Deborah Estrin

1
Embedded Networked Sensing from Eco-systems to
Human-systems

• Deborah Estrin
• destrin_at_cs.ucla.edu
• Work summarized here is that of students, staff,
  and faculty at CENS
• We gratefully acknowledge the support of our
  sponsors, including the National Science
  Foundation, Nokia, Intel Corporation, Cisco
  Systems Inc., Crossbow Inc., Agilent, Microsoft
  Research, Sun Inc., and the participating
  campuses.
• http//research.cens.ucla.edu

2
Motivation

• Many critical issues facing science, government,
  and the public call for high fidelity and real
  time observations of the physical world
• Networks of smart, wireless sensors can reveal
  the previously unobservable
• Designing physically-coupled, robust, scalable,
  distributed-systems is challenging
• The technology will also transform the business
  enterprise (from inventory to manufacturing), and
  human interactions (from medical to social)

3
Embedded Networked Sensing
Embed numerous devices to monitor the physical
world Network to coordinate and perform
higher-level identification Sense and actuate
adaptively to maximize information return
In-network and multi-scale processing algorithms
to achieve Scalability for densely deployed
sensors Low-latency for interactivity,
triggering, adaptation Integrity for challenging
system deployments
4
Environmental Monitoring Observatories Technology
History and Themes Field Inspired Systems
Research Participatory Sensing
5
ENS Observatories
Terrestrial
Seismic
createprogrammable, distributed,
multi-modal, multi-scale, multi-use
observatories to address compelling science and
engineering issues and reveal the previously
unobservable. From the natural to the built
environment From ecosystems to human systems
Contaminant transport
Aquatic
6
Environmental monitoring applications spatial
variations and heterogeneity
Precision Agriculture, Water quality management
Impact of fragmentation on species diversity
Earth structure inhomogeneities
Algal growth as part of eutrophication
7
Visualization and Navigation through ENS Space

• Example Keyhole/Google Earth as one approach
  towards navigation, visualization, data sharing,
  and attracting a community of users via the Web.

8
NIMS RD Merced and San Joaquin River Confluence
confluence
Sonar-based bathymetry (depth)
Kaiser, Harmon, et al
9
Data from Mexico Seismic Array Pakistan
Earthquake
Davis, Guy, Husker, Lukac, et al
10
Proposed NEON replicated array deployments
Graphics by Jason Fisher
11
Environmental Monitoring Observatories Technology
History and Themes Field Inspired Systems
Research Participatory Sensing
12
Technology challenges
Objectives
Constraints

• Embeddable, low-cost sensor devices
• Robust, portable, interactive systems
• Data integrity, system dependability
• Programmable, transparent systems
• Multiscale sensing and actuation

• Sensing channel uncertainties
• Environmentally compatible deployment
• Limited resources node, infrastructure
• Complexity of distributed systems
• No ground truth

13
Decade of Wireless Sensor Node Developments
LWIM III UCLA, 1996 Geophone, RFM radio, PIC,
star network
AWAIRS I UCLA/RSC 1998 Geophone, DS/SS Radio,
strongARM, Multi-hop networks
Mica Mote UCB, 2000 RFM radio, Atmel
Telos Mote UCB, 2004 802.15.4 TI processor
LEAP Node/ENS-boxUCLA, 2006 TIMSP430
w/802.15.4 AND XScale PXA255 w/802.11
14
Informal history (previous version)
Early history (wired) Distributed Tracking,
Industrial Monitoring, Ubiquitous computing DARPA
DSN, Tracking, Sensor Fusion, Industrial
monitoring ORL, Xerox PARC, MIT Media lab, IBM,
HP, MSR
Past decade of wireless sensor networks research
programs DARPA LWIM and AWAIRS (UCLA)
94-96 Smart Dust paper 96 IETF MANET WG (ad
hoc routing), 802.11 WG ISAT Simple Systems
study 98 Actuated sensing ACFR, CMU, USC, ISAT
Robotic Ecology DARPA SenseIT 99 USC/ISI,
Cornell, Xerox, UCB, BBN, Penn State, Univ Ill.,
MIT NRC Embedded Everywhere 00, TOS paper
00 Intel Berkeley lablet, Startups Crossbow,
Ember, Dust, Sensicast DARPA NEST 01 UCB
TinyOS, Ohio State, Univ Virginia, MIT,
Intel/Xbow International Australia,
Switzerland, UK, Korea, Singapore. NSF ITRs,
CENS STC(UCLA-USC), CITRIS (UCB) NSF Sensors
and sensor networks NETS/NOSS Industrial RD
MSR, Nokia, IBM, HP, PARC, Motorolla, Sun,
Agilent, Intel
Sigcomm
SOSP/OSDI
TOSN
IPSN
Sensys
Ubicomp
Mobicom
Mobihoc
Emnets
DCOSS
ICASP
Secon
15
A Walk Through History
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
Srivastava, et al
16
Status Many first generation hw/sw system
components exist
Localization Time Synchronization
System Mgmt
In Network Processing
Multiscale
Power on demand
Event Detection
Routing and Transport

