Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

HiComm: Hierarchical Communication for Multi-agent Reinforcement Learning

About

Cooperative multi-agent reinforcement learning (MARL) often relies on communication to mitigate partial observability, yet most existing protocols treat messages as flat dense vectors detached from the structure of the observations they summarize. This design overlooks an important source of inductive bias in many cooperative environments, where observations naturally follow a hierarchy such as groups and entities. We propose \textsc{HiComm}, a plug-in communication module that grounds messages in the sender's hierarchical observation. \textsc{HiComm} is receiver-driven: the receiver issues a query, and the hierarchy is resolved through a three-stage decoding process that first selects a group, then a sender, and then an entity within that group, returning the corresponding feature slice as the message. This converts communication from unstructured vector transmission into structured information retrieval over the sender's observation hierarchy. We instantiate this mechanism with Straight-Through Gumbel-Softmax for differentiable discrete selection and a lightweight shared projection design that attaches to standard MARL pipelines. Experiments across cooperative MARL tasks with different observation structures and coordination demands show that \textsc{HiComm} matches or outperforms representative learned communication baselines while reducing communication volume by up to $23\times$ per receiver per episode.

Runze Zhao, Dongruo Zhou, Sumit Kumar Jha, Nathaniel D. Bastian, Ankit Shah• 2026

Related benchmarks

TaskDatasetResultRank
Multi-Agent Reinforcement LearningSMAC 5v5 full 3x5 grid v2
Win Rate (Protoss)78
15
Multi-Agent Reinforcement LearningCC4
Reward1.94e+3
10
Multi-Agent Reinforcement LearningGRF cthard 4v3
Reward0.86
10
Multi-Agent Reinforcement LearningSMAC 5v5 v2
Protoss Reward16.7
10
Multi-Agent Reinforcement LearningGRF academy_counterattack_hard (4v3)
Mean Win Rate66
10
Multi-agent coordinationCC4
Reward1.94e+3
9
Multi-Agent Reinforcement LearningSMAC protoss_5v5 v2 (test)
Reward17.4
5
Multi-Agent Reinforcement LearningSMAC terran_5v5 v2 (test)
Reward17.6
5
Multi-Agent Reinforcement LearningSMAC zerg_5v5 v2 (test)
Reward13.9
5
Showing 9 of 9 rows

Other info

Follow for update