Ice hockey game using Deep Reinforcement Learning

Modern deep Reinforcement Learning (RL) techniques are quite good at choosing the best possible rules to maximise rewards. By using rich visual information for policy selection, this method combined with Deep Learning methodologies shows promise for difficult tasks. In this work, we have used Double DQN and also three variants of Duelling DQN to implement the ice hockey environment. The Duelling DQN outperformed the Double DQN and gave better results.