kalman.industries

RL environments for physical AI, beyond robotics.

Agents will soon operate machines and industrial processes across every physical industry. Today they still fail at that work too often to be trusted. We are building a lab that develops and operates the reinforcement learning environments AI labs use to post-train their models for real-world interaction with machines.

We start with embedded software: environments where agents write, flash and debug firmware on real microcontrollers, with rewards graded by instruments on the bench rather than by simulation alone. Programmable logic controllers are next.

An open, hardware-graded evaluation across the major MCU platforms is targeted for early 2027.