When to Communicate: Belief Distributions and KL Divergence for Principled Gating in Multi-Agent RL
Not provided in the abstract
Abstract
This paper proposes a method for agents in multi-agent reinforcement learning to communicate based on the KL divergence of their belief distributions, improving coordination and reducing unnecessary communication.
Reality Card
The proposed KL divergence-based gating method improves communication efficiency and coordination in multi-agent settings, achieving competitive performance against existing methods.
On the Predator-Prey 20x20 environment, the method achieved a success rate of 42% with an average of 73.84 steps, compared to IC3Net's 31% success rate.
The paper does not provide detailed information on the implementation or the specific configurations used for experiments, which may hinder reproducibility.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.