3 ms·

The current technology is incredibly dependent upon high quality maps of the area. In theory that's a solvable problem, but it requires breakthroughs in mapping

The current technology is incredibly dependent upon high quality maps of the area. In theory that's a solvable problem, but it requires breakthroughs in mapping (it's prohibitively expensive to map to the degree needed for large areas), breakthroughs in reactions to unpredictable or untrained situations, breakthroughs in sensors that can work in poor climates (rain and snow) and breakthroughs in computing to be able to make it affordable once it's possible to process that. It's not impossible, but it's probably safe to say that it's currently insurmountable by data given current technology and the cost structures involved. Given several large, game-changing breakthroughs, of course it's possible.

lern_too_spel 11y ago
Your perception of current technology is decades behind where current technology is. Current technology can absolutely drive without high quality maps. It drives superhumanly well when it has access to those maps but just fine relative to human baseline without them.
henrikschroder 11y ago
You're moving a bunch of goalposts here. Mapping isn't required at all for self-driving, it's required for self-navigation. Also, the obvious solution to mapping is to let every car provide input as they're driving. What do you think Google is doing with both Google Maps and Waze? They collect a shit-ton of data while people are using those apps and driving. There's your mapping. For untrained or unpredictable situations, you don't need a breakthrough, you just need the steady improvement of sensors and computer vision. You only need to encounter a situation once for the entire fleet of self-driving cars to be trained on it. For computing, you don't need breakthroughs, you just need more, and we're getting more, every year, steadily. Self-driving cars is not a matter of flashes of genius, it's a boring matter of economics of making the sensors cheap enough, the data collection cheap enough, the computing cheap enough, for it to be fitted to every car and there we go.
zardo 11y ago
For mapping, you sensor up the cars you sell, SLAM, and send the data home. I'm not saying its easy, but you don't need to wait for someone to win a Nobel prize, Tesla started doing it last year.

lern_too_spel 11y ago

Your perception of current technology is decades behind where current technology is. Current technology can absolutely drive without high quality maps. It drives superhumanly well when it has access to those maps but just fine relative to human baseline without them.

henrikschroder 11y ago

You're moving a bunch of goalposts here. Mapping isn't required at all for self-driving, it's required for self-navigation. Also, the obvious solution to mapping is to let every car provide input as they're driving. What do you think Google is doing with both Google Maps and Waze? They collect a shit-ton of data while people are using those apps and driving. There's your mapping. For untrained or unpredictable situations, you don't need a breakthrough, you just need the steady improvement of sensors and computer vision. You only need to encounter a situation once for the entire fleet of self-driving cars to be trained on it. For computing, you don't need breakthroughs, you just need more, and we're getting more, every year, steadily. Self-driving cars is not a matter of flashes of genius, it's a boring matter of economics of making the sensors cheap enough, the data collection cheap enough, the computing cheap enough, for it to be fitted to every car and there we go.

zardo 11y ago

For mapping, you sensor up the cars you sell, SLAM, and send the data home. I'm not saying its easy, but you don't need to wait for someone to win a Nobel prize, Tesla started doing it last year.