Using Backpropagation Neural Network in Object Recognition for Hybrid Mechanism Mobile Robot

Navigation of mobile robots in environments with irregular terrain is a challenging task for scientists and engineers because it involves 3D environment recognition and high order dynamics of the mobile robotic systems. One way to solve this problem is to use hybrid locomotion mechanisms. In this paper, we aim to establish a novel frame work for hybrid local planner which combines motion primitives with artificial neural networks for navigation in rough terrain. This artificial neural network decision engine will find the optimal modes of locomotion for the Hybrid Mechanism Mobile Robot. The input to the neural network is the measurements of the obstruction and the output is the configurations suitable for surmounting an obstacle. Once the robot discovers a change in the terrain (such as high step, low step, etc.), number of measurements will be taken using a feature extraction method to decide the locomotion mode suitable for a particular terrain. Several measurements of the obstruction are considered, such as height, area, and depth of each surface in the scene. These are computed from the 3D representation of the environment built using the on-board sensors, stereo cameras, and a 3D laser range finder. These measurements are fed into a backpropagation neural network in order to choose the successful candidate robot configuration.