Sponsors
  • Habsida
  • Figma
  • Rezi
  • Notion for Startups
  • Habsida
  • Figma
  • Rezi
  • Notion for Startups
  • Habsida
  • Figma
  • Rezi
  • Notion for Startups

Stay updated with our latest job postings by following us on LinkedIn and join our Discord community for daily notifications.

Dev Korea

Cross-Embodiment Learning Engineer

Seongnam (On-site) • Full-time

  • KOREAN: BASIC
  • AI
  • Machine Learning
  • Python
  • Robotics
Apply ↗

Insights about this position

  • Visa sponsorship: No

    The company cannot sponsor a Korean work visa for this role.

  • Korean language proficiency: Basic

    The candidate can ask and answer simple questions and handle basic everyday interactions in Korean.

  • Workplace type: On-site

    Work is primarily performed at the office.

한국어 공고는 아래에 있습니다.

In this role, you normalize the states/actions of different robots, arms, and grippers into a shared representation and integrate robot-free human demonstration data (UMI, ego-centric video, and more) so that a single policy and dataset can transfer and learn across diverse embodiments and also own the scalability of our stack-learn once, extend to every robot.

Key Responsibility

  • Normalize states/actions across different embodiments (robots, arms, grippers) into a shared representation
  • Integrate and leverage robot-free human demonstration data (UMI, ego-centric video, etc.)
  • Develop learning methods that transfer a single policy/dataset across diverse embodiments
  • Design benchmarks that quantify cross-embodiment generalization

Requirements

  • 3–5 years of research/engineering experience in cross-embodiment learning, transfer learning, or robot manipulation
  • Experience with imitation learning, representation learning, or multi-embodiment policy learning
  • Strong model training skills in Python and PyTorch
  • Experience working with diverse robot form factors or demonstration data
  • Master's degree in CS, AI, Robotics, or equivalent experience

Preferred

  • Research track record in cross-embodiment learning, UMI, or ego-centric video based learning
  • Experience with representation learning or domain adaptation
  • Experience with diverse robot/gripper hardware
  • Master/PhD in Robotics or AI
  • Experience building/using large heterogeneous robot datasets

Benefits

  • No restrictions on the use of AI tools (Claude).
  • We minimize unnecessary meetings and make decisions quickly.
  • Modern intranet/tools — Google Workspace, Slack, Notion, Linear, Workable, Flex.team, etc.

서로 다른 로봇·arm·gripper의 state/action을 공통 표현으로 정규화하고, UMI(Universal Manipulation Interface), ego-centric video 등 robot-free human demonstration 데이터까지 통합 활용하여, 하나의 policy와 데이터를 다양한 embodiment로 전이, 학습할 수 있는 기술을 개발하는 역할입니다. '한 번 배운 것을 모든 로봇으로 확장'하는, 스택의 확장성을 책임집니다.

주요업무

  • 서로 다른 embodiment(robot·arm·gripper)의 state/action을 공통 표현으로 정규화
  • UMI, ego-centric video 등 robot-free human demonstration 데이터의 통합·활용
  • 하나의 policy·데이터를 다양한 embodiment로 전이(transfer)하는 학습 기법 개발
  • Cross-embodiment 일반화 성능을 정량 평가하는 벤치마크 설계

Requirements

  • Cross-embodiment learning, transfer learning, 또는 robot manipulation 관련 3~5년의 연구·개발 경험
  • Imitation learning, representation learning, 또는 multi-embodiment policy 학습 경험
  • Python·PyTorch 기반 모델 학습 역량
  • 다양한 로봇 형태 또는 demonstration 데이터를 다뤄본 경험
  • 컴퓨터공학·AI·로봇공학 관련 석사 이상 또는 그에 준하는 경험

우대사항

  • Cross-embodiment learning, UMI, ego-centric video 기반 학습 관련 연구 실적
  • Representation learning 또는 domain adaptation 경험
  • 다양한 robot·gripper 하드웨어를 다뤄본 경험
  • 로봇공학·AI 석/박사 학위
  • 대규모 heterogeneous 로봇 데이터셋 구축·활용 경험

Benefits

  • Claude 등 AI 도구 사용에 제한이 없습니다.
  • 불필요한 회의를 최소화하고 빠르게 의사결정합니다.
  • Modern intranet/tools - Google Workspace, Slack, Notion, Linear, Workable, Flex.team 등.
Friendly Reminder

Help Dev Korea Grow!

If you get an interview, please mention you discovered this job on Dev Korea. This helps our platform stay active, and full of great opportunities. Thanks!!

Continue to Application →