Research Scientist - World Modeling

About The Institute Of Foundation Models

The Institute of Foundation Models (IFM) at MBZUAI is dedicated to pioneering foundational model research, creating models from their inception, thoroughly analyzing them, and publishing influential research that shapes the domain. Join a team of esteemed researchers and engineers tackling challenges that define the next generation of models.

The Role

Be a part of the PAN world model initiative, which focuses on developing foundational models that can simulate, predict, and interact with the physical environment. As a Research Scientist, you will lead research on large-scale video generation, interactive and action-conditioned world models, with applications in robotics and embodied AI, publishing innovative findings and transitioning research into operational systems.

Responsibilities

• Conduct innovative research on video world models, video diffusion models, and action-conditioned generation from conceptualization to publication and deployment.
• Develop pre-training and post-training strategies for large diffusion transformers, including studies on scaling laws in video pre-training.
• Advance the development of video action models and their application in robotics and embodied agents.
• Create rigorous evaluation benchmarks assessing physical accuracy, controllability, and interactivity of models.
• Collaborate with engineering and data teams for large-scale training, data curation, and simulation-based data generation.

Requirements

• PhD in Machine Learning, Computer Science, Computer Vision, Robotics, or related disciplines with first-author publications in leading conferences such as NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, RSS, or CoRL.
• Demonstrated research experience with advanced video generative models and world models, comparable to systems like Cosmos-3, LTX 2.3, Self-Forcing, or Lingbot-World.
• Expertise in at least one of the following: comprehensive data pipelines for large-scale video/simulation data including annotation and filtering; large-scale diffusion transformer training on GPU clusters; integration of rendering engines and simulation environments using Unreal Engine or Blueprints; development of world action models and their applications in robotics.
• Strong systems engineering skills with deep learning frameworks such as PyTorch.
• Proficiency with modern AI coding agents and web-based coding platforms (e.g., Claude Code, Codex, Cursor) to boost research productivity.
• Excellent problem-solving skills and capacity to work effectively in ambiguous, fast-evolving research contexts.

Preferred Qualifications

• Experience in accelerating diffusion model inference including distillation, few-step generation, or real-time interactive generation.
• Knowledge of visual tokenization and multimodal foundational models.
• Practical experience deploying world models in robotics or embodied AI environments.