Projects

Personal and course projects in robotics, autonomy, and software engineering.

image / gif	Autonomous PX4 Drone for Indoor GPS-Denied Flight PX4, Pixhawk, Raspberry Pi, ROS 2, OptiTrack \| Jan '25 – Present Built an autonomous UAV (PX4, Pixhawk, Raspberry Pi) for indoor GPS-denied flight using ROS 2 offboard control. Fused OptiTrack motion-capture pose into the PX4 EKF2 estimator for accurate state estimation without GPS.
image / gif	Cooperative UAV-UGV Localization using Extended Kalman Filter (EKF) [code] MATLAB, EKF \| Sept '25 – Dec '25 Developed a nonlinear UAV-UGV localization system in MATLAB using an EKF and simulation-based validation. Evaluated estimator consistency using Monte Carlo testing with NEES/NIS analysis on simulated and real data.
image / gif	Optimization-based Motion Planning for 3D Environments [code] MATLAB, YALMIP, MOSEK \| Sept '25 – Dec '25 Built a convex optimization–based collision-free trajectory generation framework in MATLAB using YALMIP/MOSEK. Modeled cuboidal and cylindrical obstacles as convex sets using superellipse-based formulations for tractable optimization.
image / gif	Dynamic Visual SLAM for Crowded Indoor Environments ORB-SLAM3, SegFormer, C++, Python \| Jan '25 – May '25 Integrated an ORB-SLAM3 pipeline with NVIDIA SegFormer to detect and reject moving objects in crowded environments. Evaluated in real university environments, achieving 0.077 m ATE and 0.073 m / 1.33° RPE, showing robust localization.
	Clothbot — Cloth Manipulation using Self-Supervised Value Network [poster, web] Python, ROS, PyTorch, UR5 \| Sept '24 – Dec '24 Developed a self-supervised value network policy using spatial action maps for dynamic cloth unfolding on a dual-arm UR5 setup. Achieved 95% coverage on rectangular cloths and 87.68% on unseen garments (T-shirts) with zero-shot sim-to-real transfer.
	Flying Ad-hoc Network Simulator for Multi-UAV Exploration [code] ROS, C++, NS3, Gazebo, PX4 SITL \| Aug '20 – Aug '22 Developed a co-simulation platform integrating NS3 and Gazebo through ROS/C++ for testing multi-UAV swarm tasks. Implemented UAV swarm motion planning (C++) and analyzed network metrics like PDR, hop-by-hop and end-to-end delay. Simulated a wildfire rescue UAV swarm (PX4 SITL and ROS) for surveillance applications. Published in ACM LANC '22.
	Mars Rover for University Rover Challenge — Controls Team Lead ROS, C++, NVIDIA Jetson \| Aug '19 – Apr '21 Designed path planning and tracking (PID, Stanley, Pure Pursuit) controllers for the manipulation and locomotion systems. Developed ROS/C++-based drivers and software interfaces to integrate sensors and motor drivers with an NVIDIA Jetson.

image / gif	Autonomous PX4 Drone for Indoor GPS-Denied Flight PX4, Pixhawk, Raspberry Pi, ROS 2, OptiTrack \| Jan '25 – Present Built an autonomous UAV (PX4, Pixhawk, Raspberry Pi) for indoor GPS-denied flight using ROS 2 offboard control. Fused OptiTrack motion-capture pose into the PX4 EKF2 estimator for accurate state estimation without GPS.
image / gif	Cooperative UAV-UGV Localization using Extended Kalman Filter (EKF) [code] MATLAB, EKF \| Sept '25 – Dec '25 Developed a nonlinear UAV-UGV localization system in MATLAB using an EKF and simulation-based validation. Evaluated estimator consistency using Monte Carlo testing with NEES/NIS analysis on simulated and real data.
image / gif	Optimization-based Motion Planning for 3D Environments [code] MATLAB, YALMIP, MOSEK \| Sept '25 – Dec '25 Built a convex optimization–based collision-free trajectory generation framework in MATLAB using YALMIP/MOSEK. Modeled cuboidal and cylindrical obstacles as convex sets using superellipse-based formulations for tractable optimization.
image / gif	Dynamic Visual SLAM for Crowded Indoor Environments ORB-SLAM3, SegFormer, C++, Python \| Jan '25 – May '25 Integrated an ORB-SLAM3 pipeline with NVIDIA SegFormer to detect and reject moving objects in crowded environments. Evaluated in real university environments, achieving 0.077 m ATE and 0.073 m / 1.33° RPE, showing robust localization.
	Clothbot — Cloth Manipulation using Self-Supervised Value Network [poster, web] Python, ROS, PyTorch, UR5 \| Sept '24 – Dec '24 Developed a self-supervised value network policy using spatial action maps for dynamic cloth unfolding on a dual-arm UR5 setup. Achieved 95% coverage on rectangular cloths and 87.68% on unseen garments (T-shirts) with zero-shot sim-to-real transfer.
	Flying Ad-hoc Network Simulator for Multi-UAV Exploration [code] ROS, C++, NS3, Gazebo, PX4 SITL \| Aug '20 – Aug '22 Developed a co-simulation platform integrating NS3 and Gazebo through ROS/C++ for testing multi-UAV swarm tasks. Implemented UAV swarm motion planning (C++) and analyzed network metrics like PDR, hop-by-hop and end-to-end delay. Simulated a wildfire rescue UAV swarm (PX4 SITL and ROS) for surveillance applications. Published in ACM LANC '22.
	Mars Rover for University Rover Challenge — Controls Team Lead ROS, C++, NVIDIA Jetson \| Aug '19 – Apr '21 Designed path planning and tracking (PID, Stanley, Pure Pursuit) controllers for the manipulation and locomotion systems. Developed ROS/C++-based drivers and software interfaces to integrate sensors and motor drivers with an NVIDIA Jetson.