<ul><li>Equivariant Graph Neural Networks (EGNNs) have emerged as a promising approach in Multi-Agent Reinforcement Learning (MARL), leveraging symmetry guarantees to improve sample efficiency and generalization.</li><li>A novel architecture, Partially Equivariant Graph NeUral Networks (PEnGUiN), is introduced to address challenges in real-world environments that exhibit inherent asymmetries.</li><li>PEnGUiN is capable of learning both fully equivariant (EGNN) and non-equivariant (GNN) representations within a unified framework.</li><li>Extensive experiments validate the efficacy of PEnGUiN, outperforming both EGNNs and standard GNNs in asymmetric environments and improving the robustness and applicability of graph-based MARL algorithms.</li></ul>

PEnGUiN: Partially Equivariant Graph NeUral Networks for Sample Efficient MARL

Discover more