People who have lost a significant part of one or both arms, or a significant degree of arm function, may perform activities of daily living through a robotic prosthesis. Any such arm is complicated to learn to use, requiring a few distinguishable actions from the user to be mapped to motion of the robot's many joints. In wearable prostheses for users with amputation or limb difference, the joints are usually controlled sequentially, switching between modes. In wheelchair-mounted robots for users with mobility impairments, it is common to control the translation and rotation of the gripper instead (thus operating in the task space, rather than the joint space), which also involves switching between modes.

Task space control is generally expected to be easier than joint space control, and is more naturally suited for autonomous assistance. Yet, there is little evidence for or against its use in wearable prosthetic arms, as research has been stunted by the inaccessibility of hardware. In wheelchair-mounted arms, where task space control exists as an established paradigm, entire regions of the workspace are typically blocked off to avoid singularities in the inverse kinematic map (the mapping from desired motion in the task space to the corresponding joint movements). Moreover, mode-switching can be confusing and time-consuming in both paradigms. While autonomous assistance through goal detection and motion planning can technically reduce the need for mode-switching, many users find it unhelpful, as it robs them of agency and is at risk of making mistakes in the real world.

This talk will first describe the development and use of an open-source immersive simulation platform for researchers to iteratively design task space control strategies for wearable prostheses, testing varying degrees of autonomy, and varying forms of informative feedback during operation. We will then propose ways to address current challenges in task space control of wheelchair-mounted arms, by making singular configurations safely reachable, displaying the mode clearly during operation, and designing autonomous assistance for only those degrees of freedom that are less intuitive to navigate with direct control.