RL_Transmission_Control

Open source code for paper: "Distributed Transmission Control for Wireless Networks using Multi-Agent Reinforcement Learning " https://arxiv.org/abs/2205.06800

Problem Description

A multi-agent reinfocement learning (MARL) problem where agents decide if and when to transmit in a highly abstracted wireless network setting. A threshold, k, is defined such that only k or fewer agents can transmit successfully on the same time step. Given the level of abstraction, our environment and approach may be applied to other cooperative MARL problems where only a limited number of agents can take the same action on the same step without incurring a reward penalty.

Description of Files

• custom_env.py is the custom environment built in OpenAI Gym
• agent.py is where agents are defined and actions are taken
• argparse_agent.py allows for command line arguments and can be used with driver.sh for automating multiple experiments
• DQN.py contains the code for Deep Q-Network algorithm
• ReplayMemory.py contains the code for the experience replay memory for the DQN agents
• CSMA_agent.py contains the code for the CSMA algorithms used for benchmarking