The posting
Action-conditioned world model을 학습하여 로봇의 미래 상태를 예측하고, virtual rollout과 action search를 통해 더 높은 성공 가능성을 가진 action을 생성·선별하는 역할입니다. 로봇이 '행동하기 전에 상상하고 검증'할 수 있게 만드는, Laplacian AI Core의 예측·계획 축을 담당합니다
In this role, you will train the action-conditioned world models to predict a robot's future states, and use virtual rollouts and action search to generate and select actions with higher probability of success. You will also own the "prediction-and-planning axis of AI Core", letting robots imagine and verify before they act.
주요업무 (Key Responsibility)
- Action-conditioned world model 학습 및 로봇 미래 상태 예측 모델 개발
- Virtual rollout을 통한 action 후보 시뮬레이션 및 성공 가능성 평가
- Action search·planning을 통해 더 나은 action을 생성·선별하는 파이프라인 구축
- World Model의 예측 정확도·계획 성능을 정량 평가하는 지표 설계
- VLA·policy 팀과 협업하여 world model 기반 planning을 실제 policy에 통합
- Train action-conditioned world models and develop future-state prediction for robots
- Simulate candidate actions via virtual rollout and assess success likelihood
- Build action-search/planning pipelines that generate and select better actions
- Design metrics that quantify world-model prediction accuracy and planning performance
- Collaborate with VLA/policy teams to integrate world-model-based planning into real policies
Requirements
- World model, model-based RL, 또는 video/dynamics prediction 관련 3~5년의 연구·개발 경험
- Action-conditioned generative model 설계 및 학습 실무 경험
- Python·PyTorch 기반 시퀀스 모델(transformer, diffusion 등) 학습 및 분산 학습 역량
- 로봇 또는 시뮬레이션 환경에서 dynamics·prediction 모델을 활용하거나 평가해본 경험
- 컴퓨터공학·AI·로봇공학 관련 석사 이상 또는 그에 준하는 경험
- 3–5 years of research/engineering experience in world models, model-based RL, or video/dynamics prediction
- Hands-on experience designing and training action-conditioned generative models
- Strong skills in training sequence models (transformers, diffusion, etc.) in Python and PyTorch, including distributed training
- Experience utilizing or evaluating dynamics/prediction models in robotic or simulated environments
- Master's degree in CS, AI, Robotics, or equivalent experience
우대사항 (Preferred)
- World model, model-based RL, video prediction 분야 탑티어 학회 (NeurIPS, ICML, ICLR, CoRL, RSS 등) 논문 실적
- Diffusion·transformer 기반 action-conditioned dynamics 모델을 대규모로 학습해본 경험
- Long-horizon rollout 안정화, sim-to-real transfer 등 world model 실전 적용 관련 연구 경험
- 실 로봇에 model-based planning 또는 world model 기반 policy를 배포해본 경험
- 컴퓨터공학·AI·로봇공학 관련 석/박사 학위
- Publication record at top-tier venues (NeurIPS, ICML, ICLR, CoRL, RSS, etc.) in world models, model-based RL, or video prediction
- Experience training diffusion- or transformer-based action-conditioned dynamics models at scale
- Research experience in practical world model deployment, such as long-horizon rollout stabilization or sim-to-real transfer
- Experience deploying model-based planning or world-model-based policies on real robots
- Master/PhD in CS, AI, or Robotics
Benefits
- Claude & ChatGPT subscriptions provided (Claude · ChatGPT 유료 플랜 지원)
- Minimal meetings with fast decision-making (불필요한 회의를 최소화하고 빠르게 의사결정합니다.)
- Modern intranet/tools (Google Workspace, Slack, Notion, Linear, Workable, Flex.team 등)
