Distributed Assembly

Reinforcement learning code to train multiple agents to collaboratively build user-specified structures in a 3D discrete gridworld.

File list

• A3C_RNN.ipynb: Multi-agent training code.
• minecraft.py: Multi-agent "minecraft" gym environment, in which agents learn collective construction.

Requirements

• OpenAI Gym 0.9.4
• Tensorflow 1.3.1
• Numpy 1.13.3
• matplotlib
• moviepy (for GIFs creation)
• ffmpeg (for GIFs creation)