Reinforcement learning isn't a NN architecture

[etano]

RL is a method of training. The actual NN used are often pretty standard feed-forward NN.

One counter-example could be value iteration networks: https://arxiv.org/abs/1602.02867

But I would classify these separately.