About
A research lab for Physical AI.
We work on the core problems of Physical AI — perception, control, world modeling, and the safety challenges of putting learned policies on hardware.
Mission
Make progress on the open problems of Physical AI — closing the gap between systems that model the world and physical systems that reliably act within it.
We focus where the bottleneck is genuine: contact-rich manipulation, robust locomotion, predictive world models, and transfer from simulation to systems that have to keep working when nothing in the test set looks quite like training.
Vision
Physical systems that are competent, safe, and useful in the real world — not as demos, but as systems that hold up outside the lab.
Getting there requires research that takes evaluation, generalization, and safety as seriously as it takes algorithms.
Principles
How we work
- Long horizons
- Hard problems take years. We pick them deliberately and stay with them, rather than chasing the quarter's benchmark.
- Reality as the test set
- We measure progress against the world the systems will eventually be deployed in, not against simulators alone.
- Honest evaluation
- Strong baselines, careful ablations, and negative results that get written up. Confidence calibrated to evidence.
- Safety as research
- Embodied systems act in the world. Safety is a first-class research problem, not a checklist appended at the end.
Path
From applied AI studio to a focused lab
- Jun 2025
Founded
Started as an applied AI studio working across machine learning and automation.
- Apr 2026
Refocused on Physical AI
Narrowed scope to embodied intelligence: robotics, world models, and sim-to-real.
Curious about the work? We're always glad to hear from people thinking about the same problems.