Place Lab: Enabling LargeScale, LocationEnhanced Computing Ian Smith

1
Place Lab Enabling Large-Scale,
Location-Enhanced ComputingIan Smith
Anthony LaMarca Yatin Chawathe, Sunny Consolvo,
Jeffrey Hightower, James Scott (IRC) November
11, 2004
2
Why am I here?

• Not headed for a Camus answer
• Let you know whats going on at Intel Research
  Seattle
• Encourage potential interns to apply

Indeed, why am I here, when my copy of Halo2 is
in Seattle?!?
3
What do you want to enable?

• Widely available location-enhanced computing
• Build a positioning system that is
• Wide-scale, indoor outdoor
• Can be used everywhere
• Privacy observant, low barrier to participation
• Can be used by everyone

4
What new ideas enable virtually free device
positioning?

• Exploit wide-scale WiFi deployment
• Urban areas have dense WiFi coverage
• WiFi base stations broadcast unique IDs
• Position WiFi devices using map of
  base-station-IDs to location

5
Demo
6
How will you go about it?Place Lab Toolkit

• Develop toolkit to foster external engagement
• XP, MacOS and Linux notebooks, PocketPC PDAs
• Recently Nokia Series 60 phones
• Sample client- and server-side apps
• 24/7 deployment of Mapper service on PlanetLab
• War-driving devices based on personal servers
• A plugin architecture to encourage
  experimentation
• Signed-up for the alpha release
• UW, UCB, UCSD, GaTech
• IRB, IRC
• Edinburgh festival

7
What is the meat of the project?1 Scanning RF
beacons in the wild

• Problem Signal strength not sufficient to
  accurately map base station location
• Visibility is another key feature..
• Solution Add contextual information to improve
  accuracy
• Apply techniques developed at IRS for correcting
  GPS data
• Mine GIS maps for structural information, e.g.,
  roads, buildings
• Incorporate models of obstacles and antennas

t4 56 db
t3 40 db
t1 95 db
t2 46 db
8
What is the meat of the project?1 Scanning RF
beacons in the wild (cont)

• War drivers will not map all base stations
• Leverage users RF traces to complete the map
• Users can opt to submit traces of base stations
  they see
• (t1,b1) , (t2,b2) , (tn, bn)
• Refine base station locations using these traces

9
What is the meat of the project?1 Scanning RF
beacons in the wild (cont)

• Recently weve incorporated wigle.nets data
• We dont have redistribution rights
• Gives us significant coverage in US and Western
  Europe (about 2M access points)
• War driving is not a crime

10
Wigle.net Coverage Map
11
What is the meat of the project?2 Mapper Service

• Distribute for scalability and trust
• 1 server hour 1 hr of war-driving or 20 hrs of
  steady-state traces
• Avoid single trusted entity
• Four key problems
• Routing AP trace data from clientsto servers
• Retrieving location maps from distributed servers
• Detecting misbehaving servers
• Detecting bad data submitted by clients

AP sightings
Mapper Service
Regional location maps
12
What is the meat of the project?3 Building a
critical mass of data

• Incorporate data from existing sources
• Wigle, WiFi clubs (NYC wireless)
• Distribute personal server spotters to research
  community
• Compact, long-lived war-driving tool
• 50 devices, 20 with GPS

13
What is the meat of the project?4 Designing the
user experience
Pilot StudyFall 04
Research Trajectory
14
What is the meat of the project?4 Designing the
user experience (ESM)

• Objective Assess attitudes towards location
  disclosure to social relations
• 16 participants, Seattle area, no tech. careers
• All got out of house daily, all were cell-phone
  users
• Phases
• Pre-study interview (assess social network)
• 2 Weeks of ESM questions (max 10 day, 9am-9pm)
• Post-study interview
• ESM questions were customized per-subject

15
What is the meat of the project?4 Designing the
user experience (ESM)

• Participants want to disclose what they felt was
  most useful to the requestor
• Recipient design (place rather than address)
• No evidence of blurring for privacy (but blur for
  utility)
• 77 of the time, something specific was disclosed
• Westin privacy classification is not a good
  predictor
• Westin concerns disclosure to business, so not
  surprising
• Highest comfort level with spouse/SO
• All of this was hypothetical, so

16
What is the meat of the project?4 Designing the
user experience (Reno)

• People share their location with their social
  network frequently
• Mobile phones
• SMS/Email/IM
• Postcards
• Coming soon Social location disclosure apps?
• Can we build a location-sharing application that
  is compelling and privacy sensitive?

