activity-based ubicomp for health
DESCRIPTION
Describes the UbiFit project and how it relates to the general idea of activity-based computing. UbiFit was a join collaboration between Intel Labs Seattle and the University of Washington. The project attempts to use low-cost sensing, inference, and feedback to allow people to stay physically active. This project is an example of the larger thrust of activity-based ubiquitous computing. This was presented at the 3rd U.S.-China Computer Science Leadership Summit at Peking University, Beijing China on June 14, 2010.TRANSCRIPT
James LandayShort-Dooley ProfessorComputer Science & EngineeringUniversity of Washington* Joint work with Intel Labs
3rd US-China CS Leadership SummitPeking UniversityJune 14-15, 2010
Activity-based UbiComp for Health
Visiting Faculty ResearcherMicrosoft Research Asia
Activity-based UbiComp Can Help Improve our Lives
Long-lived activities in our everyday lives • e.g., staying healthy, graceful aging, learning a language• high-level, physical, dynamic, & high value
Key elements: social, natural UIs, always at hand
Hard to create successfully solely with traditional CogSci-based HCI or Art Studio-based Design
Importance of Physical ActivityRegular physical activity is critical to
physical & psychological health
Spending on fitness gadgets & equipment is on the rise
Rates of inactivity also rising
How can we encourage people to be physically active?
ubifitActivity-based Application
Problem• overweight/obesity a global epidemic• have hard time fitting exercise into lives
Solution: Ambient feedback of activity
Intel/UW: Consolvo, McDonald, Landay …
CHI 2008, Ubicomp 2008, CHI 2009
3 Main Components of UbiFit Garden
glanceabledisplay
interactiveapplication
fitness device
+ +
collects data about physical activities
communicates data about physical activities
The Glanceable Display
strength
cardio
flexibility
walk
this week’s goal met
recent goal met
Fitness DeviceIntel Mobile Sensing Platform – MSP
Infers physical activities & their durations, specifically• walking• running• cycling• use of elliptical trainer• use of stair machine
IEEE Pervasive Computing, 7(2), 2008
Choudhury, Lester,Borriello, LaMarca,LeGrand, …
Activity Journaling
manuallyon
phone
any physical activity including those not inferred by the fitness device
automatically
Inside the Intel MSP
mobile sensing platform (msp)
battery
9 sensors 3d accelerometer barometric pressure humidity visible light infrared light temperature temperature (a 2nd) 44 khz microphone compass (optional)
2
Robust Action Inference:Human Actions from Motion
collect rawsensor readings
calculatefeatures
producemargins
measure of confidencefor particular activities
mean, median,range, etc.
smooth margins intomeaningful actions
Send margins to phonevia bluetooth
Intel MSP
> 95% accuracy on smartphones (Android, iPhone, Windows) for walking, running, biking, standing
IEEE Pervasive Computing, 7(2), 2008
Choudhury, Lester,Borriello, Landay, Fogarty, Saponas…
3-week pilot field trial (n=12)• shake out system & get feedback on UI/inference
3-month field trial (n=28), 3 conditions• full system (n=10): interactive app + MSP + garden• no MSP (n=9): interactive app + garden• no garden (n=9): interactive app + MSP
Results of the field trial• participants with garden maintained weekly activity level over the study• participants without the garden showed a significant decrease over time• strong enthusiasm for a garden or similar metaphor on phone’s background• participants wanted system with automatic activity inference
UbiFit Evaluation
garden
UbiFit Lessons LearnedOver 2 Years of Development & Testing
Activity inference difficult - to collect data for, train, & tune
Design → coded system = BAD!- hard to change & iterate
Evaluation time consuming- 2 full time researchers
Left “mass of data on the table”- no easy way to understand
Activity-based UbiCompKey Challenges & New Ideas
Physical actions are tedious to record & manageBuild applications using action inference
Natural interactions are ambiguousImprove disambiguation using dynamic context
Must study in situ over extended periodsUse new methods & tools to improve data collection,
analysis & application prototyping
Analytical Design Studio
Novel ToolsMyExperience – context-aware experience sampling*ActivityDesigner – design & prototyping for designersActivityViz – visual analytics of activity data
Landay, Edge, Kientz, Kolko, Lee, Patel, Philipose, Ramey, Riche, Roesler, Zhao
*AKA Ecological Momentary Assessment
Activity-Based UbiComp SummarySolve high value problems, improving our lives by using
• inference: actions & high level activities• tools: for visualization, design, & user
studies• natural UIs: improve recognition using context
UbiFit uses Activity-based UbiComp for Health
+ +@MSRA: Looking at higher level concept of wellness, mobile phone for sensing, & cultural differences
James LandayShort-Dooley ProfessorComputer Science & EngineeringUniversity of Washington
3rd US-China CS Leadership Summit Peking UniversityJune 14-15, 2010
Activity-based UbiComp for Health
Visiting Faculty ResearcherMicrosoft Research Asia
[email protected]://dub.washington.edu