• Reusable, Modular, Flexible, Well-characterized
  Services/Tools
• Routing, Reliable transport, Plug and Play
• Time synchronization, Energy Harvesting, Power on
  demand, Localization, Self-Test
• In Network Processing Tasking, Filtering,
  Triggering, Fault detection, Multiscale
  coordinated / actuation
• Simulation, Testbeds, Programming Abstractions,
  Application authoring tools, embedded statistical
  tools

17
Lessons thus far...
Early themes Thousands of small devices Minimize
individual node resource needs Exploit large
numbers Fully autonomous systems In-network and
collaborative processing for longevity
optimize communication

• New themes
• Heterogeneity
• Tiered systems architecture to optimize system as
  a whole
• Inevitable under-sampling (in time or space) with
  homogeneous sensing
• Exploit multiple modalities, multiple scales, and
  mobility
• Interactivity
• Coupled human-observational systems online
  tasking, analysis, visualization
• In-network and collaborative processing for
  responsiveness, data quality, rapid and
  iterative deployment
• Monitoring the monitors calibration, self
  test, validation

18
Environmental Monitoring Observatories Technology
History and Themes Field-inspired Systems
Research Participatory Sensing
19
Systems Deployments to Instruments

• CENS systems address challenges of field-deployed
  networks
• Difficult environments, challenging connectivity
  issues, heterogeneity
• Requires center scale to integrate algorithms,
  systems, science

20
Three research thrusts that emerged from the field

• Heterogeneous, and mobilized, sensing systems
• Rapid, iterative and robust deployment tools
• Exploit context to turn image/acoustic inputs
  into biological sensors

21
Heterogenous and Mobilized Sensing Systems
22
Heterogeneity is key to deployed systems and
the field as a whole

• Several classes of systems
• Mote herds Scale
• Collaborative processing arrays Sampling rate
• Networked Info-Mechanical Systems Autonomy
• Achieve longevity/autonomy, scalability,
  functionality with
• heterogeneous systems
• in-network processing, triggering, actuation
• Optimize across the system as a whole

lifetime/autonomy
Mote Clusters
Infrastructure- based mobility(NIMS)
scale
Collaborative processing arrays (imaging,
acoustics)
sampling rate
23
Tenet Tiered system software architecture

• Goal Rapid sensor network deployments
• Quick application changes, adaptation to field
  circumstances
• Architecture A constrained environment for
  developing applications
• Optimization only when required, more reusable
  pieces
• Example Internet
• Tenet architecture takes advantage of more
  powerful network nodes

Govindan, Kohler, et al
24
CentRoute Centralized Routing for Mote Networks

• CentRoute centralized on-demand routing protocol
  for motes
• Exploits heterogeneity by centralizing routing
  decisions on resource-rich microservers
• Reduces memory requirements and control overhead
  on motes
• Eliminates routing instabilities and loops
• Provides global view of the network at each sink

Very low control overhead at medium and high
densities
Very good network connectivity at medium and high
densities
Stahopoulos, Heidemann, et al
25
Deployment modes offer different information
return and tradeoffs

• Other key tradeoffs
• Temporal vs Spatial density
• Temporal density vs. extent
• Sensible features vs. density

Automated Mobility
Spatial Density
Static
Handheld Mobility
Remote Sensing
Spatial Extent
26
Mobility/Actuation is an important dimension of
heterogeneity