Hello? Hey, its Ian.Where are you?
drianesmith Hey, Im at home. Can you pick me up
on your way?
17
What is the meat of the project?4 Designing the
user experience (Reno)

• Do it on a device that is really ubiquitous
• Battery life, form factor, etc.
• Device should position itself
• Many options GPS/A-GPS, Place Lab, etc.
• Use place instead of location
• Follows directly from Summer study about location
  disclosure
• Dont do anything worse with privacy
• If they are already using a GSM cell phone

18
What is the meat of the project?4 Designing the
user experience (Reno)
19
(No Transcript)
20
What is the meat of the project?4 Designing the
user experience (Reno)

• Location information is a low-bandwidth channel
• It can be extremely powerful in the presence of
  shared context
• Location, path, activity

Late tonight, f pushed her location to me from
SEA-TAC airport. That reminded me that I
wanted get together with them socially when they
arrived. I had forgotten it was this
weekend. Participant g, Day 2 (Thurs)
21
What is the meat of the project?5 Moving from
location to place
(Hightower)

• Solving this problem requires
• Discovering and mapping boundaries of place
• Learning place labels
• From the WWW
• Mine place boundaries using GIS maps
• Extract place labels from GIS, Yellow Pages,
• From the RF beacon data
• Infer place boundaries from behavior and usage
• From the user
• Direct entry of place labels

Café
22
Thank You.
23
Deliverables
Place Lab Toolkit
Deploy Place Lab in university classes
Publish evaluation of academic use of Place Lab
Release toolkit with location-to-place service
Release toolkit with ULF or location stack
Release Place Lab Architecture
Mapper Service
Demo client/server and server/server reputation
algorithms
Deploy distributed Rev 2 Mapper service
Deploy centralized Rev 1 service
Deploy PS GPS spotter
Capstone Application
Publish evaluation of capstone application
Release application design and architecture
Deploy application to real users for evaluation
Define user requirements
Mapping / Learning
Demo beacon placement inference
Demo location-to-place service
Demo data-cleaning filters
Demo Arrangements layer operation primitives
Privacy
Generate user requirements for trace gathering
Report on tradeoffs of explicit vs. implicit
privacy management
Demo explicit mechanisms for privacy management
of location data
2003
2004
2005
Dec 03
24
The Three Major Aspects of Place Lab

• Build a software artifact for doing location
• Get others using and developing Place Lab
  applications and extensions
• Within Intel
• Within the research community
• Within the broader hacking community
• Perform research in location-enhanced computing
• Privacy
• Distributed systems
• Machine learning
• Applications

25
Develop Place Lab toolkitLaMarca, Chawathe, I
Smith, J Scott

• Java (J2ME)
• Many supported platforms
• 802.11, Bluetooth, GPS, GSM spotting
• Approximately 500 downloads/month
• Done
• Rev 1.0 released 1/04
• Rev 1.1 released 3/04
• To Do
• Rev 2.0 late summer 04 with Nokia support

26
Download Statistics
27
Populate beacon mapping databaseChawathe,
Powledge, Rea

• 28475 mapped WiFi beacons as of July 04
• Largely in US cities (Seattle, Berkeley, Boston,
  etc)
• Done
• Deployed non-distributed Rev 1 mapper service
• Developed Stargate Spotter, Place Lab stumbler
  software
• To Do
• Deploy Stargate Spotters to collaborators
• Convince war drivers to submit logs
• Investigate ways to drastically increase database
  size
• Contest
• Instrument public-service vehicles
• Key Collaborators
• T. Pering (IR)

28
Distribute beacon mapping databaseChawathe,
LaMarca, Smith

• Distribute via DHT using Prefix hash Trees (PHT)
  for indexing
• Done
• Developed 2D extensions to original PHT
• Measured performance using realistic data set
• Submitted paper to OSDI 04 (rejected)
• To Do
• Submit joint IRB/IRS PHT paper to NSDI 05
• Develop and test concurrency control measures for
  PHTs
• Replace public database with distributed version
• Key Collaborators
• S. Ratnasamy (IRB), S. Shenker (ISCI), J.
  Hellerstein (IRB/UCB)

29
Foster use of Place LabLaMarca, I Smith, Chawathe

• Place Lab usage in both research and teaching
  venues
• Seeing uptake beyond collaborators and friends
• Done
• Demoed Prototype system at UbiComp 03
• Taught 10 half-day Place Lab boot camp
  workshops
• Dartmouth, UW, Trinity, UCSD, ICL, Glasgow,
  Strathclyde, Cambridge
• Supported UW and Dartmouth use of Place Lab in
  the classroom
• Open source development of Place Lab on Source
  Forge
• Prepared demo for Google Place Bar
• To Do
• Continued support of class, academic use of Place
  Lab
• Increase engagement with relevant Intel groups
• Continue and increase support of Kelvin institute
  in Edinburgh deployment

