Project

Neural Physics Engine

We've developed an object-based neural network architecture for learning predictive models of intuitive physics that extrapolates to variable object count and variable scene configurations with only spatially and temporally local computation.