Distributed-In/Distributed-Out: Intelligent Sensor Networks for Coordination Constantine (Dean) Christakos Viral Communications

Distributed-In/Distributed-Out: Intelligent Sensor Networks

for Coordination

Constantine (Dean) ChristakosViral Communications

[email protected]

Introduction An everyday problem How sensor networks can be applied The concerns about current designs Our proposed solution Preliminary work Where do we go from here?

How Do We Find Our Way Around? We follow the signs! Getting out of lower Manhattan:

It Doesn’t Always Work Smoothly While following the signs, a driver could

encounter: Congestion A car accident Construction

If the driver is lucky, public works will have time to put up detour signs

Would Sensor Networks Help? Sensors monitor environmental conditions

on remote islands and volcanoes. So why not in Manhattan?

Distributed Input of Sensor Data The sensors are placed all over the

monitored area and record separate pieces of data.

The responsibility for monitoring this wide area is distributed over dozens or even hundreds of tiny sensors.

Distributed Input, Centralized Output Normally, sensor

networks are used to monitor an area and send that data back somewhere to be analyzed

Why can’t observers within the sensor network get data directly?

Base Station/Middleware

Sensor InquirerB

Sensor InquirerA

Sensor InquirerC

sens

or d

ata

Sensor network

Why is this a problem? Centralized Control leads to centralized

failures. New York’s Office of Emergency

Management was located in the World Trade Center.

Re-coordination requires the global control center to be contacted even when the problem itself is localized.

Distributed Output Each sensor acts an output. Instead of acquiring data from a central

point, each sensor node displays relevant local information to the observer.

No Central Control! Localized failures don’t interfere with the

ability to find a solution to global problems. Intelligence moves away from a central

coordinator and out to the individual nodes.

Using Local Information in Problem Solving Using just local information, we can get

closer to our goal. Signs give us the next step to the tunnel or

the bridge. Sensors can tell users the next step to a

needed resource and can warn users of upcoming dangers.

Resource Discovery The sensors direct users towards

resources, based on data they have acquired.

Sensors can keep track of the direction of the nearest route to the exit

Danger Avoidance Nodes that can see a clear path to the

destination guide users towards them Nodes that are informed of nearby danger

guide users away from those areas

The Protocol – Finding the Nearest Exit Each node keeps track of which of its

neighbors is closest to the exit and sends users in that direction.

The next node keeps aware of which of its neighbors is next closest to the exit and sends users in that direction.

And so forth, until the user arrives.

Disaster Recovery In the event of a blocked passageway, the

protocol finds alternates. Only the next step in the process is important.

Example: during building escape, nodes need to direct the users on upper floors to the ground floor. The nodes don’t need to know that users on the ground floor need to use the rear exit, rather than the front exit.

Local Recovery Scales As the network (or building or city) gets

larger, disaster recovery will remain a function of communication within a constant radius. Briefly: the larger the graph, the more paths

that exist

Simulation Using a sensor network simulator, we

have implemented a simple path-finding algorithm and are currently testing.

In this case, nodes trace a path to the center node:

Implementing These Cheaply Recently, the advent of inexpensive

bluetooth devices has allowed us to make wireless multihop networks out of computers throughout the lab:

Future Uses Intelligent decision-making can be pushed

outwards to the agents “in the field.” Examples:

UPS drivers communicate and coordinate locally to makeup lost time due to canceled drop-offs by finding nearby drivers.

“Pullers” on election day can coordinate with others while doing get out-the-vote

All these are done without a central coordinator and by leveraging large numbers of sensors to find what people need.

Questions?