AI learns to play Gradius (NES)

A.I. learns to play retro games!! 0:00 Intro 0:28 Human expert plays 11:49 AI training 12:13 AI beginner plays 13:19 AI intermediate plays 14:32 AI expert plays 16:49 Summary ================================================================================ Can AI learn to play games? Will AI be better than humans? You've probably heard that DeepMind (creators of AlphaGo) trained AI to play Atari games. Don't you wonder how well AI plays other retro games? Let's find out through experiments, one at a time !! ================================================================================ A.I. training specifications: Reinforcement Learning Algorithm: Proximal Policy Optimization Network: Convolutional neural network 1st layer: 32 8x8 filters, stride=4 2nd layer: 64 4x4 filters, stride=2 3rd layer: 64 3x3 filters, stride=1 Toolkit / environment: OpenAI baselines, OpenAI retro Data Input: Stream of 2D pixel images (game display) and Rewards (game score changes)