Learning the motion map of a robot arm with neural networks

Abstract

The integration of neural self-organization and circular reaction into a robot guidance system (Neurobot) are discussed. A two-degree-of-freedom robot arm learns a cognitive map which contains both the visual workspace and also the robot's joint angles in a biologically inspired neural network. The range of positions of the robot's end effector is called the workspace, and the corresponding joint angle space is called the configuration space. In other words, the workspace is the physical world of the robot, whereas the configuration space is an abstract space necessary for controlling the arm motions. Neurobot creates an association between its visual position and its joint position by training a self-organizing neural network using both spaces as inputs