Model-Based Autonomous Land Rover
Built a ground-up autonomous differential-drive Land Rover integrating GNC, EKF-based localization, ROS/MAVROS telemetry, and embedded motor control for closed-loop navigation.
Projects
Motion Planning and Real-Time Nonlinear Model Predictive Control for the Toyota HSR
Research Project
Real-time autonomous navigation stack for the Toyota Human Support Robot, combining nonlinear model predictive control with PRM* and RRT* global planning for safe, collision-free motion through constrained environments.
Gait-Aware Quadruped Locomotion with Temporal Logic
Research Project
Applied Signal Temporal Logic specifications to shape reinforcement-learning rewards for gait-aware quadruped locomotion, targeting safer and more interpretable policy learning.
Steering Vision-Language-Action Models for Safe Robotic Execution
Steered Vision-Language-Action robot policies at inference time using activation-space interventions, exploring ActAdd and CAST for safer, constraint-aware execution without retraining.
Autonomous Vehicle: Motion Planning & Electronics
Developed ROS2/NAV2 autonomy software and vehicle electronics for an IGVC-style autonomous ground vehicle, integrating GPS waypoint navigation, LiDAR-based obstacle avoidance, Cube Orange communication, and custom omnidirectional wheel hardware.
Simulated Aquatic Vehicle for Cave Exploration
Built a simulated underwater autonomy stack for cave exploration, integrating sonar-based SLAM, frontier detection, 3D A* planning, waypoint tracking, and STL runtime safety monitoring.
LEGO EV3 Mini-Vehicle
Built LEGO EV3 autonomous robot behaviors for grid-based navigation, fire detection, wall following, and IR-ball soccer using MicroPython, motor control, and onboard sensors.
Cart Pole Stability through Q-Learning
Implemented a PyTorch Double DQN agent for Gymnasium CartPole, using experience replay, epsilon-greedy exploration, target-network updates, and reward/epsilon logging to learn stable pole-balancing behavior.
DuckDB: Learned Indexing Extension
Extended ALEX, a C++ adaptive learned index, by prototyping polynomial/quadratic model nodes to better handle nonlinear key distributions in sorted range-index workloads. Implemented in DuckDB.
USC CPS-VIDA Lab Website Redesign
Redesigned website for USC's CPS-VIDA LAB, using Next.js. Hosted on GitHub pages