Robotic Work-Space Sensing and Control

Abstract: Industrial robots are traditionally programmed using only the internal joint position sensors, in a sense leaving the robot blind and numb. Using external sensors, such as cameras and force sensors, allows the robot to detect the existence and position of objects in an unstructured environment, and to handle contact situations not possible using only position control.

This thesis presents work on how external sensors can be used in robot control. A vision-based robotic ball-catcher was implemented, showing how high-speed computer vision can be used for robot control with hard time constraints. Special attention is payed to tracking of a flying ball with an arbitrary number of cameras, how to initialize the tracker when no information about the initial state is available, and how to dynamically update the robot trajectory when the end point of
the trajectory is modified due to new measurements. In another application example, force control was used to perform robotic assembly. It is shown how force sensing can be used to handle uncertain positions