Minku Kim

Quadruped Control

Obstacle Avoidance Using Control Barrier Functions

This project applies control barrier functions to obstacle identification and avoidance on a Unitree Go2 quadruped. AprilTags provide obstacle observations and a safety filter modifies the robot's motion to maintain collision-free behavior.

Legged Locomotion

Comparative Analysis of MPC, LQR, and RL Footstep Planners on Uneven Terrain

This project compares model-predictive control, linear-quadratic regulation, and reinforcement-learning footstep planners for bipedal locomotion. Velocity tracking and success rates are evaluated across varied initial conditions and uneven terrain in a Drake simulation of Cassie.

Optimization and Control

Optimization-Based Obstacle Estimation from Human Demonstrations

This project learns obstacle position and size from expert demonstrations using control-Lyapunov and control-barrier-function constraints. Gaussian mixture models and regression provide a probabilistic estimate without requiring direct obstacle sensing.

Paper Poster

Robot Manipulation

Vision-Aided Pick-and-Stack with a 7-DOF Manipulator

This project uses real-time AprilTag feedback and gradient-based inverse kinematics to pick and stack blocks with a 7-DOF Franka Emika Panda. Collision-aware waypoint trajectories support both static and moving-block manipulation tasks.