Non-learning artificial neural network approach to real-time motion planning for the Pioneer robot.

摘要

The real-time motion planning of mobile robotics is a stumbling block for the expansion of mobile robotics into more complex fields. The solution of this problem is the focus of this thesis. This thesis builds upon the work of Dr. Simon Yang and Prof. Max Meng, which implements a biologically-inspired non-learning artificial neural network (ANN) This ANN computes a motion path for the Pioneer 2 DX mobile robot under the Saphira operating system. It also independently confirms the findings in the earlier work. This method is a variation on the approximate cell decomposition method where neurons represent free space or regions occupied by obstacles. Each neuron in the neural network is characterized by an additive or shunting equation that models the interaction of obstacle neurons, free space neurons and the neuron representing the target pose. This method is able to find a path if one exists from any arbitrary initial and final pose.