Sewon Kim

Optimal Control and Reinfocement Learning

I am Sewon Kim (pronunciation: "seh-won"), an undergrad student in electrical and computer engineering at the University of Seoul. I am interested in developing scalable learning and control methods and their theoretical foundations for robotics.

I am originally from Seoul, South Korea 🇰🇷. I have a passion for jazz and heavy metal 🎸, and I love workout and traveling.

First-Order Methods for Trilevel Optimization

Sewon Kim, Hyunwoo Lee, Jeongyeol Kwon, and Dohyun Kwon. Under review at NeurIPS 2026.

[11/27/2025-01/21/2026] Autonomous Dual Arm Gearbox Assembly in Isaac Sim

Finite State Machine Object Pose Estimation Inverse Kinematics

[08/25/2025-09/25/2025] Autonomous Air Purification Agent

🏆 4th Place (/200) ($2,200) — SK NAMUH X A1: Autonomous Driving Path Optimization

I independently implemented full-stack autonomous air purifying agent in NumPy from scratch, covering MCL, MPPI, and TSP.

Monte Carlo Localization Traveling Salesman Problem Model Predictive Path Integral Control

[03/01/2024-12/19/2024] Autonomous Racing F1TENTH

5th Place (/12) — 22nd F1TENTH Autonomous Grand Prix at the Conference on Decision and Control 2024, Milan

F1TENTH is a 1/10-scale autonomous racing platform designed for developing algorithms for high-speed, collision-free autonomous driving. I formulated autonomous racing as a two-stage problem: generating a time-optimal racing trajectory offline (global planning via trajectory optimization), and tracking the reference while avoiding collisions online (local planning via MPPI). We built a full-stack autonomous navigation system for racing completely from scratch, covering SLAM, localization, trajectory optimization, MPPI, moter control, and hardware. I led the project with two undergraduate students.

Website | Code | LinkedIn

SLAM Monte Carlo Localization Race Line Optimization Model Predictive Path Integral Control

[07/25/2024–09/27/2024] Cart Pole System Design and Control with NMPC and PID

🏆 3rd Place (/25) ($220) — UOS ECE Innovation Fair

The Cart-Pole, a classic benchmark problem in control, consists of a cart driven by a linear actuator and a pole attached via an unactuated joint. If the pole is positioned vertically, its dynamics predict it will fall due to its instability. The goal of the controller is to swing the pole up and balance it above the cart. I independently built a physical Cart-Pole system for $272 and applied NMPC, covering 3D modeling, sensor integration, motor control, state estimation, and embedded systems. It marked the beginning of my journey in robotics.

Website | Code | LinkedIn

Model Predictive Control PID Hardware Mechanical Design Embedded Electronics State Estimation Stepper Motor

[01/05/2024-01/11/2025] InBody AI Scale

Consumer Electronic Show 2025, Las Vegas

We (7 members and I) developed InBody AI Scale, fully-automated body composition analyzer, from scratch that reduces measurement time by 67% compared to the original model. I developed height estimation model (acc 99.4%) and face recognition system of "InBody AI Scale". We developed a near-production-ready product.

Website

Computer Vision AI Height Estimation Face Recognition Segmentation Detection

[07/01/2023-07/31/2023] ECG-Based Cardiovascular Diagnosis Device Using ML

🏆 3rd Place (/12) ($1460) — InBody Makathon 2023

We rapidly prototyped the product.

Website | Code

CNN ML ECG

[07/24/2023-07/28/2023] Autonomous Mission Mobile Robot

🏆 3rd Place (/20) ($220) — Object Detection-Based Solar Tracking Robot Challenge, Korea University

Website | Code

PID Monocular Camera Ultrasonic Sensor

[03/07/2022-01/31/2023] Sidaeting

1,700 User Base

Our team, UOSLIFE, launched the dating service “Sidaeting,” serving 1,700 users. I developed the app’s frontend.