📂 AI-based Odor Source Localization

We present a Knowledge-Driven framework to use large language models to process olfactory sensor data in odor source localization problems.

Developed a multi-modal navigation system that employs an LLM to process both visual and olfactory sensor readings to find a hidden odor source.

Proposed a Hierachical Control framework to control a robot finding the odor source by coordinating obstacle avoidance, vision-based navigation, and olfaction-based navigation behaviors.

Presented a robotic platform developed for the odor source localization task with both visual and olfactory detecting capabilities.

Integrated Twin Delayed Deep Deterministic Policy Gradient (TD3) with recurrent neural network (RNN) and implemented in the robotic odor source localization task.

Designed a Deep Learning model for End-to-End control of a mobile robot in finding the odor source location.

Improved the traditional moth-inspired odor source localization algorithm with a fuzzy inference system to adaptively change search parameters and behaviors.

Trained two neural networks, FNN and CNN, using Supervised Learning to control robots in finding odor source locations.

Designed a robot navigation algorithm for finding a hidden odor source. Two steps are involved: 1. Source & Plume Mapping (fuzzy inference) 2. Path Planning (RL).

Designed an End-to-End robot control model for finding odor source location using ANFIS.

📂 Wildfire Early Detection with Unmanned Aerial Vehicles

Enhance wildfire detection via predicting bounding boxes locations of smokes using the trend estimation.

Presented a new vision-based wildfire smoke detection from drone's imageries, which involves image segmentationa and object detection to reduce the false alarm rate.

Equipped a drone with both olfactory (smoke detector) and visual (camera) sensors to detect wildfire existence.

📂 Chemical Plume Tracing with Autonomous Underwater Vehicles

Present a deep reinforcement learning-base decision-making model to combine bio-inspired and engineering-based odor source localization algorithms.

Designed a 3-D hydrothermal vent localization algorithm using moth-inspired navigation method for using on autonomous underwater vehicles.

Design a chemical plume tracing algorithm for locating hydrothermal vents using A-star and POMDP source mapping algorithms.

📂 Multi-Agent Collaboration and Robot Localization

Designed a multi-agent coordination algorithm for controlling multiple unmanned surface vehicles in detecting moving objects.

Implemented Extended Kalman Filter and Unscented Kalman Filter to determine AUV locations based on IMU (acceleration), Doppler Velocity Log (velocity), and Short Baseline (position) measurements.

📂 Deep Learning-based Applications

Develop deep learning models to replace time-consuming vector radiative transfer (VRT) calculations in polarimeter sensor data retriveal.

Developed and compared multiple deep learning models for predicting physical properties of polymers.

Developed a Graphic Neural Network-based time-series prediction model to predict Thermomechanical of thermoset shape memory polymers.