Robots can already do amazing things but dealing with visually complex structures of varying texture is extremely challenging. For example, it is very easy for a human to reach into a pile of nuts and bolts and specifically select just the bolts. This seems simple, but teaching a computer how to distinguish a bolt that may be positioned at any angle, while surrounded by a visually cluttered environment of other nuts and bolts, is quite complex.
Here a robot is demonstrating its mastery of a truly complex visual and tactile task. Consider combining this ability with other robots, such as the one that can be seen at this link.
What is even more interesting is that the programming is actually not that complex. The research paper can be found by clicking here. It is based on a fairly simple flowchart, as seen below:
