Transcript Computer Engineering Program

Smart Environments:
Artificial Intelligence in
the Home and Beyond
Diane J. Cook
[email protected]
http://eecs.wsu.edu/~cook
January 2008
1
BodyNets Topics
Smart Environments
Both generate large
amounts of data
Both need:
• data mining
• prediction
• classification /
identification
2
What Is a Smart Environment?
An environment able to acquire and apply
knowledge about the resident and the
physical surroundings to improve the
resident’s experience.
Imagine…
Automated blinds
Door/lock controllers,
Face
recognition, automated door entry
Surveillance
system
Climate control
Intelligent appliances
Robot vacuum cleaner
Lighting control
Remote
site
monitoring
and
control
Robot lawnmower
Smart sprinklers
What Good Is a Smart Environment?
• Comfort and Productivity
• Health and Safety
• Energy and Sustainability
Our Initial Hypothesis
Resident interactions with smart
environments can be automated
by analyzing and predicting resident
behavior and using the information
to learn an automation model.
Smart Apartment
Smart Workplace
Technology
Perception
Reason
Sensor
network
Data mining
Prediction
Decision maker
Rule engine
Off-the-shelf
sensors
Action
PLC
Sensor
network
Data Mining
Raw data
Patterns
2005-01-03 09:47:30 i 5 1 100 mavpad
2005-01-03 09:56:17 i 5 0 0 mavpad
2005-01-03 13:04:45 a 1 1 100 mavpad
2005-01-03 13:05:37 i 3 1 100 mavpad
2005-01-03 13:06:11 c 4 1 100 mavpad
2005-01-03 13:06:22 c 4 0 0 mavpad
2005-01-03 13:16:33 S 19 1 100 mavpad
2005-01-03 13:16:33 S 70 1 100 mavpad
2005-01-03 13:16:34 i 5 1 100 mavpad
2005-01-03 13:16:44 i 3 0 0 mavpad
…
% Turn on kitchen light when in room
[PRODUCTION_PATTERN]
ID = 56008634
4 V_19_1 V_19_0 i_1_1 x_0_0
NumberOccurrences = 5
V_19_1 V_19_0 V_19_1 V_19_0…x_0_0 1
% Kitchen Activities
[ABSTRACT_PATTERN]
ID = 54964
4 1487735365 1487712271 56008634 100100
NumberOccurrences = 1
1487735365 1487712271 56008634 100100 1
101000101101000…
Prediction

a (5)
a (2)
b(2)
a (1)
c(1)
b(4)
a (1)
b(2)
“aaababbbbbaabccddcbaaaa”
c(1)
d(2)
c(1)
a (1)
• Parse sensor events using text compression
• Given recent history, predict next symbol
(sensor event) with highest probability
Decision Model
• Learn hierarchical
model from data
• Learn automation
rules with user and
rule engine
feedback
Results - Workplace
Manual interactions
200
150
100
Inhabitant
Prediction
50
0
Flat model
Hierarchical model
Results - Apartment
Manual interactions
20
15
10
5
0
Inhabitant
Prediction
Flat model
Hierarchical model
Prototype Smart Apartment
What Good Is a Smart Environment?
• Comfort and Productivity
• Health and Safety
• Energy and Sustainability
The Crisis of an Aging Population
3.5
3
2.5
2
1.5
1
0.5
0
0-14
15-59
60+
80+
All
By 2040, 23% of U.S. population
will be 65+
• Facilities cannot handle coming
“age wave”
The Crisis for Individuals
40% cannot
afford assisted
care
Many live alone
in remote
locations
Caregivers live
far away
Removing them
results in
cognitive and
physical decline
Virtually ALL want to stay at home
Sense
Assess
Assist
Financial Impact
Americans spend $44
billion/year for home care
Family members provide
$197 million/year
U.S. healthcare spending
will increase by 25%
Identify
Remind
10 / 2 / 15
• If we can keep 10% of population home 2 years longer,
we will save Washington State $341 million/year
• $15 billion/year nationally
Millions of Dollars
3000
Geriatric Monitoring
Acute Monitoring
Wellness Monitoring
E-visit
2000
1000
0
2005
2010
Mission
• Our vision is to keep older adults
functioning independently in their
own homes as long as possible.
• Our method integrates artificial intelligence,
machine learning, and pervasive computing
technologies into their everyday
environments, turning them into smart
environments.
What Assistance Is Needed?
Activity identification
Functional assessment
Tracking
Eating
Toileting
Medical monitoring
Medication management
Now
Partial
Future
Activities of Daily Living (ADL) Technology
Sense
Assess
Assist
Identify
Remind
23
Sense
Assess
Sense
Assist
Identify
Remind
Sense
Identify
Assess
Assist
Identify
Remind
Sense
Assess
Assist
Model
Motion
2
0.1
1.0
0.4
Motion
1
Motion
4
0.6
Motion
3
1.0
Water
On
1.0
0.9
Water
Off
1.0
END
P(“wash hands” | [Motion 1, Motion 3, Water On, Water Off]) = 0.54
P(“cooking” | [Motion 1, Motion 3, Water On, Water Off]) = ?
Identify
Remind
Sense
Remind
Assess
Assist
Identify
Remind
Sense
Assess
Assist
• Medicine compliance
Remind
Identify
Remind
Sense
Assess
Assist
Identify
Remind
Assist
3
2.5
2
1.5
1
0.5
0
Voice / Hand
Disability
29
Sense
Assess
Assist
Assess
Replace lab with everyday
environments
Detect dementia as it occurs
Monitor as it progresses
Identify
Remind
What Good Is a Smart Environment?
• Comfort and Productivity
• Health and Safety
• Energy and Sustainability
31
Load Profiling for Smart Home
Visualization of
energy usage
• Play in real time or rewind
• Annotation with activity IDs
Energy-Efficient Automation
Alternative Automation of
Daily Activities
• Open south windows to warm room
• Automatically learn strategy based
on
• Positive feedback (reduced energy
usage) and
• Negative feedback (resident
reverses automation)
Smart Communities
Learn usage profiles for
entire community
• Use information to balance load
• Avoid brownouts and blackouts
• Reward customers who sign up
for conservation plan
What’s Next?
Multiple
residents
Multiple
perspectives
Multiple
environments
35
To Learn More
http://eecs.wsu.edu/~cook
[email protected]
36