hijkzzz/reinforcement-learning-wechat-jump

Reinforcement Learning for WeChat Jump

Deep Reinforcement Learning Wechat Jump

End-to-end training Wechat-Jump AI using DDPG algorithm

Environment

• PyTorch
• PyOpenCV
• Android Device(1920 * 1080)
• ADB Tools

Detail

• Using screenshot as neural network input
• The actor uses tanh as the activation function
• Critic uses a linear layer to output Q values
• Noise is a normal distribution with a std=0.2
• Get game state with template matching of OpenCV

Train

python train.py

Infer

python infer.py

Reference

• Continuous control with deep reinforcement learning
• https://github.com/ikostrikov/pytorch-ddpg-naf
• https://github.com/floodsung/wechat_jump_end_to_end