An evolved Spiking Neural Network (SNN) can act as a brain, controlling and navigating a robot through a maze. In this work, a SNN is used to control a robot based on encoded inputs from sensors obtaining information about robot's environment. The robot's goal is to pursue the brightest light in a maze using light sensors, while avoiding obstacles using infrared sensors. In this control model the tunable parameters are the synapse delays and weights from the SNN, and are evolved using a genetic algorithm (Eshelman's CHC algorithm). Training is accomplished using input and output values obtained from the robot under control of heuristic rules programmed with the same goal. This work extends the GA-SNN paradigm pioneered by H. Sichtig (2009) by linking the simulator to the robot hardware. A potential application of the technology herein developed is control of smart prosthetics.
展开▼
