Content area
This thesis addresses fundamental challenges in chemical plume tracking and odor source localization across large outdoor scales, with applications in environmental monitoring, disaster response, and public safety. Utilizing extensive field data collected from diverse environments such as the Black Rock Desert and Whittell Forest, I demonstrate that integrating statistical features of odor encounters over time enables accurate source distance estimation, with achievable median errors of 3-8 meters. This work reveals the importance of using a memory and highlights the most relevant statistics that could be used to localize the source. It also reveals that high temporal resolution sensing (at least 20 Hz) is essential for extracting useful distance information from odor signals. The thesis also introduces COSMOS (Configurable Odor Simulation Model Over Scalable Spaces), a novel probabilistic data driven simulation framework that combines empirical data with adaptive state transitions based on historical experience of odor encounters and autoregressive modeling to realistically reproduce the odor experiences of an agent moving through complex plumes across large spatial scales while maintaining computational efficiency. Comprehensive validation confirms the framework's ability to accurately replicate key statistical features of natural odor plumes, making it a valuable tool for developing and evaluating odor-tracking algorithms. Finally, a ROS-based simulation environment incorporating quadrotor and wind dynamics extends these insights to aerial vehicle applications. Together, these three complementary components—empirical distance estimation, data-driven odor experience simulation, and real-time simulation in a physics enabled robotics simulator (Gazebo-ROS) that can be directly translated to a real system—establish a comprehensive pathway toward developing autonomous outdoor plume-tracking robots for real-world deployment. The thesis concludes by identifying promising directions for future research, including multi-modal sensing integration, enhancements to the COSMOS framework, and swarm-based approaches to odor source localization.