gumpfly/pytorch-mace

A pytorch implementation of Structured Exploration via Deep Hierarchical Coordination

An implementation of MACE

1. Introduction

This is a pytorch implementation of Structured Exploration via Deep Hierarchical Coordination.

The experimental environment is a modified version of Waterworld named MAWaterWorld_modified_mixed based on MADRL.

2. Environment

The main features (different from MADRL and that of MADDPG) of the modified Waterworld environment are:

• evaders and poisons now bounce at the wall obeying physical rules
• sizes of the evaders, pursuers and poisons are now the same so that random actions will lead to average rewards around 0.
• need exactly n_coop agents to catch food.
• discrete actions: up, down, left, right.

3. Dependency

• pytorch
• visdom
• python==3.6.1 (recommend using the anaconda/miniconda)

4. Install

• Install MADRL.
• Replace the files in madrl_environments/pursuit directory with the ones in this repo.
• python main.py will run the training.

5. Results

Two agents, cooperation = 2, evader = 3, poison = 3

The two agents need to cooperate to achieve the food for reward 10.