Recently, there is an increasing interest in obtaining the relational
structures of the environment in the reinforcement learning community. However,
the resulting "relations" are not the discrete, logical predicates compatible
to the symbolic reasoning such as classical planning or go