I. Introduction
The aim of assistive robotic arms is to restore manipulation capabilities of people with disabilities, thereby enabling them to perform tasks of daily living. An example is EDAN (EMG-controlled Daily AssistaNt), which consists of a DLR Light-Weight Robot III with a DLR-HIT hand, mounted on a power-wheelchair, see Fig. 1.a. Since goal-directed physical manipulation of the environment is often complex and intricate, controlling the robotic arm can be difficult, and may lead to a high cognitive workload. This is especially the case for mobile manipulation systems such as EDAN, as they have many degrees of freedom (DoFs) which all need to be controlled appropriately to achieve a task. For instance, opening and going through a door poses a real challenge, as grasping the handle is intricate, and opening the door requires the coordination of both arm and wheelchair movements [1].