Cellular Automata Simulation on FPGA for Training Neural Networks with Virtual World Imagery

摘要

We present ongoing work on a tool that consists of two parts: (i) A rawmicro-level abstract world simulator with an interface to (ii) a 3D gameengine, translator of raw abstract simulator data to photorealistic graphics.Part (i) implements a dedicated cellular automata (CA) on reconfigurablehardware (FPGA) and part (ii) interfaces with a deep learning framework fortraining neural networks. The bottleneck of such an architecture usually liesin the fact that transferring the state of the whole CA significantly slowsdown the simulation. We bypass this by sending only a small subset of thegeneral state, which we call a 'locus of visibility', akin to a torchlight in adarkened 3D space, into the simulation. The torchlight concept exists in manygames but these games generally only simulate what is in or near the locus. Ourchosen architecture will enable us to simulate on a micro level outside thelocus. This will give us the advantage of being able to create a larger andmore fine-grained simulation which can be used to train neural networks for usein games.