I've seen a single paper on the topic of adapting fully recurrent networks to a reinforcement learning setting, but according to google scholar its had no citations and no code has been released implementing the algorithm it describes. Before I go out and roll my own implementation of this algorithm, I just wanted to check that a less obscure algorithm wasn't out there (hopefully with an open-source implementation).
If there isn't a more noteworthy algorithm of this sort, then a secondary question would be why not? Is the some research showing that recurrent architectures don't buy you anything when used for reinforcement learning? Perhaps extending temporal differencing to such an architecture results in intractable computational complexity?