Junhyeok Choi

I am a Combined MS/PhD student at Seoul National University, where I am a member of PI Lab, and a Research Associate at Maum.ai WoRV (Frontier Lab).

Before graduate school, I worked at Maum.ai as a software engineer and robotics & AI researcher, including leading the robot development team. My research interest is in making learning-based systems more capable and dependable in real technical workflows, and in the software architecture and tooling that shape what researchers can build reliably.

Email  /  CV  /  Google Scholar  /  GitHub  /  LinkedIn

Photo of Junhyeok Choi

News

Research & Projects

Selected publications, research, and engineering projects.

Mine-JEPA: In-Domain Self-Supervised Learning for Mine-Like Object Classification in Side-Scan Sonar Taeyoun Kwon, Youngwon Choi, Hyeonyu Kim, Myeongkyun Cho, Junhyeok Choi, Moon Hwan Kim CVPR 2026 Workshop on Maritime Computer Vision (MaCVi)

Side-scan sonar mine classification suffers from extreme data scarcity and a large domain gap from natural images. Mine-JEPA is an in-domain self-supervised pretraining pipeline that, trained on only about a thousand unlabeled sonar images, outperforms a fine-tuned DINOv3 foundation model on mine vs. non-mine and three-class mine-like object classification, with a compact ViT-Tiny backbone using 4× fewer parameters.

[arXiv]
RAL: Robot Abstraction Layer Junhyeok Choi, Maum.ai robot development team Maum.ai, 2025–2026 — proprietary, in production use

Robot fleets are built from hardware with different sensors, control stacks, and vendor protocols, so application logic tends to become tied to a single robot model. RAL is a unified control interface and background service that lets the same high-level code run across quadrupeds from multiple vendors, while centrally managing command and control-session lifecycles so behavior stays predictable and safe. It serves as the common foundation for the company’s robot intelligence and fleet-management applications.

Open-source tool, library, or experiment Junhyeok Choi Personal project

One or two sentences on why it exists, what you learned, and how it connects to your research direction.

[code]

Education

Experience

Certifications