Research track · mathground.ai
MathGround
Math-Ground AI replaces learned, opaque control with closed-form primitives grounded in physics and math. Every controller is traced to a Lyapunov certificate, every call carries a picojoule energy receipt, and the whole runtime fits in a WebGPU browser under 400 KB.
Swap-2C · provable control · an OpenIE project of the Charlot Lab
State of the art
Where this sits.
Embodied control today is dominated by large learned policies. They are capable, but unprovable and power-hungry. The frontier is provable, energy-bounded control. Math-Ground grounds behavior in closed-form math, so stability is certified rather than hoped for, and the power cost is known before deployment.
More research tracks
The Charlot Lab in the open.
Every track is a live, working site.