PloS Computational Biology

How much brain power do you need to control your body? It’s a key question for robotics and artificial intelligence researchers, and the answer is, well, not all that much. The key, according to new research from SFI Professor Nihat Ay and colleagues, is that bodies have to obey certain rules — the laws of physics and biomechanics, for example — that constrain the control problem and make it easier to solve.

To arrive at their results, Ay and fellow researchers first set mathematical bounds on exactly how complex a brain must be in an abstract environment. But to test their ideas in a (somewhat) more real world, the researchers used computer simulations of an insect-like robot trying to walk with a tripod gait -- that is, moving a leg on one side and two on the other at once.

Despite the complexity of walking in a particular way with six legs, it took a brain with the equivalent of just 65 neurons to tackle that goal. The researchers estimate that the traditional approach to programming, which does not capitalize on the mechanics of the brain-body system, would have required a whopping 1014 neurons to effect the same walk.

This successful prediction of the minimal brain size required for a given movement is a first in the field of artificial intelligence and may contribute to the future design of intelligent embodied agents (a.k.a. robots).

Read the paper in PLoS Computational Biology (September 1, 2015)

Read the SFI working paper (December 2014)

Watch the agent's walking behavior at different brain sizes

Read more about SFI research on embodied intelligence