An open lab building environments for the slow, irreversible parts of the physical world — where an episode costs a year and can’t be reset. Starting with growing systems.

Reinforcement learning has advanced fastest where experience is cheap and reversible. Real growing systems are the opposite — governed by processes that unfold over seasons and can’t be rolled back.

We build fast, faithful environments for these systems — simulators that reflect how the real world behaves, so decisions can be studied at the speed of computation rather than the speed of seasons. Our first domain is agricultural growing systems, and the method is meant to reach beyond them.