Modelling and learning of building blocks for shared control templates

Abstract

In the field of assistive robotic robots aim to help users with impairments to execute daily living activities in everyday household environments. Controlling the high degrees of freedom of those robots directly, with a joystick, is often tedious and can cause a high cognitive workload. Using Shared control methods, like Shared Control Templates, provide task-specific guidance and constraints to facilitate control. However, designing SCTs requires robotic expertise and was previously hand-crafted. To make such designs easier and faster, we extend the SCT approach to facilitate the generation of SCT templates on the basis of demonstrations. We show the capabilities of this approach in a set of daily living tasks such as opening a drawer and pouring with the wheelchair-mounted robot EDAN.