FRoSt Lab

GPS-Denied Path Planning

Uncertainty-Constrained

I am developing path-planning algorithms to minimize localization error for robot swarms in GPS-denied environments. Simulating an underwater robot swarm that localizes based on landmarks and shared factor graphs, I am testing approaches to determine optimal paths.

RCSP Dijkstra's

I wrote a Julia script to determine the optimal path to a goal given a maximum threshold on risk or path length. I determine risk by convolving our projected uncertainty with landmark uncertainty and summing the combined distribution that falls in our visibility radius.

Value Iteration for Best Path

With two classmates in my decision-making under uncertainty class, we used value-iteration to solve for the best path to minimize goal localization uncertainty for an agent. This is a proof of concept showing advantageous paths for minimizing time to goal with a maximum uncertainty threshold. Code can be found here.

Terrain Classification Algorithms

I developed three techniques for improving terrain and object classification in robotic systems using machine learning algorithms FAST-LIO, SAM2, YOLO, and CLIP.

Category and Subcategories

Aggregating Subcategories

Weighted Subcategories + Aggregation

Clahe Filter for Enhanced Segmentation