Javascript must be enabled for the correct page display

Isolating Wildfires using a waypoint generating CNN-based Multi-agent system

Visee, Thijs and Bäuerlein, Markus and Hewlett, Johnathan (2021) Isolating Wildfires using a waypoint generating CNN-based Multi-agent system. Bachelor's Thesis, Artificial Intelligence.

[img]
Preview
Text
Wildfire_Bachelors_Thesis_v1.5.pdf

Download (434kB) | Preview
[img] Text
toestemming.pdf
Restricted to Registered users only

Download (128kB)

Abstract

As a result of climate change, the forest fire season is becoming longer and more ferocious. One technique for controlling and mitigating the damage of forest fires is to use heavy machinery to clear paths of undergrowth, removing the fuel for the fire to continue spreading. In this paper we investigate the feasibility of using a convolutional neural network (CNN) to determine waypoints for multiple agents in a simulated environment, so that these agents can dig the firebreaks between these points to contain the fire. The CNNs are trained by learning from demonstration with the varying environmental conditions such as the wind direction and propagation speed of the fire. Different CNN architecture designs were used to predict the way-point position per agent and the overall results of each of these variants were compared. The results show that this approach to controlling forest fires in the simulated environment is not only possible but very effective for CNN-architectures that represent relative positions as continuous values. While for discrete output representations there remains some improvement, this paper provides ample foundations for future research to investigate the limitations of our approaches and to model more realistic simulations.

Item Type: Thesis (Bachelor's Thesis)
Supervisor name: Wiering, M.A. and Maathuis, H. and Visser, J.
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 24 Aug 2021 11:02
Last Modified: 24 Aug 2021 11:02
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/25764

Actions (login required)

View Item View Item