Yongseok Kwon

I am a robotics researcher in the ROAHM Lab, where I advised by Prof. Ram Vasudevan. Also, I received my BS(‘20) at UNIST, and MS(‘22) at the University of Michigan.

My research aims to create reliable and safe algorithms that bring autonomous robots into our daily lives. To achieve this, I focus on developing methods that identify, represent, and enforce safety constraints for variety of robotic tasks, ensuring the reliable deployment of robotics in real-world applications.

CV • Github • G. Scholar • LinkedIn

kwonys [at] umich [dot] edu

Publications & Preprints

Conformalized Reachable Sets for Obstacle Avoidance With Spheres

Yongseok Kwon, Jonathan Michaux, Seth Isaacson, Bohao Zhang, Matthew Ejakov, Katherine A. Skinner, and Ram Vasudevan

In Submission, 2024

project page • paper • code

Reachability-based Trajectory Design with Neural Implicit Safety Constraints

Jonathan Michaux, Qingyi Chen, Yongseok Kwon, and Ram Vasudevan

Robotics: Science and Systems (RSS), 2023

project page • paper • code

Software Contributions

ZONOPY: Zonotopes in Python

A Python library for handling various continuous sets (e.g., intervals, zonotopes, and polynomial zonotopes) to compute reachable sets in robotic arm kinematics and dynamics, with support for parallel computation.

▸ zonopy : code • documentation

▸ zonopy-robots : code • documentation

Projects

Transformer for Reaching Task

Fall 2021 at the University of Michigan

Trained a Decision Transformer for the reaching task of a 7-DOF robotic arm using an offline reinforcement learning

Contact Simulation via Linear Complementary Problem

Fall 2021 at the University of Michigan

Implemented contact simulator with LCP (Linear Complementary Problem) formulation of contact dynamics.

Trajectory Optimization for Autonoumous Car

Fall 2021 at the University of Michigan

Implemented convex collision avoidance constraint for lane changing task of autonomous car.

Extended Kalman Filter on Neural Gait Model

Fall 2021 at the University of Michigan

Implemented extended Kalman filter on a gait measurement model trained with a neural network and tested the state estimation algorithm on an open-source robotic leg.