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

Meta•Redmond, WA

10d•$7,650 - $12,134

About The Position

Reality Labs Research is seeking a Research Scientist Intern to help develop the next generation of assistance systems that guide users in contextual and adaptive environments. We welcome candidates with expertise in AI-driven 3D spatial understanding and generative modeling, building and iterating on deep learning systems, with specific experience at the intersection of generative AI (e.g., diffusion models), world simulation (predicting physics and future states), and 3D computer vision (reconstruction, camera geometry, depth estimation, motion/trajectory forecasting). Our internship is for Fall 2026 and is twelve (12) to twenty-four (24) weeks long.

Requirements

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

Nice To Haves

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

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.