Robotics & robot learning.
The problems, tools, and decisions behind the work.
PROFESSIONAL WORK · SAMSUNG ELECTRO-MECHANICS
Learning-based manipulation in simulation.
I develop robotic manipulation environments and learning pipelines using NVIDIA Isaac Sim and Isaac Lab. The work connects task definition, observations and actions, reward design, policy training, rollout evaluation, and failure analysis.
Teleoperation and useful demonstrations.
VR teleoperation connects an operator’s motion to a robot’s target pose. I build workflows for collecting demonstrations and preparing observations and actions for imitation-learning experiments.
The engineering questions are practical: which coordinate frame does a pose belong to, what does an action represent, and how do you tell a usable demonstration from a misleading one?
RELATED ENGINEERING NOTE
From a gamepad to VR: teleoperating an SO-101 in Isaac Sim
The IK behind moving a robot with a VR controller, how my Quest connects to Isaac Sim, and what a recorded demonstration looks like before it becomes a LeRobot dataset.
Read the walkthrough ↗From perception to action.
My background in production computer vision informs the way I approach RGB observations, robot proprioception, and experimental evaluation. The goal is a control pipeline whose failures can be understood and improved.