1 mobil computing © 2010-2013 daniel p. siewiorek end-to-end application considerations daniel p....
TRANSCRIPT
1Mobil Computing© 2010-2013 Daniel P. Siewiorek
End-to-End Application End-to-End Application ConsiderationsConsiderations
Daniel P. SiewiorekDaniel P. SiewiorekCarnegie Mellon Carnegie Mellon
UniversityUniversity
February 2013February 2013
2Mobil Computing© 2010-2013 Daniel P. Siewiorek
Outline
Introduction Individual
» Wearable Computers
» Augmented Reality
Infrastructure» Museums
» Hospital
» e-Display
Ad Hoc» Wildfire
» Volcano
Concluding Lessons
3Mobil Computing© 2010-2013 Daniel P. Siewiorek
Outline
Introduction Individual
» Wearable Computers
» Augmented Reality
Infrastructure» Museums
» Hospital
» e-Display
Ad Hoc» Wildfire
» Volcano
Concluding Lessons
4Mobil Computing© 2010-2013 Daniel P. Siewiorek
5Mobil Computing© 2010-2013 Daniel P. Siewiorek
6Mobil Computing© 2010-2013 Daniel P. Siewiorek
Wearable Applications and Architecture
Procedures - upload at completion Work Orders - incremental updates Collaboration - real time interaction
» Client-Server
– Thin Client Legacy Systems
– Interactive Electronic Technical Manuals (IETMs)
7Mobil Computing© 2010-2013 Daniel P. Siewiorek
Time Rate of Change of Data Taxonomy
Procedures. Maintenance and plant operation applications are characterized by a large volume of information that varies slowly over time.
A typical request consists of approximately ten pages of text and schematic drawings. Changes to the centralized information base can occur on a weekly basis.
8Mobil Computing© 2010-2013 Daniel P. Siewiorek
Savings Using Tactical Information Assistants in Marine Heavy Vehicle MaintenanceCurrent Practice
Current Practice
SAVINGS FACTOR
VuMan 3 Field Trials
VuMan 3 Field Trials
Personnel
2:1
Inspection time
40% less
SAVINGS FACTOR
9Mobil Computing© 2010-2013 Daniel P. Siewiorek
Time Rate of Change of Data Taxonomy
(continued)
Work Orders. The trend is towards more customization in systems.
Manufacturing or maintenance personnel receive a job list that describes the tasks and includes text and schematic documentation. This information can change on a daily or even hourly basis.
10Mobil Computing© 2010-2013 Daniel P. Siewiorek
Time Rate of Change of Data Taxonomy
(continued)
Collaboration. An individual often requires assistance. In a “Help Desk” an experienced person is contacted for audio and visual assistance. The Help Desk can service many people simultaneously.
Information can change on a minute-by-minute and sometimes even a second-by-second basis.
11Mobil Computing© 2010-2013 Daniel P. Siewiorek
C-130 Help Desk
12Mobil Computing© 2010-2013 Daniel P. Siewiorek
Lessons Learned From Usage
Maximum Size, weight, energy consumption before change user behavior
No fixed relationship between input/output/display
User less patient, expect instant response Intuitive to use, no user’s manual Information overload, user may focus on
computer rather than physical world Potential to lose initiative, user does exactly
what the computer tells them to do
13Mobil Computing© 2010-2013 Daniel P. Siewiorek
Based on a LectureDan Siewiorek, Thad Starner, Asim Smailagic
Morgan & Claypool
U - User C - Corporal
» Interface physically without discomfort or distraction
A - Attention» Divided between physical and virtual world
M - Manipulation» Controls quick to find, easy to operate
P - Power
14Mobil Computing© 2010-2013 Daniel P. Siewiorek
Major Factors in Portable Electronic Systems and Their Relationship to Design Disciplines
Power
Corporal
Manipulation
User
Attention
15Mobil Computing© 2010-2013 Daniel P. Siewiorek
UCAMP
User
16Mobil Computing© 2010-2013 Daniel P. Siewiorek
Sensor-based models of human activity and situation
Example:» Using just the sensors in a laptop we have predictive
model that determines “now is a bad time to interrupt” more accurately than human observers
Full Sensor Sets Only aLaptop
No Camerasor Microphones
76.9%Human
Observers
All
Sub
ject
s
Man
ager
s
Res
earc
hers
Inte
rns
Man
ager
s
Res
earc
hers
Man
ager
s
Res
earc
hers
Inte
rns
Better than human observers with just sensors already in laptop!