30
Mapping and positioning algorithmsChawathe,
LaMarca

• Base toolkit implements centroid and simple
  particle filter based location calculation
• Done
• Preliminary data suggests Place Lab 1.1 accuracy
  of 30M
• To Do
• Comprehensive study of existing WiFi location
  algorithms aka Bake Off Submit to Pervasive
  05
• GPS-less beacon inference
• Compare Voronoi-based approach to current
  unconstrained 2D approaches
• Decide how best to cope with z dimension
• Investigate multiple-radio algorithms (802.11
  Bluetooth, etc)
• Key Collaborators D. Fox (UW CSE), J. Krumm
  (MSR)

31
Designing privacy-observant proactive
applicationsConsolvo, I Smith

• Classes of application personal, social,
  institutional
• Done
• Develop plan for Seattle-based ESM study of
  social applications
• Develop plan for Edinburgh festival-based study
  of institutional applications
• UbiComp workshop paper on location disclosure by
  app. class
• To Do
• Execute study of social app, submit study results
  to CHI 05
• Execute study of institutional app, submit study
  results to CHI 05
• Fold study results into design of upcoming Place
  Lab applications
• Key Collaborators S. Mainwaring (PAPR)

32
Create a privacy license agreementI Smith,
Consolvo

• Combine end-user privacy guarantees and open
  source into a new license agreement
• Release Place Lab under this agreement
• Done
• Partially complete certificate of origin
• Rough draft of agreement with Intel Legal
• To Do
• Submit value-sensitive design paper to ECSCW 05
• Key Collaborators B. Friedman UW I School), P.
  Kahn (UW Psych)

33
Develop Place Lab applications

• No real applications developed by the team to
  date
• Done
• Copious brainstorming about possible capstone
  application
• Small sample applications in each of the 3
  identified categories
• Pedometer personal
• Ambush social
• Google Place Bar institutional
• To Do
• Revisit capstone ideas and evaluate in light of
  user study data
• Build and deploy and evaluate a capstone
  application

34
Publication Plans
35
Break
36
Deliverables
Place Lab Toolkit
Deploy Place Lab in university classes
Publish evaluation of academic use of Place Lab
Release toolkit with location-to-place service
Release toolkit with ULF or location stack
?
Release Place Lab Architecture
?
?
Mapper Service
Demo client/server and server/server reputation
algorithms
Deploy distributed Rev 2 Mapper service
?
Deploy centralized Rev 1 service
Deploy PS GPS spotter
?
?
Capstone Application
Publish evaluation of capstone application
Release application design and architecture
Deploy application to real users for evaluation
?
?
Define user requirements
?
Mapping / Learning
Demo beacon placement inference
Demo location-to-place service
Demo data-cleaning filters
Demo Arrangements layer operation primitives
Privacy
Generate user requirements for trace gathering
Report on tradeoffs of explicit vs. implicit
privacy management
Demo explicit mechanisms for privacy management
of location data
?
?
2003
2004
2005
37
What has gone well

• Toolkit development
• Achieved sought-after portability
• Complimented from users about usability
• Development of research agenda
• Engagement of top-shelf collaborators
• PHT surprising, interesting result
• Recruitment of J Scott to project
• Strong UW collaboration/contribution

38
What has not gone well

• Building out the database has been slow
• Less WD engagement than expected
• Less background mapping than expected
• Support for CLDC 1.0 (Nokias) has required more
  engineering that expected
• APIs not as simple as before
• Bake-off, privacy agenda and application
  development have lagged expectations
• Key resource (Sunny) redirected for Q1 this year
• Technical and methodological problems with our
  data collection
• No individual or entity on trajectory to inherit
  Place Lab

39
Changes to our plans

• Nokia support give us a compelling platform
• Develop and deploy Ambush on phones (J Scott, I
  Smith)
• Investigate accuracy/coverage of heterogeneous
  beacon-based location systems (J Scott, LaMarca)
• Drop study of academic use of toolkit (Consolvo)
• Investigate PHTs and their cousins beyond the
  requirements of Place Lab (Chawathe, LaMarca)

40

• Thanks!

41
Place Lab Users

• UCSDs Active Campus Project
• Trinity College, Dublin
• University of Washington CSE
• GUIRs Topiary
• Kelvin Institute