Recently, there is an increasing interest in obtaining the relational\nstructures of the environment in the Reinforcement Learning community. However,\nthe resulting "relations" are not the discrete, logical predicates compatible\nto the symbolic reasoning such as classical planning or goal recognition.\nMeanwhile, Latplan (Asai and Fukunaga 2018) bridged the gap between\ndeep-learning perceptual systems and symbolic classical planners. One key\ncomponent of the system is a Neural Network called State AutoEncoder (SAE),\nwhich encodes an image-based input into a propositional representation\ncompatible to classical planning. To get the best of both worlds, we propose\nFirst-Order State AutoEncoder, an unsupervised architecture for grounding the\nfirst-order logic predicates and facts. Each predicate models a relationship\nbetween objects by taking the interpretable arguments and returning a\npropositional value. In the experiment using 8-Puzzle and a photo-realistic\nBlocksworld environment, we show that (1) the resulting predicates capture the\ninterpretable relations (e.g. spatial), (2) they help obtaining the compact,\nabstract model of the environment, and finally, (3) the resulting model is\ncompatible to symbolic classical planning.\n