• While ENS is a revolutionary technology for dense
  sensing
• the likelihood of under-sampling critical
  phenomena is surprisingly high
• meeting sampling objectives is sometimes
  impractical with static nodes
• Mobility is a critical amplifier of system
  coverage, from highly constrained articulation,
  to longer range spatial traversals.
• Articulation magnifies effective sensor range and
  spatial diversity
• Infrastructure-supported mobility (NIMS) enables
  sensor diversity
• Enables adaptive, fidelity-driven, 3-D sampling

Networked Info Mechanical Systems (NIMS)
Kaiser, Pottie, et al
27
NIMS in the field
Deployments James Reserve Phenology, Wind River
Canopy Crane, Public Health Media Creek
Future Development 3-Dimensional, Portability
and rapid deployment
Kaiser, et al
28

Another essential tier/modality/scale the user

• Whereas we focused initially on very long lived,
  autonomous systems design, interactive and rapid
  deployments are high value.
• Interactive systems take advantage of human
  observer, actuator
• Addresses critical issues such as adaptive
  sampling, topology adjustment and faulty sensor
  detection
• Requires real time data access, model based
  analysis, and visualization in the field

Daily Average Temperature (Geostatistical
Analyst) Aspect (Spatial Analyst) Slope (Spatial
Analyst) Elevation (Calculated from Contour
Map) Aerial Photograph (10.16cm/pixels)
Coupled Human-Observational Systems transform
physical observations from batch to interactive
process
29
Rapid, Iterative, and Robust Deployment
30
Rapid Deployments (RD)

• Systems deployed many times for short durations
• Powerful usage model for environmental assessment
• trade temporal for spatial density and coverage
• Short deployment duration enables
• Frequent calibration and maintenance
• User presence gt increased functionality, data
  quality
• Need for rapid setup, data return

31
Toward Confident Deployment Bangladesh case
study

• Rapid deployment in Bangladesh to evaluate
  relationship between irrigation and arsenic
  contamination
• Soil pylons, ion selective electrode sensors to
  measure ions that correlate with arsenic
  (ammonium and calcium)
• 2 months of preparation preceding a 12-day
  deployment
• Data quality/quantity issue data often out of
  range--after the fact it has to be discarded in
  the field it could have been checked/calibrated

Ramanathan, Kohler, Jay, et al
32
System robustness and data quality require field
tools

• Develop fault model for use in field
• Detect and react to faults
• Take advantage of system architecture to adjust
  parameters
• Take advantage of human interaction to e.g.
  replace batteries, check sensor coupling
• Simple end-to-end model
• Classify faults based on user action
• Builds on prior experience with Sympathy, system
  for detecting network faults
• Working toward full architecture for writing
  flexible sensor network applications, tuning them
  in deployment, and recovering from faults as they
  occur

Ramanathan, Kohler, Jay, et al
33
Other RD challenges Experimental design
(placement, coverage)

• The traditional coverage problem is overly
  idealized for messiness of real world
• Radio location vs. sensor location (not
  equivalent, not of equal priority, not one to
  one)
• Physical medium constrains sensor placement
  (rocks, trees, etc.)
• Minimum separation b/w sampling sites closely
  placed sensors can disturb phenomenon, and
  interfere with one another
• Experimental design dilemma
• Build constraints into design tool/model
• Adapt idealized model while in the field.

Schoellhammer, Hansen, Graham, et al
34
Experimental Design (CAD) tools/techniques needed

• Pre-Deployment
• models that integrate previously captured data
  and physical models
• simulation to test design strategies
• In-Field
• capture of deployment meta-data, adjustments,
  justifications
• models to inform deployment adjustments as the
  deployment unfolds
• Integrate with calibration, data integrity tools
• Post-Deployment
• meta-data to clean data as it arrives from the
  field

Schoellhammer, Hansen, Graham, et al
35
Increasing role of statistical models and methods

• Experimental design and sensor layout adaptive,
  iterative schemes for deployment
• Botanical gardens microclimate system design
  Source of variability via PCA, Optimization via
  ILP
• Palmdale soil observation network design
  Geospatial statistical methods for optimal sensor
  placement

• Data integrity robust procedures for aggregation
  and analysis
• Fluorometer measurements at lake Fulmor Running
  medians

• Spatio-temporal models flexible or
  nonparametric descriptions of signals
• Media creek nitrate studies spline based
  estimators

