About
I am an M.S. student in Robotics at the University of Michigan. I work on multi-robot navigation and multi-agent path finding, with a focus on crowded spaces where robots keep blocking each other.
Before Michigan I spent three years in industry. At Floatic I built the navigation stack and middleware for a fleet of warehouse robots. At Metafarmers I led a seven-person robotics team that put two agricultural robots into production.
I am looking for PhD positions starting in Fall 2027.
Professional experience
- Built and led a seven-person robotics AI team, and set its structure, technical roadmap and system architecture.
- Owned the Omni Farmer robotics stack from architecture to production, with two robots operating in unstructured smart-farm environments.
- Designed the multi-robot management system and its infrastructure so the deployment could grow to larger fleets.
- Designed the robot software architecture for Flody, including a middleware layer that kept robot code independent of the backend, so Kafka and AWS IoT could be swapped without touching the robots.
- Built and deployed multi-robot navigation across three warehouse sites: two partner-operated production facilities and one internal testbed.
- Ran a fleet of 10 robots for more than 300 hours of live operation and cut fleet deadlocks from 3–4 per hour to about one every two hours.
Education
Research experience
Publications
Figures reproduced from the papers under CC BY-NC 3.0.
Projects
Industry
Omni Farmer: agricultural robots
Metafarmers, 2025–2026Strawberry harvesting and pollination robots. Defined the ROS 2 architecture from the motor control boards up to navigation; long-term SLAM fusing FAST-LIO with Cartographer kept localization stable across a 100 × 200 m site. Two robots went into production.
Apple-harvesting exhibit robot
Metafarmers, 2026A robot delivered to the Gwacheon National Science Museum for its farm-technology exhibition, where it recognizes apples and picks them without bruising using a gripper from the Rural Development Administration.
Flody: warehouse robot fleet
Floatic, 2022–2024Robots shared their poses and planned paths over a site-local Redis bus and added nearby robots to their own costmaps. Fleet deadlocks went from 3–4 per hour to about one every two hours. The system ran 10 robots for more than 300 hours across 3 warehouse sites.
Contract and lab work
Churo: dog behavior recognition
ROBOI contract, 2021Behavior recognition for a dog-care robot. YOLOv5 and SORT followed the dog, a labeling tool built 20-frame clips, and a ResNet-CRNN classifier sorted them into five behaviors at 90.4–96.8% test accuracy on 344 clips.
Item Assembly: stereo hole detection
AIRO Lab, 2020Pose estimation for screw insertion in a government-funded furniture-assembly challenge. YOLOv5 found screw holes, a pattern match removed false hits, and stereo triangulation gave a 3D pose that ROS services passed to the arm.
Undergraduate projects
Help: marine rescue-assist robot
Capstone, 2020–2022An unmanned robot for marine accidents. It detects capsizing from IMU tilt and alerts over LoRa, finds people in the water with RGB and thermal cameras, and reports their GPS position until rescuers arrive. Special Prize at the 2022 National Capstone Design Competition, plus two awards in 2021.
Latte Panda: recycling and donation robot
Intelligent Robot Crash Lab, 2020A ROS service robot on a LattePanda and Jetson TX2. YOLOv5 found people and classified cups as paper or plastic, a depth grid handled obstacle avoidance, and a touchscreen walked visitors through sorting and a QR donation. Team of 6.
Pattern inspection with two industrial cameras
UNIEYE, Feb–Jul 2020A C++/Qt inspection program on a Jetson Nano with two Basler cameras. It found circular marks with a hand-written Gaussian filter and Canny edge detector and passed or failed each part. I built the interface and the communication on a team of two.
Open source
nav2_core::NavigatorBase.rcl_wait returning an error when a timer is cancelled. Merged.RewrittenYaml. Merged.Study notes
robotics-study.github.ioI maintain robotics-study, a set of open study notes in English and Korean. Each algorithm is implemented in both C++ and Python and replayed step by step in the browser.
Multi-robot motion planning
Prioritized A*, CBS, PIBT, RHCR, dRRT*, db-CBS
Navigation planning
Graph search, sampling planners, DWA, MPC, MPPI
Modern Robotics
All 13 chapters of Lynch and Park, with live figures
Also: notes for ROB 501 (proofs to the Kalman filter), and 45 Korean posts on ROS 2 and Nav2 internals on my blog.