CMPSCI 791L: Sensor Networks

1
CMPSCI 791L Sensor Networks

• Experiences Building a Real Distributed Sensor
  Network
• Victor R. Lesser
• Computer Science Department
• University of Massachusetts, Amherst
• September 12, 2003

2
Acknowledgements

• Bryan Horling
• Roger Mailler
• Jiaying Shen
• Dr. Regis Vincent (SRI)
• http//mas.cs.umass.edu/bhorling/papers/02-14.ps.
  gz
• http//mas.cs.umass.edu/bhorling/papers/00-49.ps.
  gz

3
Outline

• An example DSN problem
• Issues in Distributed Resource Allocation
• An example of one approach

4
Distributed Sensor Network Challenge Problem

• Small 2D Doppler radar units (30s)
• Scan one of three 120? sectors at a time
• Commodity Processor associated with each radar
• Communicate short messages using one of 8 radio
  channels
• Triangulate radars to do tracking

5
Representative of Distributed Sensor Network
Issues

• Need for Coordination/Distributed Resource
  Allocation
• Multiple sensors need to collaborate on tasks
• View objects of interest from multiple angles
  with different types of sensors
• Sensing time windows need to be closely aligned
• Environmental Dynamics
• Sensor configuration changes as target moves
• Potential for Resource Overloads
• Multiple target in overlapping sensor regions
• Limited Communication Channels

6
Representative of DSN Issues, cont.

• Soft Real-time
• Limited time window for sensing
• Must anticipate where target is moving in order
  to effectively allocate sensor resources
• Time for coordination affects time for sensing
• Distribution communication latency/limited
  bandwidth precludes global knowledge/control
• distributed data fusion
• Scalability need to be able to handle large
  numbers of sensor nodes
• Robustness local failures should not induce
  global collapse
• Handle uncertain information, sensor/processor/com
  munication failures

7
Real-Time Tornado Tracking
Internet2
supercomputers
8
Weather/Computation/Sensor Integrated Control
Hazardous Weather Detection algorithms
Determine initial conditions for near-term
dynamic forecasting models (NWP)
Quality Control (clutter removal, de-aliasing)
Retrieval of 3D wind, other fields
signal processing
radars
Assimilation, Multiple Doppler analysis
(more Compete gridding
Resource database weather-algorithm-provided utili
ty functions
9
How to Allocate Processing/Sensing Tasks

• Avoid processing overloads
• Avoid communication overloads
• Have information/processing co-located
• Avoid failure of network based on single
  location failure
• Allocate sensing so that as many targets can be
  tracked with reasonable fidelity
• Allocate processing/sensing so that real-time
  constraints can be met

10
Additional Questions

• What tasks can be assigned statically which have
  to be dynamically allocated
• When do static and dynamically made decisions
  need to be revisited
• What is the appropriate context for making these
  decision
• What decisions can be made locally
• What decisions need to made with in a non-local
  context
• Is this context fixed or dynamically evolved

11
Sensor Processing Issues

• Integrating Target Acquisition with Target
  Tracking
• Re-acquiring lost targets
• Data-Correlation Issues
• Recognizing which data belongs to which target
• Handling Uncertainty in Sensor Information
• How to make resource allocation issues in face of
  faulty sensor data

12
Tasks, Processes and Agents

• Issue of Autonomy -- Locus of Control
• How much leeaway is allowed in what goals to
  pursue, how to do them, who to interact with,
  what resources to use,
• Where are these decisions being made
• How decentralized are these decisions
• How dynamic/context-dependent these decisions are

13
Soft vs. Hard Real-Time

• There are not catastrophic effects if events are
  occasionally not interpreted correctly
• If lose sight of target for a few time steps and
  then reacquire generally okay
• Computation/Sensing after the deadline may still
  have some value
• Reduction in certainty of target location

14
How to Evaluate a Sensor Network

• Communication Locality
• Information and Processing Bottlenecks
• Organizational Control Overhead
• Overall Effectiveness
• .
• Whats Best --
• Multi-attributed Evaluation?

15
One Approach from an MAS perspective

• Decompose environment to form a partitioned
  organization.
• Each partition (sector) will contain a set of
  sensor nodes, each with its own controlling
  agent.
• Individual sectors are relatively autonomous.
• Specialize members of the agent population to
  dynamically take on multiple, different
  goals/roles.
• Individual agents become managers of different
  aspects of the problem.
• Managers form high-level plans to address their
  goals, and negotiate with other nodes to achieve
  them.

16
Sectored-Based Agent OrganizationAgents
Multiplex among Different roles
17
Organizationally-Structured Communication among
Agents
DrA
DrQ
DrR
TB
RR
TD
PTC
RB
PC
DA
TBU
ES
18
Managing Conflicted ResourcesSensors,
Processors, Communication

• Sensors
• Conflicting Scanning Tasks from different Sector
  Managers
• Locally resolved by agent connected to sensor --
  SRTA agent
• Tracking Tasks wanting same sensor resources
• Negotiation among track managers -- SPAM protocol
• Communication
• Communication Degradation due to lack of Locality
• Track manager migration among sectors
• Communication Channel Overload
• Sector manager assignment of track manager roles
• Processors
• Data Fusion Overload/Knowledge locality
• Sector manager assignment of data fusion/track
  manager roles
• Multiplexing Roles -- SRTA agent

19
Centralizing Information in Sector
ManagerHandling Data Correlation with Multiple
Tracks

• Targets are represented by uncertainty bounds
• Bounds are affected by speed of target and age of
  supporting measurements
• Bounds are shared with sector manager, who in
  turn shares them with other track managers
• Sector manager
• Uses target uncertainty bounds to determine if
  new target detections (from scanning) are known
  targets
• Data from known target detections are used to
  focus attention of relevant track manager
• Track managers
• Uses amplitude lobe intersections to estimate
  position in times of need
• Prevents data fusion if estimated resolved
  position is within another targets bounds
• Throws out ambiguous measurements which intersect
  another targets bounds

20
Fault Tolerance

• Node information is propagated through the use of
  directory services (x, y, orientation, etc.).
• Sensors provide sector managers with their
  information.
• Track managers query sector managers for sensor
  details.
• This information is cached for future use at each
  step
• The directory held in sector manager maintains
  historical query information
• New data are analyzed for relevance to those
  queries
• Relevant information is automatically propagated
  to the query source
• This process quickly updates agents beliefs,
  allowing them to adapt to change

21
Major Issues in This Approach

• What is an appropriate organization for agents
• Scalability and Robustness
• Self-Organization and Adaptation
• What is the protocol for distributed resource
  allocation
• Soft Real-Time, Graceful Degradation, Efficient
• What is the structure of an agent architecture
  that supports
• Agents functioning in an organizational context
• Agents implementing complex distributed resource
  protocols
• Agents operating under soft real-time constraints

22
Some Final Thoughts

• Can not isolate one set of issues from another
• Strict layering of issues does not seem to work
• There is no one best approach
• Very sensitive to characteristics/capabilities of
  sensors, quality of sensor data, the character of
  required sensor fusion, amount and type of
  processing required, system objectives,
  communication and processing capabilities,
  environment