• Opportunistic measures identifying and
  integrating existing sources of data from other
  engineered systems
• Elevator tracking for structural health
  monitoring wavelet coherence

Hansen, et al
36
Imagers and acoustics as biological
sensors systems designed to exploit context
37
Capturing High-Frequency Data

• Tenet-like application for sampling acoustics,
  neural spikes
• First use of motes for flexible acoustic
  collection
• Simple filters save energy, reduce congestion
• Master updates filters dynamically to meet user
  requests and sampling conditions

Before filtering
Convolution filter
Amplitude gate Isolated signal!
Greenstein, Kohler, et al
38
Cyclops mote based imager

• A vision sensor that mates with Mote class
  devices
• Large Numbers
• Information about the statistics of the
  experiment
• Minimum Infrastructure
• Diversity of pose, distance, angle
• Applications
• Occluded environments
• Local observations in Large space

Agilent Technology
Rahimi, Srivastava, et al
39
Leverage context to apply on-board processing to
the application
Ecology and Agriculture
Spectroscopic, size, shape analysis

• Plant species studies phenology, fruiting
  conditions, trends, timing
• Animal species studies birds and reptiles
• Pitfall Traps measure population of reptiles,
  timely animal identification and notification for
  tag and release.
• Bird nestbox measure distribution of occupancy,
  occupancy vs. time of day and condition of the
  nest, number of eggs/young
• LED as flash for night images Infrared for birds

June 2006
August 2006
Ahmadian, Rahimi, et al
40
Environmental Monitoring Observatories Technology
History and Themes Systems Research Participato
ry Sensing From Ecosystems to Human Systems
41
Participatory Sensing

• ENS is revealing the previously unobservable in
  science applications
• Multi-scale data and models to achieve context,
  and in network processing and mobility to achieve
  scalability (communication, energy, latency)
• Automatically geocoded and uploaded participatory
  sensing data promises to make visible human
  concerns that were previously unobservableor
  unacceptable
• Urban sensing applications will leverage the
  millions of cell phone acoustic, image and
  bluetooth-connected sensors
• Internet search, blog, and personal feeds, along
  with automated location tags, to achieve context,
  and in network processing for privacy and
  personal control

?
?
42
Technical themes draw from sensornets and
internets

• Explore system needs and opportunities
• Multiscale sensing and actuation to achieve
  Coverage
• In-network processing
  to support Privacy
• Analysis and visualization to enable Discovery
• Define architectural elements and interfaces
• Sensor Observe, capture, forward
• Network Name, verify, tag with context
• Fabric Filter, search, store, disseminate
• Application Explore, task, re-present

43
Range of Application Types

• Directed Sensing Applications
• Eco-PDA (space/time-tagged annotation)
• Self-administered health diagnostics
  (auto-upload, verify context)
• Public/community health (spatial interface to
  data, data-gathering-protocol authoring)
• Citizen Sensing
• Participatory urban planning
• Place-aware social networking
• Distributed documentary - journalism
• Community-built histories, the new local
  library
• Enabling Elements
• Radio/online
• Local processing
• Connected instruments
• Image and acoustic as metadata

44
Common Application Style Observation Campaigns

• Real urban examples of citizen concerns (web
  based)
• Bicycling to work lack of adequate facilities
• Cell phone use in cars
• Does red light photo program work
• Fallen (public) fruit (fallenfruit.org)
• Impact of lack of sidewalks
• Items sold to children that resemble real bad
  objects
• Lawn estimated time-to-death without water
• Mobile phone Amber Alert (codeamber.org)
• Neighborhood maintenance, visible decay

Partisan targets Noise levels in different types
of locations Traffic at intersections (light
timing, stop signs) Flooded storm drains
Violations of carpool lanes Park or street
maintenance issues (uneven sidewalks) Public
transportation stop occupancy in LA Power outage
documentation scope time (05-1914)Speed
humps slowing traffic in neighborhoods
(04-1281-S2) Timelapse collage of a
location Water quality measurements (photograph
simple indicators)
Numbers in parentheses are LA City Council file
numbers.
Burke, Hansen, Srivastava, Parker, Redi, et al
45
Example of Directed Sensing
Nokia N80
45
Srivastava, Parker, Redi, et al
46
Campaign mechanics

