Date of Award


Access Type

Dissertation - Open Access

Degree Name

Doctor of Philosophy in Electrical Engineering & Computer Science


Electrical Engineering and Computer Science

Committee Chair

Shuo Pang, Ph.D.

First Committee Member

Brian Butka, Ph.D.

Second Committee Member

Hever Moncayo, Ph.D.

Third Committee Member

Richard Prazenica, Ph.D.

Fourth Committee Member

Tianyu Yang


Robotic odor source localization (OSL) is a technology that enables mobile robots or autonomous vehicles to find an odor source in unknown environments. It has been viewed as challenging due to the turbulent nature of airflows and the resulting odor plume characteristics. The key to correctly finding an odor source is designing an effective olfactory-based navigation algorithm, which guides the robot to detect emitted odor plumes as cues in finding the source. This dissertation proposes three kinds of olfactory-based navigation methods to improve search efficiency while maintaining a low computational cost, incorporating different machine learning and artificial intelligence methods.

A. Adaptive Bio-inspired Navigation via Fuzzy Inference Systems.

In nature, animals use olfaction to perform many life-essential activities, such as homing, foraging, mate-seeking, and evading predators. Inspired by the mate-seeking behaviors of male moths, this method presents a behavior-based navigation algorithm for using on a mobile robot to locate an odor source. Unlike traditional bio-inspired methods, which use fixed parameters to formulate robot search trajectories, a fuzzy inference system is designed to perceive the environment and adjust trajectory parameters based on the current search situation. The robot can automatically adapt the scale of search trajectories to fit environmental changes and balance the exploration and exploitation of the search.

B. Olfactory-based Navigation via Model-based Reinforcement Learning Methods.

This method analogizes the odor source localization as a reinforcement learning problem. During the odor plume tracing process, the belief state in a partially observable Markov decision process model is adapted to generate a source probability map that estimates possible odor source locations. A hidden Markov model is employed to produce a plume distribution map that premises plume propagation areas. Both source and plume estimates are fed to the robot. A decision-making model based on a fuzzy inference system is designed to dynamically fuse information from two maps and balance the exploitation and exploration of the search. After assigning the fused information to reward functions, a value iteration-based path planning algorithm solves the optimal action policy.

C. Robotic Odor Source Localization via Deep Learning-based Methods.

This method investigates the viability of implementing deep learning algorithms to solve the odor source localization problem. The primary objective is to obtain a deep learning model that guides a mobile robot to find an odor source without explicating search strategies. To achieve this goal, two kinds of deep learning models, including adaptive neuro-fuzzy inference system (ANFIS) and deep neural networks (DNNs), are employed to generate the olfactory-based navigation strategies. Multiple training data sets are acquired by applying two traditional methods in both simulation and on-vehicle tests to train deep learning models. After the supervised training, the deep learning models are verified with unseen search situations in simulation and real-world environments.

All proposed algorithms are implemented in simulation and on-vehicle tests to verify their effectiveness. Compared to traditional methods, experiment results show that the proposed algorithms outperform them in terms of the success rate and average search time. Finally, the future research directions are presented at the end of the dissertation.