Michael Bock

My master's project was a normalizing flow used for density estimation of Directors & Organizations (D&O) insurance cases, built within Georgia Tech's GitHub.

Graduated with a Computer Science degree while leading VT Baja's testing software and working on autonomous vehicle perception.

Built systems that sense and interpret a self-driving car's surroundings. On the Maps subsystem I used GPS and HD maps (PostgreSQL) via a ROS localizer node to look up the car's lane, lane offset, and speed limit. On the Computer Vision side I worked on neural network nodes for detecting signs, traffic lights, and obstacles using YOLOv3 and OpenVINO — including an unconventional sign classifier that reads sign text directly, made reliable by a "lidar-first" pipeline that crops images to regions of interest.

First, my team designed a poly-time algorithm for the NP-complete subgraph isomorphism problem using quantum annealing, constructing Q matrices via penalty functions so the correct answer mapped to the minimum energy state — intended to find vulnerabilities in call graphs. Second, I worked on controls for a lighter-than-air soccer-playing aircraft: a PID controller for stability, an autonomous camera search pattern, and grabber logic, all in ROS and C++.

Led the backend team for an augmented-reality startup, handling loading, storage, and placement of AR objects. Tackled monocular depth estimation using SLAM, and deployed AWS S3 + CloudFront to store and distribute AR assets quickly.

Programmed the sensor suite for the Virginia Tech Baja 2020 car in C++ — reading data from the Inertial Measurement Unit (IMU), linear potentiometers, and other sensors, plus software that logs everything to CSV files.

Designed and implemented the routing algorithm that orders package deliveries for drivers around Blacksburg and Christiansburg, constantly iterating on new approaches to increase throughput.