17Mobil Computing© 2010-2013 Daniel P. Siewiorek
Cognitive Load
18Mobil Computing© 2010-2013 Daniel P. Siewiorek
fMRI Experiment Configuration
19Mobil Computing© 2010-2013 Daniel P. Siewiorek
JUL 31, 2001
Car Calls May Leave Brain Short-Handed
By SANDRA BLAKESLEE
Scientists have bad news for people who think they can deftly drive a car while gabbing on a cell phone.
The first study using magnetic resonance images of brain activity to compare what happens in people's heads when they do one complex task, as opposed to two tasks at a time, reveals a disquieting fact: the brain appears to have a finite amount of space for tasks requiring attention.
The study, published in the Aug. 1, 2001 issue of the journal NeuroImage, was led by Dr. Marcel Just, a psychology professor and co- director of the Center for Cognitive Brain Imaging at Carnegie Mellon University in Pittsburgh...
20Mobil Computing© 2010-2013 Daniel P. Siewiorek
Right Temporal
Left Temporal Right Parietal
Left Parietal
0.0
10.0
20.0
30.0
40.0
50.0
60.0
70.0
80.0
90.0
Sentences Alone Rotations Alone Sum of Single Tasks Dual Task
Capacity constraint in association areas:Activation volume is less in dual task compared to single tasks, even for tasks without cortical overlap (auditory comprehension and mental rotation)
21Mobil Computing© 2010-2013 Daniel P. Siewiorek
Assessing measures of cognitive load
while performing tasks in divided attention
• Cognitive load can be a measure for whether, when and how to present information while performing a number of different tasks
• Need to manage human attention and understand when users can attend to new information
• Necessary while driving, learning new skills• Particularly an issue with elders
22Mobil Computing© 2010-2013 Daniel P. Siewiorek
Assessing measures of cognitive load
• We are working to develop a reliable, real-time and objective measure of cognitive load for contexts of divided attention
• Current techniques not appropriate• Post-hoc, so not real time
• Not accurate
• Subjective measures
23Mobil Computing© 2010-2013 Daniel P. Siewiorek
Study of cognitive load
• Six ECTs: visual perception and cognitive speed• Four sensors• Gaze tracking
24Mobil Computing© 2010-2013 Daniel P. Siewiorek
Assessing Cognitive Load
• 20 young adults• Compared sensors for determining cognitive load
• Best feature: > 74%
• Sensors/features: median heat flux, ECG
• > 81% across all participants
• Older adults• Same sensors useful, although heat flux also seems important
• BodyMedia device seems to be sufficient to capture cognitive load with elementary tasks
• Need to see how this transfers to real settings• Then, use information about cognitive load in virtual coaches
25Mobil Computing© 2010-2013 Daniel P. Siewiorek
User Lessons
“The user is not I”» Built two dozen systems to identify repeated patterns (e.g.
procedures, work orders, team collaboration) Mobile Users are more impatient than desktop users
» Must operate more like a flashlight than a computer rebooting Institutional Review Board
» “But madam, they are Marines and they do what they are told” Be prepared to reduce functionality
» “It’s a fine goal” Instant Creditability
» “I served with you on the Vinson’s first world cruise”
26Mobil Computing© 2010-2013 Daniel P. Siewiorek
UCAMP
Corporal
27Mobil Computing© 2010-2013 Daniel P. Siewiorek
“If it looks good,it will fly well”
MoCCA and VuMan 3 received the prestigious Industrial Design Excellence Awards (IDEA) from award co-sponsors Business Week magazine and the Industrial Designers Society of America (IDSA).
