4 ms·

Most of the projects gunning for L4 have sensor arrays comparable to Google's, with the exception of Uber, who has 3× more: 20 cameras and 7 lidars amongst all

by Fricken 10y ago

Most of the projects gunning for L4 have sensor arrays comparable to Google's, with the exception of Uber, who has 3× more: 20 cameras and 7 lidars amongst all the rest. I'm pretty sure Uber's in it to develop a commercial product, although the PR probably doesn't hurt either. Thrun guesses that Lidar will ultimately be unnessecary, and I think that by the time the hard AI problems are solved, he'll be proven right. Humans navigate with a pair of eyeballs and not much else, after all. But while extraneous sensors can always be removed, not having enough could hamper progress.

sangnoir 10y ago
> Humans navigate with a pair of eyeballs and not much else, after all Only if by "not much else", you are referring to a ridiculously performant image processor - the visual cortex does an amazing job! My guess is it will be many decades before we can get similar performance in hardware/software.
- Fricken 10y ago
  We have ridiculously performant image processors. The dynamic range and light sensitivity of top shelf CMOS sensors continues to improve, and consumer tech driven image processing software which can extract useful data from all sorts of lighing conditions has made leaps and bounds. Behind that is computer vision, revolutionized by deep learning. It can identify street signs, makes and models of vehicles and other road objects, it can keep a car centred in a lane when there's limited information such as snowy conditions, or with chaotic visual information such as a sun dappled country road covered in leaves. Google has patented a method for interpreting the hand signals of police officers. They can read cyclist signals, and interpret the body language of pedestrians to interpret whether they intend to cross the road or not to avoid false braking incidents. And Nvidia has demoed a way to build 3d point cloud fields (like Lidar) using off the shelf cameras. Robust computer vision is heavy on compute, but they're doing it. But behind all this is the higher level reasoning needed to deal with tens of thousands of edge cases- this is the hard part. Google made a special project out of programming their cars to effectively assert themselves at four-way stop signs, which is one problem amongst many. No single edge case is unsolveable, but taking them all on, and getting the whole rube goldberg machine to six sigma reliability is an epic slog. This is the part that's going to take another 10 or 20 years, though we'll probably see L4 applied in a constrained capacity fairly soon.
  - sangnoir 10y ago
    > We have ridiculously performant image processors. The dynamic range and light sensitivity of top shelf CMOS sensors continues to improve, and consumer tech driven image processing software which can extract useful data from all sorts of lighing conditions has made leaps and bounds. That may be true, but we are still ways from catching up to human performance on things like "where is the edge of this dirt road"

by Fricken 10y ago

sangnoir 10y ago

> Humans navigate with a pair of eyeballs and not much else, after all Only if by "not much else", you are referring to a ridiculously performant image processor - the visual cortex does an amazing job! My guess is it will be many decades before we can get similar performance in hardware/software.

Fricken 10y ago
We have ridiculously performant image processors. The dynamic range and light sensitivity of top shelf CMOS sensors continues to improve, and consumer tech driven image processing software which can extract useful data from all sorts of lighing conditions has made leaps and bounds. Behind that is computer vision, revolutionized by deep learning. It can identify street signs, makes and models of vehicles and other road objects, it can keep a car centred in a lane when there's limited information such as snowy conditions, or with chaotic visual information such as a sun dappled country road covered in leaves. Google has patented a method for interpreting the hand signals of police officers. They can read cyclist signals, and interpret the body language of pedestrians to interpret whether they intend to cross the road or not to avoid false braking incidents. And Nvidia has demoed a way to build 3d point cloud fields (like Lidar) using off the shelf cameras. Robust computer vision is heavy on compute, but they're doing it. But behind all this is the higher level reasoning needed to deal with tens of thousands of edge cases- this is the hard part. Google made a special project out of programming their cars to effectively assert themselves at four-way stop signs, which is one problem amongst many. No single edge case is unsolveable, but taking them all on, and getting the whole rube goldberg machine to six sigma reliability is an epic slog. This is the part that's going to take another 10 or 20 years, though we'll probably see L4 applied in a constrained capacity fairly soon.
- sangnoir 10y ago
  > We have ridiculously performant image processors. The dynamic range and light sensitivity of top shelf CMOS sensors continues to improve, and consumer tech driven image processing software which can extract useful data from all sorts of lighing conditions has made leaps and bounds. That may be true, but we are still ways from catching up to human performance on things like "where is the edge of this dirt road"