An efficiently trainable neural network based vision-guided robot arm

A robotic system using simple visual processing and controlled by neural networks is constructed and tested. The robot performs docking and target reaching without prior geometric calibration of its components. All effects of control signals on the robot are learned by the controller through visual observation during a training period and refined during actual operation. This method avoids computation of the inverse perspective projection and robot arm inverse kinematic transformations. This approach features small, efficiently trainable neural networks which exhibit sufficiently accurate performance for the authors´ reaching tasks. Their design confirms that successful arm control can be achieved without calculating the camera baseline parameters explicitly. Rather than recalibrating the system off-line following a minor change to its configuration, it is possible for the robot to adapt to the new mappings on-line