This is just a minor tweak from this repo.
The algorithm is DDPG. See this post as a great intro to the algorithm.
The main difference is that this repo uses roboschool instead of plain OpenAi Gym. Roboschool features the awesome Bullet 3D physics engine.
The repo defaults to the classic cartpole problem, and with Roboschool the actions for this env are continuous (in this case representing motor voltage), which really lets the DDPG algorithm stretch its legs.
Dependencies:
Tested with Python 2.7 and Python 3.6