Tri Wahyu Guntara

Applied Research Engineer (RL), KRAFTON

I build reinforcement learning agents at KRAFTON, first for games (PUBG) and now for humanoid robots, and I am currently on dispatch to Ludo Robotics, KRAFTON's robotics spin-off. I did my master's at KAIST, advised by Kee-Eung Kim.

wahyu.guntara@gmail.com

GitHub LinkedIn

Experience

2026.07 - Present

Research Engineer, Ludo Robotics Seoul, South Korea

Concurrent role, on dispatch from KRAFTON

Built a Pico VR teleoperation system for the Unitree G1 humanoid with locomotion, waist, and arm control, used to collect real-robot data
Fine-tuned vision-language-action (VLA) models on the collected data and deployed them on the Unitree G1
Systems integration for Ludi 0.1, an agentic system for socially intelligent robots, on the real robot

2024.11 - Present

Applied Research Engineer (RL), KRAFTON Inc. Seoul, South Korea

2025.12 - Present Physical AI Team

Trained RL agents for non-perceptive locomotion, and for perceptive navigation on top of the locomotion policy, on the Unitree G1 humanoid
Fine-tuned GR00T and π-family vision-language-action (VLA) models on the AgiBot G1 humanoid in simulation and added a HAMLET memory module, which significantly improved task success
Owned the simulation evaluation pipeline: ported GenieSim, an asynchronous Isaac Sim environment for the AgiBot G1, to a synchronous Isaac Lab implementation that runs parallel instances for scalable evaluation

2024.11 - 2025.11 Reinforcement Learning Team

Trained an RL agent for the PUBG parachute-landing task to replace the rule-based bot, cutting mean landing error from 40 / 60 / 523 m to 18 / 19 / 24 m on short (<1 km), medium (1–1.7 km), and far (>1.7 km) drops
Trained an RL agent that plays the full PUBG game, deployed as the PUBG AI Chicken Challenge, a limited-time player-vs-AI event at PUBG UNITED: HOMECOMING (Dec 2025 – Jan 2026)
Early development of a full-game RL agent for PUBG: Blindspot, a top-down tactical shooter in development at KRAFTON

2022.01 - 2024.08

Graduate Student Research Assistant, Korea Advanced Institute of Science and Technology (KAIST) South Korea

2023.01 - 2024.08 Artificial Intelligence & Probabilistic Reasoning Lab

Reinforcement learning and thesis project under the advisory of Prof. Kee-Eung Kim

2022.01 - 2022.12 Autonomous Vehicles and Embedded Systems Lab

Self-driving car and reinforcement learning under the advisory of Prof. Seung-Hyun Kong

2019.06 - 2019.12

Research Scientist Intern, Kata.ai Indonesia

Benchmarked models on Indonesian sequence labeling tasks (POS tagging, NER, SRL)
Collected scattered Indonesian-English data and benchmarked models on neural machine translation

Publications

2026

P2Ludi 0.1: An Agentic System for Socially Intelligent Robots

Wooseong Chung, William Cong, Jakub Dworakowski, Ethan Ewer, Tri Wahyu Guntara, Yeonwoo Jeong, Tianchong Jiang, Chaewon Kim, Hyunseo Kim, Jinwoo Kim, Jinyeon Kim, Yea-Seul Kim, Jack Kunde, Kangwook Lee, Sangheon Lee, Robert Nowak, Junha Roh

Ludo Robotics Technical Report

paper project page
C2ACPO: Agent-Chained Policy Optimization for Multi-Agent Reinforcement Learning

Daiki E. Matsunaga, Junho Na, Tri Wahyu Guntara, Scott Sanner, Pascal Poupart, Jongmin Lee, Kee-Eung Kim

Reinforcement Learning Conference (RLC) 2026

paper

2025

P1CLEAR: An Information-Theoretic Framework for Distraction-Free Representation Learning in Visual Offline RL

Tri Wahyu Guntara, Daiki E. Matsunaga, HyeongJoo Hwang, Kee-Eung Kim

Preprint

paper code project page

2024

C1Kernel Metric Learning for In-Sample Off-Policy Evaluation of Deterministic RL Policies

Haanvid Lee, Tri Wahyu Guntara, Jongmin Lee, Yung-Kyun Noh, Kee-Eung Kim

ICLR 2024

paper code Spotlight

2020

W1Benchmarking Multidomain English-Indonesian Machine Translation

Tri Wahyu Guntara, Alham Fikri Aji, Radityo Eko Prasojo

LREC Workshop on Building and Using Comparable Corpora (BUCC) 2020

paper

Education

2021.09 - 2024.08

M.S. in Artificial Intelligence, Korea Advanced Institute of Science and Technology (KAIST) (Advisor: Kee-Eung Kim)

Thesis: Information-Theoretic Distraction-Free Representation Learning for Visual Offline RL

2016.09 - 2020.08

B.Eng. in Electrical Engineering, Universitas Indonesia

Projects

Ludi 0.1: An Agentic System for Socially Intelligent Robots

Agentic Robotics System: a robot that holds a spoken conversation while it perceives, navigates, and manipulates, driven by a fine-tuned vision-language model that reasons over context and calls the navigation and manipulation policies as tools. Contributed the vision-language-action (VLA) policies and the systems integration behind deploying it on the real robot.

project page technical report
CLEAR: Controllable Latent State Extractor for Visual Offline RL

Master's thesis project on learning agent-centric latent representations that remain robust when the visual observation is full of distractions.

project page paper code
PALAPA-707: An Autonomous VTOL Drone for the Indonesia Flying Robot Contest 2018

Undergraduate team project at Universitas Indonesia: an autonomous drone for payload pick-up and delivery, built around a PID controller with vision-based target detection.

project page source code

Scholarships & Awards

2021.09 - 2023.08

Hyundai Motor Chung Mong-Koo Global Scholarship

Full scholarship for the duration of the 2-year master's study at KAIST

2018.11

Indonesia Flying Robot Contest 2018 — Honorable Mention

Vertical take-off and landing (VTOL) category