Keonyoung Koh

I'm a master's student at RIRO Lab at KAIST, supervised by Daehyung Park.

My current research focuses on data-driven decision-making for mobile robotics, particularly in mobile manipulation under real-world conditions.

I am also interested in applying imitation learning, multi-modal large language models (MLLMs), and vision-language-action (VLA) models to robot learning and planning.

Email / CV / Scholar / LinkedIn / Github

Publications

A Real-World Based 3D Simulation and Robotic Framework for Semantic-Aware Outdoor Navigation
Keonyoung Koh, Moonkyeong Jung, Samuel Seungsup Lee, Daehyung Park
Submitted to Korea Robotics Society Annual Conference (KRoC) (Under Review), 2026

SuReNav: Superpixel Graph-based Constraint Relaxation for Navigation in Over-constrained Environments
Keonyoung Koh, Moonkyeong Jung, Samuel Seungsup Lee, Daehyung Park
Submitted to International Conference on Robotics and Automation (ICRA) (Under Review), 2026

Constraint Relaxation for Urban Environment Navigation
Jinwoo Kim*, Keonyoung Koh*, Samuel Seungsup Lee, Yohan Park, Daehyung Park
The 12th International Conference on Robot Intelligence Technology and Applications (RiTA), 2024
Best Student Paper Award
[PDF]

Researches and Projects

[Research] Safe Local Planner Using Generative Models
RIRO Lab, 2025 ~ Current

Trained and reproduced diffusion policy-based navigation model (NoMaD)
Working on safe local planning research based on diffusion steering

[Research] Imitation Learning-based Human Like Constraint-aware Robot Navigation
RIRO Lab, 2025 ~ Current

Formulated constraint relaxation problem for human-like safe and efficient robot navigation in real-world outdoor environments
Applied imitation learning based on search-guided differentiable planner to imitate human decision-making in semi-static, over-constrained situations
Working on latent intention imitation of heterogeneous, mixed-modality demonstration dataset using conditional variational inference

[Project] Humanoid SLAM and VLA-based Control
RIRO Lab, 2025

Implemented metric SLAM system and applied on Unitree G1 robot for autonomous navigation “[VIDEO LINK]”
Fine-tuned NVIDIA's GR00T model using G1 block stacking dataset & tested on Isaac Sim

[Research, Project] Autonomous Navigation System Development for Quadrupedal Robot
RIRO Lab, 2023–2024

Developed semantic SLAM & intelligent planning system for quadrupedal robot's autonomous navigation
Real-world deployment on Boston Dynamics SPOT robot to test mapless/map-based outdoor pointgoal navigation

[Project] Reproduction of “On the Planning Abilities of Large Language Models”
AI602: Advanced Deep Learning, Fall 2024

Reimplemented key experiments from “On the Planning Abilities of Large Language Models: A Critical Investigation” by Valmeekam et al. (NeurIPS 2023)
Evaluated LLMs such as GPT-4 on task planning benchmarks and analyzed error modes
Extended the evaluation to SOTA LLMs with advanced prompting techniques and improved performance

[Project] Graph Relation Prediction for Robotic Goal State Design
CS491: Graph Machine Learning and Mining, Spring 2024

Curated knowledge graph dataset by parsing WikiHow data, consists of task–quadruple set
Implemented & trained composite knowledge embedding for goal state prediction of robotic task planning

[Project] Hovercraft Design
ME400: Capstone Design & ME454: Dynamics System Programming, Spring 2023

Designed HW/SW architecture of hovercraft, mainly worked on control software development
Implemented localization & subgoal-based reactive path planning algorithm
Applied adaptive controller/sliding mode controller to hovercraft, won 1st place on the competition

[Project] Traversability Estimation for Robust Robot Locomotion
CS470: Introduction to Artificial Intelligence, Fall 2022

Reimplemented traversability estimation network from "Locomotion Policy Guided Traversability Learning using Volumetric Representations of Complex Environments" by Frey et al. (IROS 2022)
Generated randomized terrain dataset and trained the network to deploy on real-world quadrupedal robot

[Project] Reproduction of "PoseScript: 3D Human Poses from Natural Language"
CS492(I): Special Topics in Computer Science – Introduction to Deep Learning, Fall 2022

Reimplemented pose generation/retrieval network from "PoseScript: 3D Human Poses from Natural Language" by Delmas et al. (ECCV 2022)
Altered network configuration and trained on paper's dataset to improve generation & retrieval performances

[Project] Internship at Angel Robotics
Co-op at Angel Robotics, Spring 2022

Worked as a robot software engineer at Angel Robotics, a startup developing exoskeleton robot to assist patients with paraplegia
Developed core softwares (firmware update code based on hierarchical SW architecture, adaptive controller for robotic parameter estimation, IMU filtering algorithm for robot pose estimation)

[Project] Pick-and-place Robotic Manipulation Simulation
CS492(I): Special Topics in Computer Science – Introduction to Intelligent Robotics, Fall 2021

Simulated GAZEBO-based UR5 manipulator
Applied subgoal-based planning method for Amazon pick-and-place contest

Extracurricular Activities

Served as a teaching assistant of CS470: Introduction to Artificial Intelligence (Fall 2024), Outstanding TA Award
Served as a teaching assistant of CS477: Introduction to Intelligent Robotics (Spring 2024)
Software engineering intern at a startup developing exoskeleton robots for paraplegic patients (Spring 2022)
Served as Proctor, mentoring incoming freshmen as part of a student guidance program (2022)
Served as Vice Chair of the Student Welfare Committee, contributing to the planning of student events and the development of welfare policies (2019-2021)

Website adapted from Jon Barron. Source available on GitHub.