Heterogeneous Policy Networks for Composite Robot Team Communication and Coordination

2026-06-18

AI Summary

Existing approaches struggle to effectively model differentiated communication and coordination among heterogeneous agents, limiting overall team performance. This work proposes Heterogeneous Policy Network (HetNet), the first framework to enable end-to-end joint learning of collaborative policies and binary communication for large-scale heterogeneous robotic teams. Built upon a heterogeneous graph attention mechanism and integrated with multi-agent reinforcement learning, HetNet achieves substantial communication bandwidth reduction—up to 200-fold compression—while outperforming state-of-the-art baselines by 5.84% to 707.65% across multiple tasks, thereby simultaneously ensuring highly efficient communication and superior collaborative performance.

Abstract

High-performing human-human teams learn intelligent and efficient communication and coordination strategies to maximize their joint utility. These teams implicitly understand the different roles of heterogeneous team members and adapt their communication protocols accordingly. Multi-Agent Reinforcement Learning (MARL) has attempted to develop computational methods for synthesizing such joint coordination-communication strategies, but emulating heterogeneous communication patterns across agents with different state, action, and observation spaces has remained a challenge. Without properly modeling agent heterogeneity, as in prior MARL work that leverages homogeneous graph networks, communication becomes less helpful and can even deteriorate the team's performance. In the past, we proposed Heterogeneous Policy Networks (HetNet) to learn efficient and diverse communication models for coordinating cooperative heterogeneous teams. In this extended work, we extend Heterogeneous Policy Networks (HetNet) to support scaling heterogeneous robot teams. Building on heterogeneous graph-attention networks, we show that HetNet not only facilitates learning heterogeneous collaborative policies but also enables end-to-end training for learning highly efficient binarized messaging. Our empirical evaluation shows that HetNet sets a new state of the art in learning coordination and communication strategies for heterogeneous multi-agent teams by achieving an 5.84% to 707.65% performance improvement over the next-best baseline across multiple domains while simultaneously achieving a 200x reduction in the required communication bandwidth.

Problem

• heterogeneous multi-agent systems
• communication and coordination
• multi-agent reinforcement learning
• agent heterogeneity
• cooperative teams

Innovation

• Heterogeneous Policy Networks
• Multi-Agent Reinforcement Learning
• Graph Attention Networks
• Binarized Messaging
• Team Coordination