Abstract
Robot manipulators may be used as flexible camera platforms by mounting cameras on the wrist of the robot. In this paper we present a new way of interaction between an operator and the camera platform, where the operator wants to get a visual overview of a remote operation. The goal is to relieve the operator from controlling both the operation and the camera platform simultaneously, and allow the operator to focus only on the operation while the camera plaform is automatically controlled based on learned operator preferences. We describe an architecture for learning from operator inputs, and use an active camera control algorithm as a base for learning. An M-RAN sequential function approximator is used as memory function. Experimental results on a demonstration case indicate that the camera platform responds to and remembers differences in operator preference
