Finding Intuitive Action Spaces for Wheelchair-Mounted Robotic Arms

Operating high degree of freedom robots can be difficult for users of wheelchair mounted robotic manipulators. Mode switching in Cartesian action space has several drawbacks such as unintuitive control reference frames, separate translation and orientation control, and limited movement capabilities that hinder performance. Autonomous methods aim to improve the user experience by prompting the user for high-level goal input so that the system can autonomously completes the task. However, studies show users prefer to maintain control authority over the robot, and do not like explicitly stating their task goals. To accommodate for this, shared-autonomy methods have been proposed to assist the user through learning action maps or predicting goals and providing assistance. These methods rely on learning task specific representations or require ground truth information, both of which may impede real world implementation. Additionally, many control systems may be limited by choosing Cartesian mode switching as the basis for their method's control paradigm.

We assert that we can create new mode switching action spaces that reallocate movement reference frames into a more intuitive action space. We create a framework for mode switching that introduces wrist motions to the robot's action library, in addition to base-frame and end-effector aligned actions. With this framework, we create an action space representation of basis vectors to formally define different control spaces. After conducting a pilot study to find optimal action reference frames for activities of daily living, we propose Point and Go mode switching, consisting of a novel translation and rotation mode with many quality of life improvements to reduce the user's mental load.

We use a novel sweeping motion to point the gripper, which defines the new translation axis along the robot base frame's horizontal plane. This creates an intuitive `point and go' translation mode that allows the user to easily perform complex, human-like movements that would not be possible with Cartesian mode switching. The system's rotation mode combines position control with a refined end-effector oriented frame that provides precise and consistent robot actions in various end-effector poses. We verified its effectiveness through initial experiments that evaluated each feature of our system on its own. It was followed by a three-task user study that compared our method to Cartesian mode switching and learned State Conditioned Linear Maps. Results show that when comparing Point and Go mode switching to Cartesian-base control, we reduced completion times by 31%, workload by 12%, pauses by 41%, and mode switches by 33%, while receiving significantly favorable responses in user surveys and significantly better control smoothness. Additionally, we matched performance of State Conditioned Linear Maps on easy tasks, and exceeded it in more complex tasks. Point and Go mode switching can be implemented 'out of the box', without requiring sensors or data for learning, in current wheelchair-mounted robotic manipulators, and we hope can also serve as a more optimized control system for future methods to build upon.