In addition to Subie Patel's work on the wireless insole project, I also supervised three other senior honors projects at UMass during the 2004-05 academic year: | |
Electromyography-based interface for human-computer interaction | |
| Anthony Leonardi developed a simple game where a player maneuvers a cursor across a busy highway. (Remember Frogger?) The catch is that cursor movements depend on the electrical activity of a player's muscles, detected with noninvasive EMG sensors at the skin. Most EMG-based interfaces use a threshold to classify the electrical activity as either on or off. For the test path in the figure, Anthony demonstrated statistically significant improvement when cursor movements were instead proportional to muscle activity. Future work toward intelligent interfaces for prosthetic limbs will need to account for contextual information and will also need to consider learning and adaptive methods for customizing the interface. |
A web-based interface for human-robot interaction | |
| Scott Myers designed and built a small robot that can sketch simple figures on a magnetic drawing toy. The photo shows a web-based interface that allows a user to control the robot with various assistive strategies. One interesting result is that a trail superimposed on the video display leads to statistically significant improvement in performance, quantified as total time to move the stylus through a sequence of targets while avoiding obstacles. The trail, however, only shows previous locations of the stylus with no explicit predictive information about future movement. Scott's conjecture is that the trail nevertheless improves the user's perception of the stylus movement and, therefore, supports more aggressive control of the robot. |
Haptic feedback for on-screen target selection | |
| Michael Mintz explored the benefits of force feedback for moving a cursor toward various targets without overshoot. Cursor movement was controlled using an old Logitech Wingman mouse (figure inset) that can simulate viscoelastic effects and simple textures. Michael tested various strategies for predicting which target that the user intended to click. The mouse then provided assistance by guiding the user toward the predicted target and by simulating a "bump" when the target was reached. The best strategy incorporated proximity information as well as directional information, quantified by the dot product between the cursor velocity vector and the vector pointing from the cursor toward the putative target. |
|
updated 14-Dec-2005 |