Page 47 - AI for Good-Innovate for Impact

AI for Good-Innovate for Impact

Use Case – 7: Smart UAV Networks for Efficient Disaster Response

Country: Turkey 7 - ITU

Organization: Istanbul Technical University
Contact person: Dr. Nazim Kemal Ure, ure@ itu .edu .tr

7�1� Use case summary table

Domain Disaster Response
Problem to be To solve delays, resource limitations, and logistical challenges during
addressed disaster response.
Key aspects of the Combo of drones, object detection and satellite based coordination
solution for rescue operations. Drone-2-drone or drone-base station commu-
nication.
Adhoc network design.

Technology keywords Multi-agent, collaborative intelligent solution.
Data availability Video and still images, satellite images to determine the location and
network location.

Metadata (type of Video and Images
data)

Testbeds or pilot sim2real
deployments

7�2� Use case description

7�2�1 Description

The proposed use case aims to harness the advancements in reinforcement learning (RL)
to enhance the deployment, route selection, and coordination of unmanned aerial vehicles
(UAVs) in disaster scenarios, especially for scenarios that require immediate response such as
earthquakes and floods. Traditional disaster response efforts are often hampered by delays,
resource limitations, and logistical challenges. To overcome these obstacles, the use case
develops a coordinated UAV network designed to autonomously perform essential tasks within
disaster-stricken areas. Utilizing RL algorithms, UAVs can learn and adjust their operations
(including route navigation, returning to charging stations, and data detection and transmission)
based on feedback from the environment. In particular, the project integrates several state of
the art RL approaches, such as multiagent learning (for achieving efficient cooperation among
UAVs), sim2real transfer (for leveraging simulated data) and curriculum learning (for achieving
smoother learning curve from simple to complex scenarios). This combination of approaches
allows for the optimization of task distribution and resource management in real time, while
ensuring generalization across a rich variety of disaster scenarios.

31