About the role#
This internship focuses on advancing foundation models for robotics, specifically through mid-training. You will join the Reality Labs Research team to develop scalable methods for world action models (WAMs) using large-scale egocentric video and heterogeneous robot data. You will conduct fundamental and applied research to build the next generation of embodied AI, working alongside research scientists and engineers to develop algorithms and evaluate model capabilities.
What you'll do#
- Develop scalable mid-training methods for WAMs using large-scale egocentric video and heterogeneous robot data.
- Investigate representation and objective design for learning from human egocentric video.
- Address challenges in training on web-scale, weakly-labeled data, including embodiment and viewpoint gaps, data curation, and domain mixture weighting.
- Build efficient training and data pipelines while studying how downstream policy performance scales with dataset size, model capacity, and compute.
- Design experiments and benchmarks to isolate the effect of mid-training on downstream post-training and generalization across tasks.
- Collaborate with the team to evaluate mid-trained models in simulation and on real robots.
- Communicate research findings and publish results at leading machine learning, computer vision, and robotics venues.
What you'll need#
- Currently pursuing a Ph.D. in Computer Science, Machine Learning, Artificial Intelligence, Robotics, or a related STEM field.
- Research experience in training video models or robotics foundation models.
- 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 and maintain work authorization in the country of employment.
- Intent to return to your degree program after the internship.
- Experience training VLAs, WAMs, or other foundation models for embodied decision-making.
- Experience with diffusion or flow-based policies, action-sequence prediction, 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.
- A proven track record of significant results, such as grants, fellowships, patents, or accepted papers in top conferences or journals.
Location & details#
- Location: Menlo Park, California.
- Modality: On-site.
- Duration: 12 to 24 weeks.
- Schedule: Full-time.
- Compensation: $7,650 to $12,667 per month.
About Meta
Meta develops software and social media applications. It operates platforms such as Facebook, Messenger, Instagram, and WhatsApp. The company maintains headquarters in Menlo Park, California. Current work focuses on augmented and virtual reality alongside artificial intelligence and machine learning.
How to get in at Meta
Landing an internship at Meta requires a proactive approach to the application process. Intern Insider sends an instant alert the moment a role matching your target is published, allowing you to apply among the first. Early applicants often have their materials reviewed before the applicant pool grows significantly. You can use Intern Insider to surface the recruiters behind specific roles at Meta. Reaching out directly to these individuals to ask about the position or a referral often improves your response rates compared to applying into a queue.
