Add reinforcement learning

[Adrian Roman]

Something like deep q-learning.

Perhaps an example related with physics. Pole balancing is a typical subject.