Knowledge-Based Planning for Human-Robot Collaborative Tasks

Abstract

Human-robot collaboration is a promising alternative to full automation and manual labour. Collaborative robots are considered safe and individual robot actions can often be easily programmed, for example, by physical hand-guiding. Coordinated collaboration, where tasks and the environment are shared, cannot be so easily achieved, due to continuously changing conditions and actions that need to be triggered at unknown instances. Besides, knowledge required for collaboration is difficult to program into robotic systems. This paper presents a system architecture that aims at facilitating human-robot collaboration by alleviating the need for pre-programmed information and exploring ways to teach new skills. This is done by reducing the amount of information required to program tasks by utilizing a knowledge base that represents knowledge on tasks, actions and the world. Automatic reasoning over conditions and properties of the knowledge is then utilized to generate available actions and action plans in order to complete the shared tasks. Moreover, learning new tasks is enabled by extending the original knowledge base, concatenating available actions and tasks into news bricks of knowledge. Two examples, a kitting task and a handover task, serve to validate the system architecture and exemplify its usage. Experiments demonstrate that by combining reasoning methods and knowledge-based planning, high-level shared tasks can be generated and executed, and robots can act reliable as teammate to human operators.