• Post a campaign request
• Issue / problem statement
• Type of data needed
• Sampling density, extents, other parameters
• Geographic and temporal limits
• Wait for people to agree to contribute
• Offer coverage to take samples
• Offer availability to classify / verify samples
  if necessary
• Opt-in to submit location, receive SMS msgs
  triggering sampling
• Campaign executes
• System listens to published locations of
  citizen-sensors
• Trigger sampling according to geographic
  temporal coverage needs
• Adjust windows, triggers (via SMS) to achieve
  coverage
• Pass samples to distributed analysts who
  verify/classify
• Accept and post (map, visualize) results
• Closure

Burke, Hansen, Srivastava, Parker, Redi, et al
47
Motivations to participate

• Network vouches for the context.
• Organized use by community partners
• Individual agreement with / interest in issues
• Disagreement with past campaign
• Gaining ability to post Challenge of ones own.
• Simple API open intrinsic capabilities of the
  framework to mashups
• Converge entry / estimation / management
• Opportunistic triggering of sampling
• Distributed classification / verification
• Online tools for analysis

Burke, Hansen, Srivastava, Parker, Redi, et al
48
Concerns about participation Privacy

• Need for personal configuration and control of
  shared data
• Close to the sensor source not on the backend
• Lessons from microdata release Resolution
  control, blurring, subsampling, local buffering
  and filtering
• Guidelines for Privacy and Selective Sharing
• Context of data should be verifiable to a
  resolution with which provider is comfortable
  and as needed by application
• Policies for selective sharing should be
  implemented as an automated component of a
  sensing system.
• Decisions about data sharing depend often on
  location and time.
• HCI for configurability of privacy/security
  policies is critical (Bellovin)
• Data Integrity also matters
• Verify geocoding
• Corroborate sensor data

Burke, Hansen, Srivastava, Parker, Redi, et al
49
Sensing Contexts Matter

• Private Citizens, Private Spaces
• Personal applications
• Data strictly personal and citizens
  expect privacy (health monitoring)
• Social applications
• Share data with a small circle of
  friends (Flickr)
• Urban applications
• Citizens share data as part of city or
  state-wide project (blogs)
• Private Citizens, Public Spaces
• Monitoring the public domain space by private
  citizens.
• Can we keep Partisans from becoming
  Vigilanty-net? (Levis)

50
Network Architecture

• Challenging questions posed by new sensing
  contexts require a new network architecture
  Partisan
• Credibility Network contributes to credibility
  of many autonomous individuals data sources.
  Verified space-time context as network
  primitives.
• Dissemination Make it easy for people to share
  data. Policy-mediated rendezvous based on data
  properties and meta-data.
• Selecting Sharing Respect privacy concerns and
  encourage sharing.
• Verification Provide basic quality checks on
  data and context.
• Reliability Aggregation-based reliability -
  good enough equivalence.

Burke, Hansen, Srivastava, Parker, Redi, et al
51
Partisan Overview
Data sinks individuals and network apps
Selective sharing, verification, dissemination
Discovery and binding to data
End user applications
Data sources
Burke, Hansen, Srivastava, Parker, Redi, et al
52
How is Partisan different?

• Network support for these decentralized apps
• Sensors are typically not controlled by a central
  authority. Instead they are owned by users or
  groups.
• Network architecture contributes credibility,
  privacy protection, and services to support
  decentralized approach.
• Leverages deployed base of multi-use devices
• Over 2 x 109 users worldwide of cell phones.
• Consider cell phones, wireless cameras, etc. in
  lieu of motes
• Different sensing emphasis Location, video, and
  audio are vital because of device types and
  application requirements.

Burke, Hansen, Srivastava, Parker, Redi, et al
53
Key Partisan Network Component Mediators

• Mediators role protect both data and contextual
  information
• protect at a network level through interposition,
  indirection, physical proximity
• protect statistically through aggregation
• Physical Context and Sensor Data Validation
• Physical context information useful for
  validating integrity
• Aggregation can aid in verifying sensor data,
  down-sampling, blurring, and other anonymization
  techniques
• Mediator functions to explore
• Provides selected in-network functions on sensor
  data streams.
• Selective sharing policy enforcement and
  negotiation on behalf of a publisher or
  subscriber.
• Performing anonymization by removing
  identification information.
• Verification
• Enhancing streams with attested contextual
  information.
• Performing simple range/proximity checks.
• Dissemination
• Stream replication and reliability in presence of
  disconnections.

