Object recognition is quite a tough problem. It’s easy to trivialize it since humans are great at detecting objects and patterns, but it is difficult to write an algorithm that detects all cases of how an object appears at different distances, angles, and if it’s partially obscured. Just look at face detection. It’s easy for a human to tell statues from humans, but the techniques for a computer to do so are easily fooled, and probably not worthwhile to train against. Likewise, object character recognition can fail by simply using a different typeface that the computer hasn’t been trained on. Likely the maps google generates reduces a lot of overhead the car would need to compute on the fly otherwise, for example static vs moving hazards.
Yeah, I’m going with ‘potential hurdle to implementation’ != FATAL FLAW THAT WILL DESTROY US ALL (AND GOOGLE DOESN’T WANT YOU TO KNOW), as implied. Tone, people.
I was thinking the same thing. The upward progression of course difficulty and number of finishers in the DARPA challenge series for self-driving vehicles seems to speak to this.
We still take a lot of factors affecting object recognition for granted. For example, it’s a trivial matter for a human to look at a scene like this and understand that they aren’t actually looking at a stop sign. Your iPhone? Not so much.
3 Likes
If it leads to ban on use of such reflexive-attention-getting important symbols in outdoor advertising, it cannot be bad.
2 Likes
then again, we’re pretty bad at recognizing this sign:
11101010010101100110101010011010111101010101010110101000010101010101010101
2 Likes
Can I do any better? Of course, every day:
I drive for my work, a lot. And it’s on different kinds of roads, different kinds of weather, different conditions. And I’ve driven much more than 700,000 miles.
And with only a few accidents, less than half a dozen - which I all blame on the weather or on the other drivers.
The engineers can do that too, themselves. The question as intended is, can you make a better self-driving car than they got? 
On a side note, a friend once did some calculation for a busy intersection, some years ago (some semi-drunk napkin drafting it was), the needed spacing and positioning of cars driving at certain fixed speed, so there can be intersecting flows of cars from two or more directions without the flows having to stop at the ligths - the cars in the other directions were zooming through the spaces between the perpendicular flow of cars. You cannot do this with a human driver. By getting humans out of the control loop, a whole new universe of possibilities opens up.
And also closed off from deer and other large critters (often with thumbs directed outward) that occasionally wander the highways?
I just had to thought about Little Chicago, the model of the city in your friendly wizard’s basement.
It doesn’t matter since they’re both worth the same number of points.
1 Like
the real dirty secret is that they all have oompa loompa’s hiding in the trunks operating the vehicles and they aren’t self driving at all. true fact. google it.
1 Like
Mechanical Turk cars?
4 Likes
OMG!! Why can’t they just do this with magic? It’s totally simple!
This just in: duh.
The only other large public self driving car testing that’s happening right now is in Sweden being run by volvo. There’s a handful of cars working right now and they plan to have 100 autonomous vehicles on the road by 2017. It will come as a surprise to no one that they’ve got the same issue, but their workaround is to work closely with the local council to share the costs of upgrading infrastructure to have standardised and predictable markings and signs.
I did. And I couldn’t find anything out about it…
… which is exactly what I’d do if I had control of all of the information.
Mind=blown.
3 Likes
Information systems can assist drivers with situational awareness that they do not have now. A few months ago, a woman drove over a pile of leaves on a street near here. She could not have known that a small child was hiding in those leaves. A system that watched the streets for infrared sources like children or pets, or a ball that rolled into the street before a driver turned the corner, could see and remember their movements, and inform drivers with a heads-up display, The same display could help drivers estimate stopping distance on an icy street, etc. Use machines for what they are good at, people ditto, and migrate towards more automation as it becomes practical. We should concentrate on removing hazards, not drivers.
This topic was automatically closed after 5 days. New replies are no longer allowed.
