depth first search

True or False: acting is more difficult than perceiving.

UPDATE: Again with the comically vague questions — that’s sort of how my mind works (or starts working). Anyway, to get a bit more specific, imagine you need to teach a humanoid robot 1) how to move and act in the world and 2) how to understand what it sees. You’d want your robot to do both these things at something approaching the abilities of an adult human. These are both very active research questions, but are often explored in much narrower forms (e.g. bipedal walking or object recognition). I think it is safe to say that the state of the art in both areas is still far from the capabilities of an average adult human. But we can speculate on which question will be answered first.

I used to think computer vision was the harder problem and that robot locomotion and action in humanoid robots would approach the capabilities of an adult human much faster than any form of visual understanding by computers. Now I’m not so sure.

UPDATE UPDATE: Another, different way to think about this question is, suppose we had a humanoid robot that successfully learns both 1 and 2. Do we expect the learning process for 1 to be more sample efficient or require less time than 2, or would 2 be more complex a learning problem than 1?

•