Abstract
dc:description.abstractAtmospheric smoke plumes play a critical role in environmental monitoring and wildfire management, yet accurately tracking these highly dynamic flows remains challenging due to their unsteady motion and continuously evolving shape. This thesis presents an autonomous drone system capable of dynamically tracking smoke plumes under highly unsteady atmospheric conditions. The system integrates a quadrotor with onboard edge computing for real-time perception-based control. A two-phase framework is employed: an initial descent using vision-based smoke detection and optical flow, followed by in-plume tracking governed by a PID or a DRL controller trained with Proximal Policy Optimization (PPO). The DRL approach uses binary smoke segmentation masks as state representation, enabling robust adaptation without explicit flow measurements. Evaluation in a high-fidelity Unreal Engine simulation and real-world field experiments demonstrates that the DRL-based controller exhibits superior robustness and adaptability in complex, rapidly changing conditions, advancing the state of the art in autonomous environmental monitoring.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Pal, Srijan Kumar
Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Citation
- Pal, Srijan Kumar. (2026). Autonomous Drone for Dynamic Smoke Plume Tracking. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/282264.