Research Scientist Intern, 3D Vision & World Simulation (PhD) Responsibilities:

• Develop, implement, and evaluate methods for improving the performance and interpretability of VLMs and related AI/ML models.
• Write modular, reusable research code and utilize Meta’s large infrastructure to scale experimentation.
• Collaborate cross-functionally with researchers and engineers to prototype and test models at scale.
• Deliver clear, compelling, and creative solutions to challenging problems.
• Work should result in publishable research in top-tier journals or conferences (e.g., NeurIPS, ICLR, CVPR, ECCV, ICML, ICCV, AAAI, IJCAI, ICRA, IEEE T-PAMI, IJCV, IEEE RA-L etc.).
• Design and develop generative AI models (e.g., diffusion models) to learn environment dynamics, predict future states, and simulate physical interactions within world modeling frameworks.
• Build and advance 3D computer vision and reconstruction pipelines, leveraging camera geometry, depth estimation, and related techniques to create accurate representations of real-world environments.
• Research and implement methods for 3D spatial understanding, including motion forecasting and trajectory prediction, to enable intelligent reasoning about how objects and agents move through space.

Minimum Qualifications:

• Currently has, or is in the process of obtaining, a PhD in Computer Science, Artificial Intelligence, Robotics, or related field
• Strong programming proficiency in Python and extensive hands-on experience building, training, and debugging deep learning models using PyTorch
• Experience with generative AI architectures (e.g., diffusion models) and world modeling concepts (e.g., learning environment dynamics, predicting future states, or simulating physical interactions)
• Solid foundation in 3D computer vision and 3D reconstruction, including core concepts like camera geometry and depth estimation
• Experience working with 3D spatial data, motion forecasting, or trajectory prediction

Preferred Qualifications:

• Practical experience with advanced generative frameworks, specifically flow-matching or Diffusion Transformers (DiTs)
• Hands-on experience with modern 3D representations like 3D Gaussian Splatting (3DGS) and forecasting dense 3D point trajectories in physical world coordinates
• Background in robotics planning and manipulation, with experience transferring learned representations to downstream tasks like closed-loop pick-and-place or real-world robot control
• Familiarity with large Vision-Language Models (VLMs) and their application in object grounding, understanding natural language instructions, and processing multi-modal tokens
• Knowledge of extending world models to physical applications, such as latent-action world models or trajectory-conditioned video generation
• A proven track record of academic research, demonstrated by publications in top-tier conferences (e.g., CVPR, ICCV, ECCV, ICLR, NeurIPS, ICRA, RSS) in areas related to 3D vision, generative modeling, or robot learning
• Demonstrated software engineer experience via an internship, work experience, coding competitions, or widely used contributions in open source repositories (e.g. GitHub)
• Intent to return to degree-program after the completion of the internship/co-op

About Meta: