes by building locations and such when the lanes are unclear. As long as not much changes, which probably requires very regular remapping of regions (or an awful lot of data upload from cars), that's the most feasible approach, though it's not a general purpose solution given the amount of data required to do high precision mapping of the entire country, much less the world.
We'll see, but "a couple cameras and a car-scale computer" just doesn't seem a solution that's going to work.