Safety-critical control with measurement and actuation uncertainties:

• Developed a provably safe feedback controller using sampled-data Control Barrier Functions (CBFs) and convex optimization for nonlinear systems with measurement and actuation uncertainties.
• Implemented the controller on Franka Research 3 robot manipulator and Crazyflie quadrotor, achieving real-time obstacle avoidance where conventional controllers failed.

Scalable verification of learning-enabled systems:

• Designed an efficient and tunable method for safety verification of perception-based autonomous systems using set-based computation and optimization techniques (LP, MILP, SDP).
• Accelerated robustness and sensitivity analysis of large-scale neural networks by 10x compared to baseline tools, through a novel local NN compression method and MILP cutting-plane techniques.
• Integrated verification results into controller design, achieving provably safe goal-reaching and obstacle avoidance for robotic systems with machine learning components.
• Applied parallelization (CUDA, OpenMP) to accelerate Monte Carlo reachability and real-time state estimation for high-dimensional nonlinear systems.
• Supervised undergraduate research on optimizing algorithm runtime and scalability; guided one mentee to co-author a publication in a top venue

Robust stability of neural network control systems:

• Derived new robust stability conditions via Linear Matrix Inequalities and quadratic constraints, enlarging the certified stability region by 5x over baseline methods.
• Developed a data-driven, provably stable controller design and verification framework for uncertain and disturbed systems without requiring system identification.

Hierarchical control for autonomous aircraft taxiing:

• Built a hierarchical architecture for autonomous taxiing, including high-level graph-based path planning, mid-level trajectory generation, and low-level tracking control.
• Ensured safety-by-design control of aircraft taxiing by integrating ATC-level constraints (e.g., runway selection) and ground-level requirements (e.g., speed limits, obstacle avoidance) using graph theory and MPC.

Perception-based autonomous aircraft landing:

• Developed vision-based landing control using camera sensing, PID control, YOLO detection, SIFT feature extraction, and geometric state estimation via Perspective-n-Point algorithms.
• Validated the proposed algorithms in the X-Plane 11 high-fidelity flight simulator.

• Grading for graduate level course: ECE 560 - Linear Systems Theory

Robot kinematics and dynamics:

• Implemented PID control for robotic manipulators in joint-space and Cartesian-space.
• Generated optimal manipulator trajectories in cluttered environments using advanced planning and optimization methods, including PRM, RRT, A*, and potential field
• Applied state estimation algorithms (KF, EKF, UKF, MHE) to achieve high-accuracy estimation of the manipulator position.

Computational fluid dynamics of MILD combustion:

• Built high-fidelity 3D models of boilers and performed CFD simulations of methanol MILD combustion under varying operating conditions.
• Conducted experimental validation in a mid-scale boiler, demonstrating higher thermal efficiency and lower emissions compared to traditional combustion.

Human kinetic energy harvesting:

• Designed a frameork to collect kinetic energy from human motion and vibrations under the supervision of Prof. Wei-Hsin Liao.