Research Scientist Intern – Robotics, World Models & Embodied AI (PhD)

Meta Reality Labs Research (RL Research) brings together a world-class R&D team of researchers and engineers with the shared goal of developing AI, Robotics, and AR/VR technology across the spectrum. The Surreal Spatial AI group is seeking Research Scientist Interns to help build the machine perception, 3D generative systems, and control technology that allow robotic agents and AI systems to perceive, understand, reason about, and interact with the world around them.

Our research puts multi-modal generative world models at the core of machine perception and robotic control, using data from egocentric devices such as Project Aria alongside data from physical robots. As an intern, you will work with our research scientists and engineers on robot foundation models, including world action models (WAMs) and vision-language-action models (VLAs). Projects can span several research streams, including exploring novel architectures, grounding these models in 3D, learning action and goal representations, and integrating new sensor modalities such as touch.

Interns in this role are proficient in research areas such as action-conditioned world models, 3D computer vision, representation learning, learning from human video and demonstrations, cross-embodiment transfer, tactile sensing, reinforcement learning, and dexterous manipulation. You should have a keen interest in producing new, open science that makes embodied agents more intelligent.

Our internships are twelve (12) to twenty-four (24) weeks long and we have various start dates throughout the year.

Responsibilities

Perform fundamental and applied research at the intersection of multi-modal generative AI, predictive world modeling, embodied reasoning, and robotic manipulation.
Collaborate with research scientists and engineers to explore novel architectures for robot foundation models, such as WAMs and VLAs.
Ground robot foundation models in 3D, using 3D computer vision and scene understanding.
Develop action and goal representations that transfer across embodiments, from human video to different robots.
Integrate new sensor modalities, such as tactile sensing and audio, into robot foundation models alongside vision and language.
Design and run empirical evaluations, writing robust code for physical hardware and simulators.
Contribute to publications and open-source releases.

Qualifications

Currently has, or is in the process of obtaining, a PhD degree in Computer Vision, Robotics, Machine Learning, Artificial Intelligence, Computer Science, or a related STEM field
Research experience in robotics, 3D computer vision, deep learning, such as multimodal or 3D generative modeling, predictive world models, scene understanding, egocentric perception, learning from demonstration, reinforcement learning, or learning-based robotic control
Experience with deep learning frameworks (e.g., PyTorch, TensorFlow) and Python
Experience communicating research to public audiences of peers
Interpersonal experience: cross-group and cross-culture collaboration
Must obtain work authorization in the country of employment at the time of hire and maintain ongoing work authorization during employment Intent to return to the degree program after the completion of the internship/co-op
Experience working with physics simulation platforms (e.g., Isaac Sim, Isaac Lab, MuJoCo)
Hands-on experience with real robot hardware, such as collecting data on, debugging, or deploying learned policies to physical robots, beyond simulation alone
Experience training or evaluating large-scale robot foundation models (e.g., VLAs, WAMs), world models, or generative video models
Experience applying 3D computer vision to embodied AI, such as 3D scene understanding, reconstruction, or 3D-aware policies and world models
Experience with action or goal representation learning, such as latent action models or goal-conditioned policy learning
Experience integrating tactile, force/torque, or other non-visual sensing into robot learning
Experience with reinforcement learning, long-horizon planning, kinematics, or sim-to-real transfer
Demonstrated research and software engineering experience via an internship, work experience, coding competitions, or widely used contributions in open source repositories (e.g., GitHub)
Proven track record of achieving significant results as demonstrated by grants, fellowships, patents, or publications at leading conferences or journals such as CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, CoRL, RSS, ICRA, or IROS
Ability to commit to 24 consecutive weeks