Composition and Incremental Refinement of Skill Models for Robotic Assembly Tasks

Abstract

Skill-based approaches for programming robots promise many benefits such as easier reuse of functionality across applications, encapsulation and hiding of process details, and often a certain level of hardware abstraction. While many previous approaches allow to compose high-level skills, the basic skills themselves are usually atomic, non-extensible, or are supposed to be created by experts beforehand. We designed a prototype-based skill model to create forcecontrolled manipulation skills for robotic assembly tasks. In this paper, we show how skills can be created by composing subelements, each modeling a different aspect of a skill such as kinematics, task description, or coordination. Using prototype-based inheritance, skills can be incrementally refined and extended to create new skills, instead of creating each skill from scratch. Our skill model uses iTaSC for task specification, hierarchical statecharts for coordinating the execution of skills, and provides a domain-specific language that allows to quickly compose and parameterize skills and applications.