This repository is meant to offer a way to reproduce the experiments of some of the papers Unity published using ML-Agents.
Multi-Agent POsthumous Credit Assignment (MA-POCA) is a multi-agent learning algorithm that can gracefully deal with spawning and despawning agents without the use of absorbing states. See here for more information.