Developed multi-agent maritime simulation framework by adapting Waymax repository to model ship trajectory planning and navigation scenarios with multiple autonomous vessels Built and trained a behavioral cloning (BC) policy to mimic expert ship navigation in complex multi-vessel scenarios Reproduced ShipNavism results by creating custom environments and benchmarking performance against published metrics Developed actor-based visualization tools to analyze trajectories, multi-vessel interactions, and collision avoidance

Conducted deep learning experiments using PyTorch, working with both pretrained and custom models on clean and noisy datasets. Applied conformal prediction techniques for model evaluation and utilized Weights & Biases for experiment tracking. Utilized pose detection models (YOLOv7, YOLOv8, Detectron2, SAM2, Sapiens) for walking analysis, modifying code from popular repositories to meet specific project requirements. Implemented Label Studio for comprehensive data annotation and labeling, contributing to high-quality training data preparation