VuMan 3MoCCA
28Mobil Computing© 2010-2013 Daniel P. Siewiorek
NATO Soldier Using CMU TIA-P Language Translator in Bosnia
29Mobil Computing© 2010-2013 Daniel P. Siewiorek
Corporal Lessons
Air Logistics Command, Sacramento CA» The vapor barrier on a hot August day
Fort Gorden Battle Command» “Does this bend?!”
Six units to Bosnia ….. none returned
30Mobil Computing© 2010-2013 Daniel P. Siewiorek
UCAMP
Attention
31Mobil Computing© 2010-2013 Daniel P. Siewiorek
On
A KC-135 Aircraft was being pressurized at ground level. The outflow valves which are used to regulate the pressure of the aircraft were capped off during a 5 year overhaul and never opened back up. The post-investigation revealed: that a civilian depot technician who, "had always done it that way," was using a homemade gauge, and no procedure.
The technician's gauge didn't even have a max "peg" for the needle and so it was no surprise he missed it when the needle went around the gauge the first time. As the technician continued to pressurize the aircraft, and as the needle was on its second trip around the gauge the aircraft went "boom" - the rear hatch was blown over 70 yards away, behind a blast fence!
But, I’ve Always Done it This Way!
32Mobil Computing© 2010-2013 Daniel P. Siewiorek
KC-135 Pressure Test Results
33Mobil Computing© 2010-2013 Daniel P. Siewiorek
Attention Lessons
The more information and more engaging the virtual world, the less interaction with the physical world
Use systems for reference (e.g. information dashboard) then focus on physical world
34Mobil Computing© 2010-2013 Daniel P. Siewiorek
UCAMP
Manipulation
35Mobil Computing© 2010-2013 Daniel P. Siewiorek
Selection of “hot links” with CMU’s Wheel/Pointer
36Mobil Computing© 2010-2013 Daniel P. Siewiorek
Manipulation Lessons
Strong user interaction mental model» Not only trained in the field in less than ten minutes but
also able to train the next user» No users manual» Training without access to what the user is seeing
Without stationary reference, users easily become disoriented and confused with traditional input devices
» “Which is the left button” Control devices have to be flexible to mount
on different parts of the body» Boeing – shoulder holster» Sailors on aircraft carrier – shoulders used to carry tie-
down chains
37Mobil Computing© 2010-2013 Daniel P. Siewiorek
UCAMP
Power
38Mobil Computing© 2010-2013 Daniel P. Siewiorek
Power Lessons
It’s the peak, not the average » Muffled pops
Warning labels are no substitute for good design
More to be gained on reducing demand than increasing source
» Current batteries have half the energy density of dynamite
» Current user interactions demand all resources to reduce response time
Land Warrior» Army standard operating system
» Nine different battery types
39Mobil Computing© 2010-2013 Daniel P. Siewiorek
Speech Enabled Augmented Reality for Maintenance [Goose]
SEAR – Speech Enabled Augmented Reality 3 D Augmented Reality graphical view with
location-sensitive 3 D speech-driven interface 3 D component specific vocabulary triggered by
proximity sensors» Display component specific commands when in vicinity
» Key word activated
Simultaneous 3 D “parameterized” synthesized speech streams
» For current values from physical plant
Coded visual markers for tracking, location» 4 X 4 matrix with dots for over 10,000 combinations
40Mobil Computing© 2010-2013 Daniel P. Siewiorek
Augmented Reality - Maintenance of Nuclear Power
Plant Components
Room “augmented” with the CAD drawing of a floorplan
41Mobil Computing© 2010-2013 Daniel P. Siewiorek
Speech Enabled Augmented Reality for Maintenance [Goose]
System Architecture and User Interface
42Mobil Computing© 2010-2013 Daniel P. Siewiorek
Outline
Introduction Individual
» Wearable Computers
» Augmented Reality
Infrastructure» Museums
» Hospital
» e-Display
Ad Hoc» Wildfire
» Volcano
Concluding Lessons
43Mobil Computing© 2010-2013 Daniel P. Siewiorek
Interactive Museums [Fleck]
San Francisco Exploratorium» Several hundred interactive exhibits
» Frequently rotated off the floor
