Deep Learning Quick Reference

Results

I've included weights for the Lunar Lander in the chapter Git, and created a script that runs those weights with visualization turned on called dqn_lunar_lander_test.py. It loads the trained model weights and runs for 10 episodes. Most of the time, the agent is able to land the Lunar Lander on it's landing pad with surprising skill and accuracy, as you can see in the following screenshot:

Hopefully, this example demonstrates that while deep Q networks aren't quite rocket science, they can be used to control a rocket.