Distributed Sensor Tracking: using the Physical Model

1
Distributed Sensor Trackingusing the Physical
Model

• Bart Selman, Carla Gomes, Scott Kirkpatrick,
• Ramon Bejar, Bhaskar Krishnamachari,
• Johannes Schneider
• Intelligent Information Systems Institute,
  Cornell University Hebrew University
• Autonomous Negotiating Teams
  Oct. 31, 2001

2
Outline

• Overview of our approach
• Ants - Challenge Problem (Sensor Domain)
• Movie 100 sensors, 10 moving targets

3
Overview of Approach

• Approach based on an energy function
• Sensors independently and asynchronously decide
  which target to track, based on closest
  neighbors status information. Make tradeoffs.
• Metropolis dynamics permits reassignments
• Simulated annealing controls process
• Constraints are relatively easy to incorporate
• Time dependent communications failures
• Communications delays, sensor orientation issues
• Tuning of method in progress, full report in Dec.

4
First results Movie

• 10x10 field of sensors
• Track 10 moving, turning targets
• Picture shows
• Target detection range (green circles)
• Target clusters (gt3 communicating sensors) are
  red lines to the target from the sensors