PSN Game — Multi-Agent Planning
Developed a game-theoretic path planning framework that models interactions between agents as dynamically selected games. A Player Selection Network (PSN) identifies which neighbors matter, reducing computational cost and achieving a 10× runtime reduction while preserving solution quality. Accepted to AAMAS 2026 (extended abstract).
Paper: Read on arXiv
Inverse Material Geometry Reconstruction
Connecting machine learning and solid mechanics, I programmed a lightweight PyTorch autoencoder to reconstruct hidden, nonlinear material geometries from 2D Digital Image Correlation (DIC) strain fields and force-displacement data. Targeting the 2025 Mechanical MNIST Challenge, I built a data pipeline to process and augment physical strain maps, achieving robust one-shot geometry identification under strict data constraints.
Context: Self-directed research under Dr. Tan Bui · Fall 2025
VTOL Transition Flight Simulation
I created a pseudo-6-DOF simulation environment to optimize propeller throttle schedules during the VTOL-to-forward-flight transition. By coupling rigid-body vehicle dynamics, propulsion power budgets, and stochastic wind perturbation models, the project identifies throttle curves that minimize altitude loss while respecting motor constraints.
Context: FlareX Senior Design · 2025–2026