About the Institute of Foundation Models
The Institute of Foundation Models (IFM) at MBZUAI is a research lab dedicated to meaningful foundation model research — building models from scratch, understanding them deeply, and publishing work that shapes the field. You’ll work alongside world-class researchers and engineers on problems that directly define the models we ship.
The Role
Join the PAN world model project — our effort to build world models: foundation models that simulate, predict, and interact with the physical world. As a Research Scientist, you’ll drive the core research behind PAN — large-scale video generation, interactive and action-conditioned world models, and their applications in robotics and embodied AI — and publish at top venues while turning breakthroughs into working systems.
What You'll Do
- Conduct original research on video world models, video diffusion models, and action-conditioned generation — from idea to publication and deployment.
- Design pre-training and post-training recipes for large-scale diffusion transformers, including scaling-law studies for video pre-training.
- Advance world action models / video action models and their applications in robotics and embodied agents.
- Develop rigorous evaluation benchmarks for physical accuracy, controllability, and interactivity.
- Collaborate with engineering and data teams on large-scale training, data curation, and simulation-based data generation.
What We're Looking For
- PhD in Machine Learning, Computer Science, Computer Vision, Robotics, or a related field, with first-author publications at top-tier venues (e.g., NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, RSS, CoRL).
- Research experience with state-of-the-art video generative models and world models (e.g., Cosmos-3, LTX 2.3, Self-Forcing, Lingbot-World, or comparable systems).
- Deep expertise in at least one of the following areas:
- Full-stack data pipelines — large-scale video data pipelines and/or simulation data collection; annotation and filtering workflows for video / world model training.
- Model training & infrastructure — training large-scale diffusion transformers on large GPU clusters.
- Rendering engines & simulation — Unreal Engine and Blueprint-based gym environments, game-engine integration, building interactive simulated environments.
- World action models & robotics — world action models / video action models, action-conditioned video generation, world-model applications in robotics.
- Strong systems and engineering expertise in deep learning frameworks such as PyTorch.
- Highly proficient with modern AI coding agents and web-based coding tools (e.g., Claude Code, Codex, Cursor), and skilled at leveraging them to dramatically accelerate research workflows.
- Exceptional problem-solving skills and the ability to navigate ambiguity in rapidly evolving research areas.
Nice To Have
- Experience accelerating diffusion model inference (distillation, few-step generation, real-time interactive generation).
- Experience with visual tokenization and multimodal foundation models.
- Experience deploying world models in robotics or embodied-AI settings.
We may use artificial intelligence (AI) tools to support parts of the hiring process, such as reviewing applications, analyzing resumes, or assessing responses and identifying potential inconsistencies or verification signals in application materials based on available information. These tools assist our recruitment team but do not replace human judgment. Final hiring decisions are ultimately made by humans. If you would like more information about how your data is processed, please contact us.
