A 2-D Mapping Approach for Monitoring and Diagnosing the Execution of Robotic Assembly Tasks

Abstract

Summary Assembly task planning is concerned with generating the sequence of operations and the required detailed execution instructions. Realtime monitoring and diagnosing of uncertain events, based on the latest feedback from sensors (vision, tactile, force, etc.), during robot assembly tasks execution plays a vital role in ensuring a reliable and robust assembly. This paper presents techniques developed for realtime monitoring of assembly operations, detection of errors due to parts mishandling such as incorrect parts manipulation, insertion and placement in the workspace or due to interference between robots sharing the workspace. Two—dimensional maps are created by adaptively projecting the typically three or more dimensional assembly problems to a simpler two—dimensional problem space which provides the flexible and efficient associative properties needed for effective realtime monitoring and diagnosis. A self—organizing neural network (Kohonen map) is used for organizing the various inputs regarding parts, tools and sensors feedback during robotic assembly and diagnosing the sources of errors. Replanning of the assembly to recover from errors would then proceed based on this feedback. The results of this research have been demonstrated in the real assembly of a dishwasher power unit.