Prior work» PDA, Acoustic Guides, IR beacons for pointing
44Mobil Computing© 2010-2013 Daniel P. Siewiorek
Interactive Museums - Services
Informer – detailed information on exhibit Suggester – what to try Remember – build record of experiences, selects what to
record Guider – suggests an order of exhibits* Communicator – instant messaging, leave notes at exhibits*
* Not implemented
45Mobil Computing© 2010-2013 Daniel P. Siewiorek
Interactive Museums – Lessons Learned
Overall positive responses Not enough hands to hold PDA and operate exhibit
interfering with exploration Undesirable demand on user attention Lost in hyper reality – focus on device rather than
exhibit Wow factor – part of positive feedback Beacons OK but sometimes picked up other beacons Browser interface – people unfamiliar with stylus Content design – did not know what was clickable Forgetting to use Remember – application complexity
46Mobil Computing© 2010-2013 Daniel P. Siewiorek
Pervasive Technologies in a Hospital [Hansen]
Hospital scheduling, coordination, and awareness system
» Location tracking, context awareness, large interactive displays, mobile phones
Awareness Media» Status of work in Operating Rooms
» Video stream for accessing current state
» Progress bar
» Chat area
» Schedule
» Location tracking of who is in Operating Rooms
47Mobil Computing© 2010-2013 Daniel P. Siewiorek
Pervasive Technologies in a Hospital - Scenario
Acute patient Find/schedule Operating Room Find surgeon not in Operating Room Send message Notify patient ward Scheduled patient notified of their surgery
postponement
48Mobil Computing© 2010-2013 Daniel P. Siewiorek
Pervasive Technologies in a Hospital – Hardware
Lessons
49Mobil Computing© 2010-2013 Daniel P. Siewiorek
Pervasive Technologies in a Hospital – Software
Lessons
50Mobil Computing© 2010-2013 Daniel P. Siewiorek
Pervasive Technologies in a Hospital – User Lessons
51Mobil Computing© 2010-2013 Daniel P. Siewiorek
e-Campus Display [Storz]
Three deployments
Conference Signage
Art Center
Underground Bus Station
52Mobil Computing© 2010-2013 Daniel P. Siewiorek
e-Campus Display [Storz]
Conference Signage» Dynamic schedule of time/location of conference events
Art Center Exhibition» 60 year celebration of VE day
» Three displays, video diary, web based diary, interactive artifact exhibit
» News footage, images of artifacts, visitors record own memories
Underground Bus Station» Three screens
» Artistic material, textual information, videos
» Context triggered
53Mobil Computing© 2010-2013 Daniel P. Siewiorek
e-Campus Display – Lessons Learned
54Mobil Computing© 2010-2013 Daniel P. Siewiorek
e-Campus Display – Lessons Learned Details – Part 1
Deployments are costly» Time, money, is there a cheaper way to get the same results
Environmental Challenges can be significant» Fire alarm tests and power cycling, diesel fumes clogging
projector air filter
After deployment comes maintenance» Underpass projector failure required road closure, insure
accessibility. Shadow system in lab
Follow the rules» Accessibility for people with disabilities. Work with local site
manager
See what public sees» Remotely monitor what public perceives. Add cameras to
view projector output
55Mobil Computing© 2010-2013 Daniel P. Siewiorek
e-Campus Display – Lessons Learned Details – Part 2
Build white boxes, not black» Be able to monitor whole system state (what displayed
where, when) rather than just observing projected images
Content is king» Content is the system. Considered late in cycle led to
mismatch between content support requirements and what was provided
Content is expensive» Generating compelling content is nontrivial requiring
specialist skills and domain knowledge
Manage your assets» Difficulties in managing content once it is created including
encodings, archiving, previewing, approval
56Mobil Computing© 2010-2013 Daniel P. Siewiorek
e-Campus Display – Lessons Learned Details – Part 3
Define the user experience» Precisely define user experiences – carefully coordinated
performances
