Well, Lewandowsky heads Google's self driving car project now. Thrun is long gone, off trying to make Udacity fly.
He's an impressive guy. I met him when he was going to Berkeley and doing the self-balancing motorcycle for the 2005 Grand Challenge. He'd already done a startup, with a specialized giant laptop for construction sites for people who needed to see blueprints.
The LIDAR Google uses is from Velodyne, which was "Team DAD" in the 2005 Grand Challenge. The first version of their LIDAR fell off their vehicle, but they improved the mechanics and produced that cone-shaped thing Google now uses. That's really a research tool; a different approach is needed for production vehicles. (I still like the Advanced Scientific Concepts flash LIDAR; it's expensive, but that's because it has custom silicon. If you had to get the price down, that's where to start. No moving parts, all electronics.)
I'm kind of disappointed with Google's self-driving effort on the hardware side. I'd expected flash LIDARs, terahertz phased array radars, and other advanced sensors by now. You need to be able to see in all directions, but the requirements to the sides and back are less than for looking ahead. The CMU/Cadillac effort is ahead on the hardware side; their self-driving car has all its sensors integrated into the vehicle so you don't notice them.
(I had an entry in the 2005 Grand Challenge: Team Overbot. Ours was too slow, and we worried about off-road capability too much.)
I suspect price is still the driver (pun intended) here. Flash LIDAR is, for the most part, shockingly expensive. Some companies like Toyota are looking into bringing the cost down with CMOS SPAD arrays, but it's still very much lab work. Similarly the Velodyne system ain't cheap and was close to $80k not too long ago. Stereo is pretty much free, but still needs more work to make it robust enough. Even smaller line scan LIDAR units (e.g. the SICK LMS models) are around $5k each.
He's an impressive guy. I met him when he was going to Berkeley and doing the self-balancing motorcycle for the 2005 Grand Challenge. He'd already done a startup, with a specialized giant laptop for construction sites for people who needed to see blueprints.
The LIDAR Google uses is from Velodyne, which was "Team DAD" in the 2005 Grand Challenge. The first version of their LIDAR fell off their vehicle, but they improved the mechanics and produced that cone-shaped thing Google now uses. That's really a research tool; a different approach is needed for production vehicles. (I still like the Advanced Scientific Concepts flash LIDAR; it's expensive, but that's because it has custom silicon. If you had to get the price down, that's where to start. No moving parts, all electronics.)
I'm kind of disappointed with Google's self-driving effort on the hardware side. I'd expected flash LIDARs, terahertz phased array radars, and other advanced sensors by now. You need to be able to see in all directions, but the requirements to the sides and back are less than for looking ahead. The CMU/Cadillac effort is ahead on the hardware side; their self-driving car has all its sensors integrated into the vehicle so you don't notice them.
(I had an entry in the 2005 Grand Challenge: Team Overbot. Ours was too slow, and we worried about off-road capability too much.)