Planning for Robots with Skills

Modern industrial robotics is characterised by a need for flexibility in robot design, in order to minimise programming and development time when a robot's tasks must be changed. To address this problem, a recent approach has proposed that robots be equipped with a set of general, reoccurring operations called `skills', e.g., picking, placing, or driving. This paper presents a method for automatically generating planning problems from existing skill definitions such that the resulting problems can be solved using off-the-shelf planning software, and the solutions can be used to control robot actions in the world. As a result, a robot can therefore perform new tasks simply by specifying the task's goals via a GUI. The approach is demonstrated on a set of common tasks in a simulated industrial environment and has also been tested successfully on a real-world robotic platform.