Text this: MetaSeeker: sketching an open invisible space with self-play reinforcement learning