Automated Planning of Industrial Logistics on a Skill-equipped Robot

Automated task planning for robots is usually implemented on a motion primitive domain, where the focus is on constructing meaningful, general motion primitives. In this work we propose planning on the higher abstraction level of robot skills. In this context, skills are general, functional blocks that contain both sensing and action, have a welldefined expected outcome and set of preconditions, and are thus immediately useful for planning. By using a world model, which is focused on the skills, we show that this is in fact the case. We show automated task planning and execution, as a sequence of skills and their parameters, based on the desired goal state and the current state from the world model. Experiments show that the approach is immediately applicable, given a skill-equipped robot, and that inconsistencies between the world model and physical world are overcome simply by replanning.