Synaptic plasticity for recurrent neural networks is derived by introducing neurons as self-regulating units. These neurons have home-ostatic properties for certain parameter domains. Depending on its underlying connectivity a neurocontroller endowed with the derived synaptic plasticity rule can generate a variety of different behaviors. The structure of these networks can be developed by evolutionary techniques. For demonstration, examples are given generating a walking behavior for a 3-joint single leg of a walking machine.
展开▼