A planning system for generating manipulation sequences for the automation of maintenance tasks

Abstract

The adaption of autonomous robots to complex assignments like automating maintenance tasks, requires the study of novel planning systems for task generation which use a priori and sensor-based data. This paper presents an approach to disassembly and assembly planning for automating maintenance tasks in production systems. The planning system combines CAD data with visual data from an RGB-D sensor. This allows the creation of a consistent environmental model, which can be used to plan a (dis)assembly task for an autonomous robot. The proposed method allows to compute manipulations directly in the task space of the robot. The algorithm uses a geometric and semantic reasoning concept to compute a relational graph which describes the individual degrees of freedom in-between the components. The proposed algorithm is highly time efficient, thus being particularly suited for the automation of maintenance tasks. With the additional integration of visual data, it is possible to compensate environmental uncertainty from the CAD data as well. This paper describes the planning architecture and algorithms in detail and concludes with an experimental validation.