A web-based version of notes on Double DQNs is available on neatrl, with explanations of the method, its motivation and implementation, plus a live training playground.
The author says the notes cover what Double DQN is, why it differs from classic DQN, and how to implement it. They describe it as an off-policy, model-free, value-based method in the same family as DQNs.
The notes identify overestimation of Q values as a problem in classic DQN. According to the author, the target network’s max operator can create an upward bias and lead to false positives in action selection.