PhD Researcher Intern, World Foundation Models

NVIDIA is seeking PhD research intern to advance world foundation models within NVIDIA Cosmos for Physical AI. Our research explores new model architectures, training objectives, and efficient training and inference for multimodal generation and action-conditioned prediction. We are particularly interested in physical plausibility, spatial and temporal consistency, and the modeling of human appearance, motion, and interactions with objects and environments. You will work with NVIDIA researchers and engineers to define and pursue a focused research project, develop new methods, and demonstrate measurable improvements through rigorous experimentation. Projects will reflect team priorities and your research strengths. We welcome candidates with relevant research experience in generative modeling, computer vision, multimodal learning, computer graphics, or robotics. What you'll be doing: • Research and develop model architectures, training objectives, and learning algorithms for video generation and world modeling, including diffusion or flow-matching approaches, multimodal conditioning, and spatial and temporal representations. • Design controlled experiments and ablation studies, combining quantitative benchmarks with qualitative evaluation to diagnose model behavior, identify failure modes, and validate improvements. • Explore training and inference efficiency through architectural changes, distillation, model compression, or other optimization techniques. • Develop data curation and evaluation workflows, including model-assisted filtering, annotation, and quality assessment, and measure their impact on model performance. • Throughout your project, you will collaborate with research and engineering teams and share findings through reproducible implementations, technical reports, and presentations. What we need to see: • Currently pursuing a PhD in Computer Science, Computer Graphics, Computer Engineering, Robotics, or a closely related field. • Research experience in one or more relevant areas, such as generative modeling, computer vision, video understanding or generation, multi-modal learning, 3D vision, or robotics. • Strong proficiency in Python and PyTorch, demonstrated through research implementations or substantial projects. • Hands-on experience implementing, training or fine-tuning, and evaluating deep learning models. • Ability to formulate research questions, design rigorous experiments, analyze results, and communicate findings clearly. • Comfortable working in Linux and using standard development practices for ML research, including version control, testing, and experiment reproducibility. Ways to stand out from the crowd: • Publications at venues such as NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, or related conferences, or substantial research projects demonstrating original technical contributions. • Background with human appearance or motion modeling, human-object interactions, 3D generative models, 3D/4D representations, or neural rendering. • Experience assessing physical plausibility, diagnosing visual and temporal artifacts, or developing evaluations using perceptual metrics, human feedback, or learned evaluators. • Experience with distributed training, GPU performance profiling, efficient data loading, or model acceleration. • Prior experience with NVIDIA Cosmos is welcome but not required. Join us to help build the next generation of world models for Physical AI! NVIDIA is committed to fostering an inclusive work environment and proud to be an equal opportunity employer. As we highly value diversity in our current and future employees, we do not discriminate (including in our hiring and promotion practices) on the basis of race, religion, color, national origin, gender, gender expression, sexu