Burke, Hansen, Srivastava, Parker, Redi, et al
54
Participatory Sensing Potential

• Participatory Sensing
• Enable massive distributed, parallel collection
  of media
• Contextualize data to data for automated
  classification, verification.
• Leverage Partisan core to increase credibility
  and privacy for participants.
• Inspire innovative algorithms for managing the
  sampling process, opt-in location info, analysis
  tools, middleware, etc.

• Public health impact
• Personal/home indicators/testing
• Human activity patterns
• Municipal public health factors
• e.g., Air quality relationship to chronic health
  issues (asthma
• retrospective analyses of chronic health problem
  causes.

• Natural resource mgmt
• Facilitate high-quality field data entry
• Leverage signal processing for validation at time
  of entry
• Analyzable image-based data entry
• Adaptive protocols that depend on data collected
  and other (multi-scale) environmental conditions

Burke, Hansen, Srivastava, Parker, Redi, et al
55
Participatory Sensing Activities at UCLA

• Slogging premise
• citizen-initiated sensing, publishing, sharing
• SensorBase.org
• Urban Sensing Summit (Held May 06)
• UCLA, USC, UCI, UMN, Iowa State
• Nokia, Cisco, Disney, IBM, Intel
• Getty Conservation Inst., Mollenhauer, Metro
  Planning Report
• CS219 Course (Held Spring06)
• Platform Nokia 770/usb audio adapter/Bluetooth
  GPS Maps and Earth
• ecoPDA prototype development for Conservation
  International
• Biodiversity protocols
• Nokia n80 based (SensorPlanet)
• Selective sharing and context verification
  (NSF-FIND project funded)
• mediator architecture, verified context
  taggingapp participatory urban planning tool.
• Integration with backend discovery (ESP) and
  Sensorbase
• Ubicomp06 Demo.
• SensorPlanet Participation

56
Conclusions

• New themes will drive the next 5-10 years of
  sensing systems research
• Smart dust ? Mobile, Multi-scale, Multi-modal
• Science, military ? Urban, public health, social,
  personal, enterprise
• Resources, communications, autonomy ? Integrity,
  sensing, participatory and interactive
• Profound impact of cell phone a wireless sensor
  with intelligent mobility
• Nokias SensorPlanet effort
• Implications on and integration with Internet
• Publishing and sharing sensor data slogging
• Architectural support for verification, privacy,
  selective sharing
• Application authoring

56
57
Science applications are historical drivers for
information technology development and deployment

• Early embedded sensing applications
• Biological and Earth Sciences
• Environmental, Civil, Bio Engineering
• Public health, Medical research
• Agriculture, Resource management

• Science is early adopter because the technology
  is transformative and research tolerates risk
• The same technology will transform the business
  enterprise
• Important historical precedents
• Weather modeling--early computing
• Scientific collaboration--Internet
• Experimental physics (CERN)--WWW
• Computational science--Grid computing

• Embeddable device developments
• Energy-conserving platforms, radios
• Miniaturized, autonomous, sensors
• Standardized software interfaces
• Self-configuration algorithms

58
Engineering, enterprise, civic, and consumer
applications will eventually dominate

• As the technology matures we expect to find
  wide-reaching applications in the built
  environment, health care, and throughout the
  business enterprise.
• Todays systems focus on early-adopter
  science users (reveal the previously
  unobservable)

59
Acknowledgments

• CENS colleagues
• Jeff Burk, Jeff Goldman, Eric Graham, Mark
  Hansen, Tom Harmon, Jenny Jay, Bill Kaiser, Eddie
  Kohler, Greg Pottie, Phil Rundel, Mani
  Srivastava, Gaurav Sukhatme, John Villasenor and
  many others...
• Students (current and recently current
• Lewis Girod, Ben Greenstein, Martin Lukac, Andrew
  Parker, Nithya Ramanathan, Sasank Reddy, Thomas
  Schmid, Tom Schoellhammer, Thanos Stathopoulos,
  and many others...
• Funding agencies and Industrial Supporters
• NSF
• Intel, Nokia, Cisco, MSR