NeurIPS 2020 Flatland challenge solution

Description

This repository contains the multi-agent reinforcement learning framework solution used for our solution to the NeurIPS 2020 Flatland Challenge.

Key Files

• ACNET4.py - Defines network architecture.
• NewAgentInitObs.py - Defines the decentralized observation of each agent.
• DRLMAPF_A3C_RNN.py - The main file which contains the entire training code, worker classes and training parameters.

Authors

Mehul Damani

Zhiyao Luo

Guillaume Sartoretti