D3HRL: A Distributed Hierarchical Reinforcement Learning Approach Based on Causal Discovery and Spurious Correlation Detection
Chenran Zhao, Dianxi Shi, Mengzhu Wang, Jianqiang Xia, Huanhuan Yang, Songchang Jin, Shaowu Yang, Chunping Qiu
TL;DR
D3HRL tackles two core issues in long-horizon tasks—delay effects and spurious correlations—by modeling delayed effects as cross-time causal relations and applying distributed causal discovery across multiple time spans up to $\tau_{max}$. It then uses conditional independence testing to prune spurious links and constructs a hierarchical policy network aligned with the learned causal chain. The approach yields higher causal-graph accuracy and improved learning efficiency, outperforming baselines like CDHRL in identifying true causal relationships and guiding sub-goal training. This yields more reliable decision-making in complex environments and demonstrates scalability across time-span configurations, with potential for extension to additional modalities and more diverse tasks.
Abstract
Current Hierarchical Reinforcement Learning (HRL) algorithms excel in long-horizon sequential decision-making tasks but still face two challenges: delay effects and spurious correlations. To address them, we propose a causal HRL approach called D3HRL. First, D3HRL models delayed effects as causal relationships across different time spans and employs distributed causal discovery to learn these relationships. Second, it employs conditional independence testing to eliminate spurious correlations. Finally, D3HRL constructs and trains hierarchical policies based on the identified true causal relationships. These three steps are iteratively executed, gradually exploring the complete causal chain of the task. Experiments conducted in 2D-MineCraft and MiniGrid show that D3HRL demonstrates superior sensitivity to delay effects and accurately identifies causal relationships, leading to reliable decision-making in complex environments.