Subgoal Representation

Subgoal representation in hierarchical reinforcement learning (HRL) focuses on effectively decomposing complex tasks into simpler sub-tasks, improving learning efficiency and performance. Current research emphasizes learning robust and temporally coherent subgoal representations, often employing probabilistic methods like Gaussian processes or contrastive learning to capture uncertainty and relationships between subgoals. These advancements aim to address the exploration-exploitation dilemma and improve generalization across tasks, leading to more efficient and adaptable AI agents. The resulting improvements in sample efficiency and performance have significant implications for real-world applications requiring complex sequential decision-making.

Papers

June 24, 2024

Probabilistic Subgoal Representations for Hierarchical Reinforcement learning
Vivienne Huiling Wang, Tinghuai Wang, Wenyan Yang, Joni-Kristian Kämäräinen, Joni Pajarinen
Hierarchical Reinforcement Learning Subgoal Representation

July 22, 2023

Balancing Exploration and Exploitation in Hierarchical Reinforcement Learning via Latent Landmark Graphs
Qingyang Zhang, Yiming Yang, Jingqing Ruan, Xuantang Xiong, Dengpeng Xing, Bo Xu
Reinforcement Learning Hierarchical Reinforcement Learning Real World eXploitation Exploration Performance Subgoal Representation

Subgoal Representation

Papers

Probabilistic Subgoal Representations for Hierarchical Reinforcement learning

Balancing Exploration and Exploitation in Hierarchical Reinforcement Learning via Latent Landmark Graphs