Emergence of coherent behaviors from homogenous sensorimotor coupling

It seems self-evident that an agent's perception of its surrounding environment is tightly coupled to its behavior. However, most robot control architectures make some assumptions about how sensory information relates to motor actions, in order to provide a set of basic behaviors (such as reflexes). It is largely unknown to what extent these biases reduce the potential for the generation of diverse or unexpected behaviors from the agent-environment interaction. In this paper, we propose a new model of robot control architecture, consisting of homogeneous, non-hierarchical coupling, which only learns the correlation of simultaneous activity between any pair of sensor or motor modalities. We show that the propagation of activity across the different modalities, modulated by the learnt correlations, can lead to the emergence of coherent complex behaviors, such as approaching and following an object, or solving a task based on the temporal relationship between an early clue and a delayed reward