Robotics Group, Dept. of Mechanical Engineering & Aeronautics University of Patras, Rio 26504, Greece
Abstract: Robots that share workspace with humans require safety capabilities to identify possible collisions and perform appropriate reactions in order to eliminate the potential injury of a human. For this purpose, an intelligent fuzzy identification method and a time series method are proposed and implemented in this paper. These systems are trained to the robot dynamics with and without collision to be able to detect contacts along the links of the robot. Since the speed of the detection is of critical importance to avoid injuries, attention is paid to recognise a collision as soon as possible. The proposed methods are evaluated in an experimental setup by using a KUKA LWR manipulator and their performance is compared with a model-based approach.
Paper appeared in RAAD 2013 Conference 22nd International Workshop on Robotics in Alpe-Adria-Danube Region September 11-13, 2013, Portoroz, Slovenia