Autonomy you can trust, not just demo.

The lab's question is the one that matters most for autonomy: not whether a robot can complete a task, but whether you can trust that it will. My work sits in the simulation-first validation loop — building perception and control pipelines for ROS-based autonomous robots and systematically breaking them before they ever touch hardware.

The workflow runs on the Duckietown framework: an image-processing pipeline finds the lane, odometry and IMU data estimate where the robot actually is, and PID control keeps it tracking. The research contribution is in the validation methodology — designing simulation environments that stress the system across dynamic scenarios and identify failure modes early.

Formal methods are the other half of trustworthy autonomy. DESops is a University of Michigan Python library for discrete event systems — finite-state automata, parallel and product compositions, observer computation, supervisory control, and opacity enforcement. I made minor contributions to the library and used it for academic research in formal methods and the control theory of autonomous systems.

Where the CASA-Goes work validates behavior empirically, discrete event systems let you prove properties about it: what a supervisor can permit, what an observer can infer, what an outside party can or cannot deduce about system state. The combination — empirical robustness testing plus formal guarantees — is the direction I find most interesting.