Research at the frontier of physical AI.
Work alongside researchers from the Stanford AI Lab, OpenAI, and DeepMind on original research in robotics and world models, backed by 2M+ hours of data and SOTA research infra.
Apply for Cohort 2- 3–6 Months
- Flexible duration
- $8K
- Monthly stipend
- Up to $1M
- In compute
- Remote / NYC
- Location
The Residency
We're looking for researchers at all stages, from undergrads and graduate students to experienced research engineers, with strong backgrounds in foundation models, robotics, world models, computer vision, simulation, or related fields.
Residents work directly with our research team to pursue original research, run experiments, and publish their findings. You can explore one of our research directions or propose a question of your own.
Our first cohort
An author of NVIDIA's Cosmos 3, a CoRL award winner, and researchers published at NeurIPS and ICRA, with backgrounds spanning CMU, UCL, and Brown.
Selected research questions
- 01Manipulation PhysicsWhat supervision beyond pixels, and what training objectives, let world models learn accurate contact dynamics for dexterous manipulation and stay consistent over long rollouts?
- 02Deployment-Faithful Industrial Manipulation BenchmarksWhat task constraints, scene variability, and failure modes must industrial manipulation benchmarks capture to reliably predict real-world task success and robustness?
- 03Inverse Dynamics & Tactile SupervisionCan inverse dynamics models trained on limited paired video-tactile data recover useful tactile supervision from large-scale egocentric video?
- 04Scaling Human-to-Robot LearningHow does pretraining on human egocentric data change the robot-data requirements of generalist manipulation policies?
Apply for Cohort 2
Share your CV, relevant work, and a short note on the research questions you'd like to pursue.
Apply for Cohort 2