Research Scientist Intern, Robotics Foundation Models, Agentic RL (PhD)

Meta•Menlo Park, CA

1d•$7,650 - $12,667

About The Position

Meta Reality Labs Research is seeking a Research Scientist Intern to develop scalable offline reinforcement learning methods for post-training world action models (WAMs) and vision-language-action models (VLAs) for robotic manipulation. The intern will conduct fundamental and applied research advancing robotics foundation models, with a focus on post-training. They will collaborate with research scientists and engineers to develop algorithms, build and evaluate model capabilities, and contribute to embodied AI. Internships last 12 to 24 weeks, with various start dates throughout the year.

Requirements

Currently has or is in the process of obtaining a Ph.D. in Computer Science, Machine Learning, Artificial Intelligence, Robotics, or a related STEM field.
Research experience in reinforcement learning, including policy optimization, value learning, or offline/off-policy learning.
Experience implementing and training deep learning models using Python and PyTorch or a comparable framework.
Experience designing controlled experiments, analyzing results, and debugging machine learning algorithms.
Interpersonal experience with cross-group and cross-culture collaboration.
Must obtain work authorization in the country of employment by the time of hire and maintain ongoing work authorization during employment.

Nice To Haves

Intent to return to a degree program after completion of the internship or co-op.
Experience developing offline RL methods, particularly for continuous control, large datasets, or learning from suboptimal experience.
Experience training or post-training VLAs, WAMs, or other foundation models for embodied decision-making.
Experience with diffusion or flow-based policies, action-sequence prediction, model-based RL, or action-conditioned world models.
Hands-on experience with distributed training across multiple GPUs or nodes, including improving training throughput, memory efficiency, and stability.
Experience evaluating learned policies on physical robots or robotics simulation benchmarks.
Track record of significant results demonstrated by grants, fellowships, patents, or accepted papers in leading conferences or journals such as NeurIPS, ICML, ICLR, CoRL, RSS, ICRA, CVPR, ICCV, ECCV, or ACL.

Responsibilities

Develop agentic RL algorithms for post-training WAMs and VLAs using large, heterogeneous robot datasets.
Investigate agentic RSI, reward learning, value estimation, and policy optimization for generative action models, including approaches using learned world models to support policy improvement.
Address challenges in learning from mixed-quality experience, including distribution shift, sparse rewards, and long-horizon credit assignment.
Build efficient training and data pipelines, and study how policy performance scales with dataset size, model capacity, and compute.
Design experiments and benchmarks to evaluate policy improvement and generalization across tasks and environments.
Collaborate with research scientists and engineers to evaluate learned policies in simulation and on real robots.
Communicate research findings and publish results at leading machine learning and robotics venues.