Neural Network based Guide Robot Navigation: An Evolutionary Approach

Intelligent robot navigation in urban environments is still a challenge. In this paper we test if it is possible to train neural networks to control the robot to reach the target location in urban dynamic environments. The robot has to rely on GPS and compass sensor to navigate from the starting point to the goal location in an environment with moving obstacles. We compare the performance of three neural architectures in different environments settings. The results show that neural controller with separated hidden neurons has a fast response to sensory input. The performance of evolved neural controllers is also tested in real robot navigation. In addition to the neural network based navigation, the robot has also to switch between other navigation algorithms to reach the target location in the university campus.