Selected work
A collection of work in robotics, machine learning, and computer vision.
ModLab · Mobile robotics
Connecting caster mechanics, visual sensing, and feedback control in a holonomic mobile robot. Work spans dual-camera caster tracking, depth-camera localization, and ROS 2 / embedded PID control.
3D reconstruction · Robot perception
Exploring 3D Gaussian Splatting for indoor scene reconstruction from RGB-D lap recordings, calibrated camera data, and robot poses.
Robot learning · Reinforcement learning
A residual reinforcement learning framework around a frozen vision-language-action policy. SAC and PPO learn bounded action corrections through a shared training and evaluation pipeline.
Robotics · Optimal control
Developed a planar boxing-arm simulator that uses Model Predictive Control with iLQR to generate fast, contact-rich attack, block, and parry motions between two opposing robotic arms. The system demonstrates real-time trajectory optimization and adaptive defensive behaviors.
Machine learning · Audio
Developed and compared three deep learning architectures to distinguish synthetic speech from real audio, exploring recurrent models and transformer-based audio representations.
Computer vision · Remote sensing
Implemented UNet3+ for settlement segmentation in satellite imagery, using the IEEE GRSS 2021 ESD dataset to study electricity access in African settlements.
Graphics · Game development
Created a Minecraft clone from scratch using OpenGL, featuring custom rendering, texturing, and player physics. The game allows players to build, destroy, and move around, closely mimicking the mechanics of the original game.
Embedded systems · HackDavis 2023
HackDavis 2023 winner, this project enhances biker safety by integrating LED blinker lights, a rear-view vehicle detection and alert system, and an emergency contact messaging system directly into the helmet.
