Intelligent Monitoring of Assembly Operations

Abstract

To ensure safety, state-of-the-art robotic assembly environments typically require separation of assembly robots from human assembly workers. This separation has the disadvantage of requiring significant space and cost to install, and more importantly, it forces humans and robots to work in isolation, leaving little opportunity for cooperation. The Intelligent Monitoring of Assembly Operations (IMAO) project aims to develop an accurate and precise safety monitoring system for industrial assembly workcells, one which will allow closer interaction of human and robotic workers than is currently possible while still maintaining a safe environment. By enabling humans and robots to work together, we can open a whole new array of cooperative assembly strategies that use the strengths of both human and robotic workers to maximum effect. In this work, we first present the IMAO system architecture and key engineering accomplishments. Our proposed safety system uses multiple 3D range cameras to detect people, compares their locations against robot locations and possible future trajectories, and uses the resulting data to predict and prevent any possible collisions. Secondly, we present in detail several research challenges that have been addressed during the course of this work. One of these is a new method in accessibility-based background subtraction for evidence grids, and the other is a fast means of computing reachability bounds within manipulator workspaces based on joint-level motion constraints. Finally, we present results of a preliminary system evaluation using a small test workcell equipped with four sensors and a seven degree-of-freedom manipulator. Initial tests indicate that IMAO is effective for detecting people and preventing collisions in assembly environments, even when people are in close proximity to robotic manipulators. Further testing and evaluation is still ongoing to improve reliability and establish robustness for operation in complex real-world environments.