Current navigation systems often underestimate the true time needed to arrive by reporting driving time, while ignoring parking search and walking to the final destination. Probability-aware parking selection addresses this issue by using parking success probabilities for individual lots to direct drivers to locations that minimize expected arrival time. However, the existing framework assumes these probabilities are known, so this project extends the framework by using user traces to estimate parking probabilities. In this SuperUROP, I will build on my preliminary work with simulated 2D-grid data by testing whether trace-based inference is realistic beyond a simplified simulation environment. I will identify suitable GPS trace data and validate inferred parking attempts, with the goal of developing the probability estimation layer needed to make probability-aware parking selection practical in real-world settings.

I am participating in SuperUROP because it builds directly on my interest in transportation modeling and real-world infrastructure systems. In a previous class, I used machine learning and a Poisson neural network to predict when BlueBikes stations would become empty or full, and I am excited to extend this approach to parking and navigation. Through this project, I hope to strengthen my ability to work with raw, imperfect data and develop models from first principles for a practical transportation problem.