Implementation of the paper "Decentralized Counterfactual Value with Threat Detection in Multi-Agent Mixed Cooperative and Competitive Environments"
This is the code for the paper "Decentralized Counterfactual Value with Threat Detection in Multi-Agent Mixed Cooperative and Competitive Environments".
- Grid Examples: The environment contains two basic grid environment (2-room and 4-room environments), which are implemented in the
GRID/ENVfile. - Classical Scenarios: Job Scheduling, Matthew Effect and Manufacturing Plant. All scenarios have limited resources, thus, the agents encounter the mixed cooperative and competitive relationship with others under the general-sum rewards.
- SSD Environments: The environment is a 2D grid game with the partially observable state with the picture of
$15 \times 15 \times 3$ . The action space is a discrete space that includes$7$ basic motions: move up, move down, move left, move right, stay, rotate clockwise and rotate counterclockwise. Each agent intends to collect more apples in the map and each apple responds with a$+1$ reward. - MAgent: MAgent is a research platform for many-agent reinforcement learning. Unlike previous research platforms that focus on reinforcement learning research with a single agent or only few agents, MAgent aims at supporting reinforcement learning research that scales up from hundreds to millions of agents.
Please follow the instruction of 'README.md' file in different environments to install Python requirements.