Distributed action selection by a brainstem neural substrate: an embodied evaluation

Theoretical approaches to the problem of action selection in autonomous agents often contrast centralised and distributed selection schemes. Here we describe a neural substrate for distributed action selection in the vertebrate brain-stem, the medial reticular formation (mRF), which may form a evolutionary precursor to centralised schemes found in the higher brain. We evaluate its competence as a selection device for robot control in a simulated resource co-ordination task, and use a genetic algorithm to evolve the mRF's inputs and internal structure. Some configurations of the mRF could sufficiently co-ordinate actions to maximise the robot's energy, but this is critically dependent on a high rate of energy acquisition, which leaves an animal (or agent) susceptible to food shortages. Thus, the inflexibility of the mRF as a distributed selection mechanism may have provided impetus for the evolution of more complex, centralised, selection mechanisms in the brain.