Toward Safe Human Robot Collaboration by Using Multiple Kinects Based Real-Time Human Tracking

Abstract

We present a multiple Kinects based exteroceptive sensing framework to achieve safe human-robot collaboration during assembly tasks. Our approach is mainly based on a real-time replication of the human and robot movements inside a physics-based simulation of the work cell. This enables the evaluation of the human-robot separation in a 3D Euclidean space, which can be used to generate safe motion goals for the robot. For this purpose, we develop an N-Kinect system to build an explicit model of the human and a roll-out strategy, in which we forward-simulate the robot's trajectory into the near future. Now, we use a precollision strategy that allows a human to operate in close proximity with the robot, while pausing the robot's motion whenever an imminent collision between the human model and any part of the robot is detected. Whereas most previous range based methods analyzed the physical separation based on depth data pertaining to 2D projections of robot and human, our approach evaluates the separation in a 3D space based on an explicit human model and a forward physical simulation of the robot. Real-time behavior (≈ 30 Hz) observed during experiments with a 5 DOF articulated robot and a human safely collaborating to perform an assembly task validate our approach.