Projects

Research Projects

Learning Mixed-Integer Solvers for MPC (2025)

Explored differentiable optimization for mixed-integer quadratic programs (MIQP) in model predictive control.
Developed a CVXPYLayer-based learning framework where neural networks predict integer decisions while the continuous optimization is solved in a differentiable QP layer.
Keywords: differentiable optimization, MIQP, learning-to-optimize, MPC

Gaussian Process Dynamics with MPPI (2025)

Built a Gaussian Process (GP) model to estimate vehicle dynamics and integrated it with Model Predictive Path Integral (MPPI) control.
Focused on uncertainty propagation and adaptive trajectory control performance.
Keywords: Gaussian Process regression, uncertainty-aware control, Bayesian learning

Autonomous Systems

F1TENTH Autonomous Racing (2025)

Participated in three autonomous racing rounds using a full ROS 2 stack for perception, planning, and control.
Developed and integrated multiple subsystems that formed a complete racing pipeline:

• Planning & Control
  • Implemented Wall-Follow and Gap-Follow algorithms with PID tracking.
  • Developed Pure Pursuit trajectory tracking and MPC-based optimal control.
  • Added dynamic obstacle avoidance.
• Perception
  • Designed vehicle detection pipeline combining YOLO object detection and LiDAR clustering for robust localization and obstacle tracking.
• Races
  • Competed in two time-trial events and one head-to-head race, achieving real-time performance under tight computational budgets.

Tools: ROS 2, Python/C++, RViz, TensorRT
Keywords: autonomous racing, motion planning, perception, MPC, embedded robotics

LLM-Guided Navigation for F1TENTH (2025)

Investigated using large language models (LLMs) for natural-language-driven navigation and control decisions.
Connected semantic task instructions to high-level control primitives in simulation.
Keywords: LLM robotics, task planning, autonomous driving

Classical Control and Learning Foundations

Quadrotor Control Paper Reproduction (2025)

Reproduced results from a published nonlinear quadrotor control paper.
Implemented attitude and trajectory controllers, sensor fusion and state estimation, validated stability and robustness, and analyzed sensitivity to modeling errors.
Tools: Python

Image-to-GPS Regression (2024)

Implemented Vision-Transformer regression to estimate geographic coordinates from street images, benchmarking against classical feature-matching baselines.
Keywords: vision transformers, geolocation, regression