Communication is crucial in multi-agent reinforcement learning when agents are not able to observe the full state of the environment. The most common approach to allow learned communication between agents is the use of a differentiable communication channel that allows gradients to flow between agents as a form of feedback. However, this is challenging when we want to use discrete messages to reduce the message size since gradients cannot flow through a discrete communication channel. Previous work proposed methods to deal with this problem. However, these methods are tested in different communication learning architectures and environments, making it hard to compare them. In this paper, we compare several state-of-the-art discretization methods as well as two methods that have not been used for communication learning before. We do this comparison in the context of communication learning using gradients from other agents and perform tests on several environments. Our results show that none of the methods is best in all environments. The best choice in discretization method greatly depends on the environment. However, the discretize regularize unit (DRU), straight through DRU and the straight through gumbel softmax show the most consistent results across all the tested environments. Therefore, these methods prove to be the best choice for general use while the straight through estimator and the gumbel softmax may provide better results in specific environments but fail completely in others.
View on arXiv