Unsupervised learning of target attraction for robots through Spike Timing Dependent Plasticity

In this paper, unsupervised learning of robots through Spiking Neural Network (SNN) is studied. Izhikevich model is used for building spiking neural network and kinematic model of the robots is considered for body modeling. Learning mechanism is based on Spike Timing Dependent Plasticity (STDP) in which variation of synapses between neurons depends on spike timing. The reciprocal architecture of network as well as timing regulation of algorithm result in tuning of spiking neural network which makes robot to be attracted to the target. It is supposed that the robot has two sensors for target detection. Signals from sensors are conducted into neural network, so that the output of network (spikes of motor neurons) derive the actuators. Neurons that fire together make special pathways from sensory neurons to motor neurons, and other synapses which are not involved, are weakened slowly. The algorithm of proposed method is illustrated and we designed two sets of simulation studies in order to test the effectiveness of the proposed method: the learning of target attraction for 1) mobile robot, 2) one-DOF robotic arm.