Toward Subtle Intimate Interfaces for Mobile Devices
Using an EMG Controller
Enrico CostanzaMedia Lab Europenow at MIT Media Lab
Samuel A. InversoMedia Lab Europe
Rebecca AllenMedia Lab Europe
Outline
• Motivation: Subtle and hands-free interaction• EMG as a solution• An EMG-based controller• Design approach• Formal user study• Conclusion
• Mobile interaction is often in public spaces• Subtle interfaces: do not disrupt the environment• Intimate interaction: only for the user• Ringing vs. vibrating alert
Importance of Subtlety in Mobile Interfaces
• Speech recognition and evident gesturing can be inappropriate
Importance of Subtlety in Mobile Interfaces
Design for Hands-free Interaction
Eyeglass displays
Design for Hands-free Interaction
• Electrical signal from muscle activity• Can measure isometric activity:
subtle or no movement• Surface Electrodes (EKG-like)• Non-contact sensing (future)
Electromyogram (EMG) as a Novel Solution
• Prosthesis control• Input devices for disabled users• Affect sensing• Music expression
EMG in CHI(Related Work)
• EMG and movement are not always related • Tanaka & Knapp report this as a limitation• We think it is an advantage!
EMG and MovementLimitation or Advantage?
• EMG greatest potential for mobile HCI• Sense subtle gestures• Example: brief contraction of the bicep
Motionless Gestures
• Self-contained in armband• Integration with Bluetooth devices
(e.g. Phones and PDAs)• No calibration for individual users
EMG-based Controller
The gesture should be:• Natural to perform• Different from normal muscle activity
User centered iterative approach:1. Select muscle & generic gesture definition
(non-detailed description to subjects)2. Definition refinement, model and algorithm3. Tuning
Design Process
• Realistic controlled environment: subjects walked around obstacles in trafficked walkway
• 10 subjects• Audio stimuli and feedback• Is training avoidable? (minimal feedback)• Push the limit: short and long contractions
Formal User Study
• 96% correct recognition• No false positives• No training necessary in 7 out of 10 cases• Cannot distinguish short and long contractions
across different subjects
Results
• EMG can be successfully used (96%)• Generally no training required• No calibration across users
Discussion
• Cannot distinguish short and long contractions• Subjective definition of “short” and “long”
Discussion
• Test in more complex scenarios• Measure subtleness• Improve algorithm• Use more muscles (alphabet definition)
Future Work
• Subtle interaction for mobile devices• New reason to use EMG in CHI• Motionless gestures• It works (96% correct recognition no false positives)
Summary and Conclusion