Adaptive behaviour control by self-regulating neurons

A local learning rule for recurrent neural networks is derived by introducing neurons as self-regulating units. Acting as controllers for the behaviour of autonomous mobile agents, neural systems endowed with this learning rule generate interesting agent-environment interactions, which depend on the underlying connectivity of the network. The dynamical phenomenology of a single neuron and a two neuron network is analysed, and, as an example, the generation of obstacle avoidance behaviour of a miniature Khepera robot is presented.