Abstract
In order to realize the interface between a neural network and an electronic circuit for neuroprosthetics, it is critical to elucidate the network dynamics of a neural network and to establish the optimal input methods and the decoding methods of the spatio-temporal firing pattern of the neural network. For the purpose, we are developing the neuro-robot as the model system for biological information processing with vital components. The behavior of the neuro-robot is generated by the response pattern of neuronal electrical activity evoked by a current stimulation as an input from outer world. Our developed neurorobot has a living neuronal network (LNN) as an epistatic processor and the robot is designed to generate purposive behavior by the regulation of the relationships between input and output of the LNN. In this study, we developed a novel type of neuro-robot with Self-Organization Map (SOM) algorithm, which enable the robot to perform non-stop learning and generation of behavior simultaneously. The 64 dimension feature vectors of the spatiotemporal electrical pattern evoked by the inputs according to the value of the IR sensors on the robot body are inputted to the SOM. The 64 dimension feature vectors are mapped to the 10 x 10 output layer of SOM. Only at the early stage of the learning, SOM selects two winner nodes premisely assigned to specific inputs for the obstacles near the L and R side of the robot body. Thus the process is teacher learning. Winner nodes are linked to the purposive behaviors adequate to the inputs. We call the process as "seeding". By seeding procedure, the distribution of winner units for the two inputs were separated each other, in the case of the spatiotemporal pattern of responses evoked by 2 inputs were not overlapped. In addition, the location of the center of gravity among winner nodes for repeated inputs are approximately same in the output layer of the SOM.
