Learning systems for a world in motion.
§ 01
Research direction
Intelligence begins with a model of what happens next.
Machines acting in the physical world need representations that connect observation, action, and consequence.
- World models
- Predictive representations of dynamic environments — built to reason about change, uncertainty, and the consequences of action.
- Scalable learning
- Data, training, and evaluation treated as one loop — so capability can improve as experience grows.
- Real-world generalization
- Methods for transferring learned behavior across tasks, environments, and forms of embodiment.
Current program withheld
We are deliberately quiet about work in progress. If you need to know more, write to us.
Where prediction becomes action.
Fig. 01 — Each dot follows its own path. One of them walks.