Neural network approach to control wall-following robot navigation

In any robotics application, the deployed robot has to navigate from source to a destination for performing task(s). Efficient control of this navigation is a major research challenge in the field. In this paper, an attempt has been made to develop a neural network (NN) based controller for navigation of wall following robot. The primary focus is to control the robot to take decision of changing direction based on a set of sensor readings, where the sensors are fit around of the waist of the robot (SCITOS G5 robot in this work). The NN is trained by these sensor readings dataset (a collection of multiple such instances) and predicts the future control strategy. The NN is trained with gradient descent algorithm. An extensive parametric study has been conducted to set the optimal number of nodes in the hidden layer and the learning rate. The experimental result shows that the proposed algorithm can control the robot with 92.67% accuracy and can take decision within 1 second.