muupan / dqn-in-the-caffe

An implementation of Deep Q-Network using Caffe

DQN-in-the-Caffe

DQN-in-the-Caffe is an implementation of Deep Q-Network using Caffe.

Requirements

• Caffe (with two pull requests and minor fixes on them)
• BVLC/caffe#1228 for step execution of solving
• BVLC/caffe#1122 for AdaDelta solver
• At the moment my forked repository's dqn branch https://github.com/muupan/caffe/tree/dqn can be used with DQN-in-the-Caffe
• Arcade Learning Environment
• etc.

Algorithm Details

See http://www.cs.toronto.edu/~vmnih/docs/dqn.pdf for the details of DQN.

Demo

https://www.youtube.com/watch?v=p88R2_3yWPA