As i said above, this is a hybrid "BEAM Robot" simulator (Biology, Electronics, Aesthetics, and Mechanics). If you watched it in slow motion it would probably become clearer what its actually doing is completely random and then forming a pattern.
A BEAM robot's "intelligence" comes from trial and error from very simple logic with no CPU involved, just simple oscilating logic from lots of sensors. In reality it has no intelligence. However, the Neural Net part just looks and learns (in a manner similar to the way a real brain works).
The COOL part is that the data from the robots mistakes in a certain situation can be used to build a neural network, so the next time the robot encounters a similar situation it could take over full control and try to do it itself. If it fails, rather than just standing there totally confused, it allows the BEAM to take over to figure the rest of the task out. (and so expanding its own intelligence and logic paths all the time)
The impressive thing about the article is that's exactly what they are doing:
The bbc reporter probably completely missed the point and just filmed some flopping about robots for 5 minutes.
A BEAM robot's "intelligence" comes from trial and error from very simple logic with no CPU involved, just simple oscilating logic from lots of sensors. In reality it has no intelligence. However, the Neural Net part just looks and learns (in a manner similar to the way a real brain works).
The COOL part is that the data from the robots mistakes in a certain situation can be used to build a neural network, so the next time the robot encounters a similar situation it could take over full control and try to do it itself. If it fails, rather than just standing there totally confused, it allows the BEAM to take over to figure the rest of the task out. (and so expanding its own intelligence and logic paths all the time)
The impressive thing about the article is that's exactly what they are doing:
For now, the network learns behaviours such as how to stand up, but promptly forgets them. Der and his colleagues are working to create a long-term memory, so that when the robot finds itself in similar situations, it knows what to do.
. A VR model is easier and faster to build and experiment with than a real one, simulations can be run and learned from in a fraction of the time. (Cheaper too :)The bbc reporter probably completely missed the point and just filmed some flopping about robots for 5 minutes.