Contributions to Conferences

Dynamic Collision Avoidance for an Anthropomorphic Manipulator using a 3D TOF camera

Abstract

This paper describes a practical approach for dynamic collision avoidance for an anthropomorphic manipulator using a 3D TOF (Time-Of-Flight) camera. These days robots are required to work closely and even interact with humans within its workspace. In order to avoid the risk of a collision with humans or machines, one has to ensure high safety standards. Generally, the global path planning is done on configuration space and in higher level of the manipulator control hierar-chy, so they are not suited for dynamic and unstructured environment. In recent years, researches have demonstrated different theories and practical approaches to handle the so-called dynamic collision avoidance. This local path planning for a manipulator is not computationally expensive and suited for dynamic environment. This paper is based on sensor based local planning approach proposed by Seraji and Bon [1]. Most of the dynamic collision avoidance methods use data from proximity sensor to avoid manipulator colliding with obstacles. Commonly used proximity sensors such as stereo cameras and artificial skin have some drawbacks. For instance, cameras possess time delays caused by complex calculations or produces errors due to bad lighting conditions. Artificial skin are computational expensive and has com-plex hardware parts. To avoid these problems we are using a 3D-TOF camera which produces fast and adequate depth information even for a moving object. Out of the visual information from the camera we are able to extract the manipula-tor and obstacles. Then a collision avoidance algorithm will move the elbow part of the manipulator away from the ob-stacle without perturbing the end-effector pose. The approach proposed here is simple and fast. It is applicable to any 7 DOF anthropomorphic manipulator, which has two spherical joints connected through a single revolute joint.