Moonyoung (Mark) Lee

I am a PhD student at the Carnegie Mellon University, Robotics Institute where I work on robot learning for manipulation. I am co-advised by Oliver Kroemer and George Kantor.

From 2017 to 2020, I worked at KAIST Humanoid Research Center, the winning team of the DARPA Robotics Challenge. As a research engineer, I worked on computer vision for real-time terrain mapping as well as 3D object pose estimation for pick-and-place tasks.

I graduated from Cornell University (B.S & M.Eng) in electrical engineering. Previously I interned at iRobot and New York University.

GitHub  /  Google Scholar  /  LinkedIn

Research

I'm interested in contact-rich manipulation using multi-modal learning (vision, tactile) & learning dynamics model of deformable objects for sim2real.

Towards Autonomous Crop Monitoring: Inserting Sensors in Cluttered Environments

We present a robot platform that autonomously detects and inserts nitrate sensors into corn stalks. The robot was deployed in a cornfield in Iowa for evaluation.

Towards Robotic Tree Manipulation: Leveraging Graph Representations

We present a framework for learning the deformation behavior of trees under contact interaction. Graph neural network is used to learn the forward model and action policy to manipulate trees.

Task-Oriented Active Learning of Model Preconditions for Inaccurate Dynamics Models

We present an active learning algorithm that selects trajectories to learn model preconditions for planning with an inaccurate pre-specified dynamics model.

3D Reconstruction-Based Seed Counting of Sorghum Panicles for Agricultural Inspection

We present a method for creating high-quality 3D models of sorghum panicles to estimate seed counts. This is acheived using seeds as semantic 3D landmarks for global registration and a novel density-based clustering approach.

Vision-based Detection and Tracking for Relative Localization of Aerial Swarms

We present evaluations on vision-based decentralized Bayesian multi-tracking filtering strategies to resolve the association between the incoming unsorted measurements obtained by a visual detector algorithm and the tracked agents. We show computation tradeoff on running inference and tracking pipeline on-board.

Dynamic Humanoid Locomotion Over Rough Terrain With Streamlined Perception-Control Pipeline

We present a geometric footstep planner and walking controller for a humanoid robot to dynamically walk across rough terrain at speeds up to 0.3 m/s. All visual sensing and compute are done on-board.

Joint Space Position/Torque Hybrid Control of the Quadruped Robot for Locomotion and Push Reaction

We present a novel algorithm for joint space position/torque hybrid control of a mammal-type quadruped robot. With this control algorithm, the robot demonstrated both dynamic locomotion and push recovery abilities without torque control in the ab/ad joints.

Fast Perception, Planning, and Execution for a Robotic Butler: Wheeled Humanoid M-Hubo

We present a new robotic butler system for a wheeled humanoid that is capable of fetching requested objects at 24% of the speed a human needs to fulfill the same task. We achieve this speedup by integrating a 3D object detection pipeline with a kinematically optimal manipulation planner to significantly increase speed performance at runtime.

Motion Generation Interface of ROS to PODO Software Framework for Wheeled Humanoid Robot

We present a new motion generation interface between ROS and PODO that enables users to generate motion trajectories through standard ROS messages while leveraging a real-time motion controller.

Design and source code from Jon Barron's website