Provide transactions» Transaction level atomicity – because multiple components
involved there should be no partial transitions
Manage expectations» Testing confused the viewing public. Communicate access
policies. Blank displays create expectation of content to come
Be accountable» Entering public domain makes work subject to increased
levels of scrutiny
57Mobil Computing© 2010-2013 Daniel P. Siewiorek
Outline
Introduction Individual
» Wearable Computers
» Augmented Reality
Infrastructure» Museums
» Hospital
» e-Display
Ad Hoc» Wildfire
» Volcano
Concluding Lessons
58Mobil Computing© 2010-2013 Daniel P. Siewiorek
Monitoring Weather Conditions in Wildland
Fires [Hartung] Predictions based on current conditions,
previous weather, and predicted weather Thermal belts, temperature inversions
» Cold air moves into valleys forcing war air to rise which is trapped by continuously moving air above ridge tops
» Fire stays more active in warmer air with lower relative humidity
Current approach» Manual Kit – 5 to 10 minutes to collect at current location
each hour
» Remote Automated Weather Stations – measure temperature, wind speed/direction, relative humidity, precipitation, barometric pressure, fuel moisture/temperature, soil moisture
59Mobil Computing© 2010-2013 Daniel P. Siewiorek
Monitoring Weather Conditions in Wildland
Fires -Background
Inversion Layer
Manual Measurement
60Mobil Computing© 2010-2013 Daniel P. Siewiorek
Monitoring Weather Conditions in Wildland
Fires -Architecture
61Mobil Computing© 2010-2013 Daniel P. Siewiorek
Monitoring Weather Conditions in Wildland Fires
– Lessons Learned 1 Fire activity changes rapidly with even small
changes at low Relative Humidity Conform to existing infrastructure/logistics
» Powered by AA batteries – quickly accepted since nearly all electronics devices operate by crews used AA batteries
Battery temperature has significant effect on battery power available
62Mobil Computing© 2010-2013 Daniel P. Siewiorek
Monitoring Weather Conditions in Wildland Fires
– Lessons Learned 2
Housing vents and open bottom» Allows airflow but protects from rain
» Prevent heat trapping
» Painted white to protect from radiation heating
» Mount 1.8 meters above ground to protect from heat damage caused by fire
Large change in elevation increased radio range by 4 to 10 times
» Further from “ground” less interference
» Fresnel Zones - measure of phase difference between reflection of waves between transmitter and receiver that cause cancelling effect
63Mobil Computing© 2010-2013 Daniel P. Siewiorek
Air Dropped Sensors for Volcano Monitoring [Song]
Goals» Synchronized sampling – time synchronized within 1 msec
» Real Time continuous raw data
» One year robust operation – harsh conditions including heavy rain, snow, ice, wind gusts to 120 mph
» On-line configurable
» Fast deployment – air drop sensors (70 pounds), network self starting and organizing
Sensors» Seismic
» Infrasonic
» Lightning – RF pulse detector
64Mobil Computing© 2010-2013 Daniel P. Siewiorek
System for Volcano Monitoring
65Mobil Computing© 2010-2013 Daniel P. Siewiorek
Air Dropped Sensors for Volcano Monitoring -
Lessons Signal strength does not reflect signal quality,
rather signal to noise ratio Problems with connectors System data delivery 91.7% Station tipped over, damaged voltage regulator Infrasonic – pressure sensor also recorded
heavy wind gusts Outage for 20 hours since data base could not
handle daylight savings time Large data loss due to low battery voltage
66Mobil Computing© 2010-2013 Daniel P. Siewiorek
Outline
Introduction Individual
» Wearable Computers
» Augmented Reality
Infrastructure» Museums
» Hospital
» e-Display
Ad Hoc» Wildfire
» Volcano
Concluding Lessons
67Mobil Computing© 2010-2013 Daniel P. Siewiorek
Common Lessons Learned
Demands on User attention Risk of user becoming engaged in virtual world
and become less aware of physical context Mobile users impatient Strong interaction mental model Fit into current procedures/logistics Environmental influences Power especially battery characteristics