Spiking Neural Network for Behavior Learning of A Mobile Robot

Neural networks, fuzzy controls, and evolutionary computation have been used for behavior learning of robots in unknown and/or dynamic environment [14]. However, it is very di cult to design the learning structure of a robot beforehand, because the dynamics of the environment is unknown. Consequently, the robot should have a high learning capability to deal with the spatio-temporal context of the facing environment. Therefore, this paper proposes a fuzzy spiking neural network (FSNN) [5] for behavior learning of a mobile robot. The Hebbian learning can be used for updating the weights of FSNN. Furthermore, the network topology should be adaptive to the environmental conditions, and so, a steady-state genetic algorithm (SSGA) is applied for deciding the network topology as a combinatorial optimization problem. In this paper, the part of fuzzy was omitted for space.