Introduction
Assistive robots are designed to help individuals with disabilities regain independence. However, a significant "dimensionality gap" exists: while a typical robot arm operates in 6D (3D position and 3D rotation), most accessible interfaces, such as wheelchair joysticks, provide only 2D input. This discrepancy often forces users to perform complex mental rotations to translate joystick movements into robot actions, leading to high cognitive load and frustration. This research introduces a dynamic input mapping framework that bridges this gap by aligning simple inputs with the robot's task geometry.
The Canal Surface Framework
To simplify control, the researchers utilize a framework called GeoSACS, which encodes robot tasks as canal surfaces. A canal surface can be visualized as a 3D tube composed of a sequence of 2D disks. The robot autonomously navigates along the center line of this tube, while the user provides 2D corrections within the plane of the current disk to fine-tune the movement, such as adjusting a brush stroke while painting.
Dynamic Input Mapping
The primary contribution of this work is a Dynamic Input Mapping Framework that ensures these 2D corrections feel natural and intuitive. The framework follows a four-step process:
- Context-Aware Alignment: The system classifies the canal's disks based on their orientation—typically "near-horizontal" (like a tabletop) or "near-vertical" (like a wall or canvas).
- Perspective Projection: Joystick axes are projected onto the robot's control frame based on the user's likely viewpoint. For instance, on a vertical surface, "forward" on the joystick is mapped to "up" for the robot.
- Smoothing and Optimization: An optimization algorithm prevents abrupt flips or rotations of the control axes as the robot moves through bends in the canal, maintaining a consistent feel.
- Boundary Flexibility: A "moving out of canals" feature allows the robot to slightly exceed the pre-defined tube limits to reach objects that may have shifted from their expected positions.
Experimental Validation and Results
The framework was validated through three user studies, including a comparative study with 20 participants and an exploratory study with wheelchair users utilizing specialized accessible gamepads. The results demonstrated significant improvements over the standard GeoSACS baseline:
- Efficiency and Accuracy: In object relocation tasks, completion time dropped from 259s to 220s. Furthermore, 100% of participants successfully placed all objects using the dynamic mapping, compared to 85% with the baseline.
- Subjective Experience: Participants rated the system significantly higher for Ease of Use (6.50 vs 3.92) and Satisfaction (6.54 vs 4.29) on a 7-point scale.
- Reduced Workload: NASA-TLX scores indicated a dramatic reduction in mental demand and frustration. Users described the experience as "like playing a game," suggesting the system successfully transitioned the task from a complex coordination puzzle to an engaging activity.
Insights and Future Directions
The study highlights that users prioritize a perceptual coordinate system over a mathematically pure one. By optimizing for common "carpentered" environments (horizontal and vertical surfaces), the system covers the majority of activities of daily living.
Future work aims to address visual occlusion—where the robot's body blocks the user's view—and move toward vision-based canal generation. This would allow the robot to automatically generate a control "tube" by observing its environment, further reducing the setup time for new assistive tasks.
Summary updated