Vision for Space Robots on a Parallel Distributed Architecture

Abstract

This article reports on two main contributions: the introduction of a parallel distributed architecture for robot vision, and robust robot vision techniques and their computationally efficient parallel implementation for grasping a polyhedral object in 3-D space. The parallel distributed architecture for robot vision is based on transputers (message-passing parallel processors) and vector signal processors. Vision algorithms including straight line detection, binocular stereo correspondence, surface reconstruction, motion estimation, recognition, and localization have been implemented in the OCCAM parallel programming language and run in adequate processor network topologies. The integration of the Robot Vision System into the DLR Telerobotics Ground Station and its application in the frame-work of the first German space robotics experiment ROTEX (“RObotics Technology EXperiment”) to the task of grasping a floating object, the DLR FreeFlyer, under μ-gravity is presented. For this task a hybrid approach to parallelism has been adopted involving geometric and algorithm parallelism. The integration of the whole set of robust robot vision techniques for this task is reported for the first time in this article.