Bird behaviour inspires fire-spotting plan

As bushfires raged across California in October 2007, Professor Tim Hendtlass was attracting a great deal of attention on the other side of the US, where he was describing to conference delegates how unmanned aerial vehicles (UAVs) could be deployed in large numbers to monitor remote and inaccessible terrain for fires.

In Australia, bushfires destroy 100 000 hectares of forest and pasture each year and take the lives of many people and countless animals. Many fires start as a result of lightning strikes in remote areas, where they may burn undetected until they have caused vast damage and are almost impossible to control.

The fires in 2003 that devastated large areas of north-east Victoria and the NSW Snowy Mountains, and descended on Canberra – destroying hundreds of houses and claiming four lives – arose from bushfires that had been burning for weeks in the Namadgi National Park to the south and west of Canberra.

It is situations such as this that Hendtlass, director of the Complex Intelligent Systems Laboratory at Swinburne University of Technology, along with PhD researcher David Howden envisage could be monitored by using swarms of unmanned spotter planes to search for fires.

They decided to use an approach known as collective intelligence to address the question of how to control a number of UAVs and their searching patterns.

Hendtlass is interested in using ideas from biology. “Nature has found some very efficient ways to do things and I’m interested in seeing if we can replicate this work in applying it to common human problems,” he said.

For example, researchers have modelled the foraging behaviour of flocks of birds – the way they weave, search and share information to converge on promising food sources – and created algorithims to describe this pattern.

Rather than birds, or planes, converging on a point, Hendtlass and Howden are interested in describing the opposite behaviour: where a swarm of individuals (the UAVs) disperse from a point. “Could we invert the algorithm so that instead of converging when searching they spread out?” was the question they posed.

The researchers have divided terrain into a grid; with each grid square a monitoring point. “Some monitoring points need to be surveyed more often than others because they are more fire-prone. Lakes for example rarely, if ever, need to be monitored, while dry, bush-filled gullies in high lightning-strike areas need to be surveyed frequently. Our aim is to survey all the points (every grid square) as close to the desired frequency as possible,” said Hendtlass.

Their model also considers what happens if two UAVs come within a certain distance of each other and ensures they experience a force that makes them diverge.

Each UAV has its own map of an area, which it uses to make decisions about what to do next. Although central control of UAVs is feasible in the lab, in real terrain communications may be too unreliable for this to be effective so the UAVs have to ‘work it out for themselves’.

“The process of sharing map information between two UAVs that come close is the key to making the independent UAVs work as a dynamic team and enables a few cooperating UAVs to do work that would take a much greater number of fully independent UAVs to do,” he said. “That’s where the collective intelligence comes in.”

Researchers are at the simulation stage and have already written the algorithm that could be developed into a software program for use by fire management agencies and which, with a little further development, will soon be ready for commercialisation.

In the meantime, Howden is working on how to get the information back from the UAVs to a central point and investigating real data about fire spread to include in the program.

Editor's Note: A story provided by Swinburne University of Technology.  This article is under copyright; permission must be sought from Swinburne University of Technology to reproduce it.