Human attention guided multiagent hierarchical reinforcement learning for heterogeneous agents

Dingbang Liu, Fenghui Ren, Jun Yan, Guoxin Su, Shohei Kato, Wen Gu, Minjie Zhang

Abstract: Highlights•We propose a novel approach to solve heterogeneous multiagent problems•Our approach tolerates suboptimal human guidance and reduces reliance on expertises.•Human attention guides parallel agent learning, enabling adaptation and dynamic knowledge requirements.•The method is algorithm-agnostic and can flexibly integrate with a variety of MARL algorithms.•The training is end-to-end, granting flexibility for customizing human attention.