I've spent years watching ML models that work in simulation fail on the factory floor. I've realized we're treating physical success as a single path when it's actually a broad manifold. This article is the first in a series about 'Negative Constraint Learning'—shifting the focus from what to do, to what must never happen.
I've spent years watching ML models that work in simulation fail on the factory floor. I've realized we're treating physical success as a single path when it's actually a broad manifold. This article is the first in a series about 'Negative Constraint Learning'—shifting the focus from what to do, to what must never happen.