Barsalou (1999) presented a simulation-based theory ofgrounded cognition called Perceptual Symbol Systems.According to this theory, a fully functional conceptual systemcan be implemented using only modal representations (akaperceptual symbols) and simulations. While the theory hasgained considerable neuroscientific and experimental support,there is an urgent need for computational accounts that fleshout the theory. The current paper explores one approach forimplementing these computational foundations. We present animplementation of perceptual symbols, simulators, simulation-based perception, and “conscious” multi-modal perceptuallearning based on generative neural networks, called