Stabilizing Tracking :Demos on Berkeley’s Mote Platform

Anish Arora, Murat Demirbas, Sandip Bapat, Sohail Munir

Ohio State

Tracking Problem

• Consider an undirected graph where any two nodes connected by an edge are called neighbors

• There is an evader that moves from a node to a neighboring node each Te seconds

• There is a pursuer that moves from any node to a neighboring node each Tp seconds, where Tp < Te

• Associate a “sensor process” with each node so that the pursuer catches up with the evader in short time

Two Approaches to Stabilization

• OSU tracking protocol sensor processes communicate periodically with neighbors stabilizes and tracks faster

• UT tracking protocol sensor processes communicate with neighbors only upon

request minimizes number of messages and energy efficient

OSU Tracking Protocol

Sensor process

• Processes communicate periodically (four times per second) with their neighbors

• Message format Src: Id of sender Parent: Id of parent of Src Dist: Distance from root to Src Value: The logical timestamp when the bug was detected by

the root mote of Src

• Processes run stabilizing tree algorithm to form the

tracking tree

Stabilizing Tree Algorithm

• Dynamic-rooted minimum-distance spanning tree algorithm

• Self stabilizing to the invariant that the tree is rooted at the bug

• Asynchronous Does not require any underlying synchronization

mechanism, or local clocks

• Ad hoc Works for any topology

A Tracking Tree Forms…with Bug at Root

• Bug is on mote 2

• Visualization tool shows tracking tree formed with root at mote 2

Video Screenshot

Tree Stabilizes…when Motes State gets Corrupted

• Tree rooted at mote 2, where bug is located

• State of mote 16 is transiently corrupted

• Two trees, with roots at motes 2 and 16, are formed

• Source of corruption at mote 16 is removed

• A single tree, potentially different from the original one, is (re)formed

Video Screenshot1 Screenshot2

Tree Stabilizes…even when Motes Failstop

• Tree is rooted at mote 2

• Mote 14 points to mote 10 as the next hop on its shortest path to the root

• Mote 10 fails (is turned off, radio unit fails, or runs out of battery)

• Mote 14 now points to mote 13 or mote 15 instead of the failed mote 5

• Tree (re)stabilizes

Video Screenshot1 Screenshot2

Spider Modes

• Line following If spider goes off line, it searches for line and follows it until it

reaches the end of the line

• Direction tracking Colored lines encode the four directions

e.g. Long Yellow + Short Black = North Spider calculates its direction after a complete line traversal

• Reading from mote Motes blink IR LEDs at four different frequencies to communicate

the four directions On reaching a mote, spider detects the frequency of emitted IR

signals and decides how to turn

The Spider Converges to the Grid...and Tracks the Tree

• Spider converges to the grid from any point within grid, without falling off the edge

• Upon converging, spider regains its sense of direction within one complete line traversal

• Upon regaining sense of direction, spider tracks tree by following direction signals from the motes

Video

The Spider Catches the Bug…assuming Spider is Faster

• Spider follows the tree all the way to the root, where the bug is located and catches the bug

• Even if the bug moves, the tree reconfigures when the bug is detected at the next mote

• So, if the bug stops or is slower than the spider, the spider eventually catches the bug

Video

The Spider Catches the Bug…even if Motes Failstop

• If spider reaches a failed mote, it cannot get a direction signal

• Spider then chooses a random direction to follow. If this direction leads it off the grid, it backs up and retries till it finds a grid direction

• On reaching the next non-failed node, the spider gets a proper direction to follow and catches the bug eventually

Video

Engineering tricks

• Main problem – Lack of a fat communication pipe between spider and motes

• Encoding directions using line colors

• Using IR LEDs on motes to communicate with spider

• Ensuring that motes do not detect spider as the bug using a light source

Future work: Hybrid approaches

• Proxy Pursuer or Agent

• Publish/Subscribe service for interest based information propagation to pursuer

• On-demand stabilization of the spanning tree