Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2509.22008 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908560757096448 |
|---|---|
| author | Qi, Yajie Wei, Wei Li, Lin Zhang, Lijun Gao, Zhidong Wang, Da Song, Huizhong |
| author_facet | Qi, Yajie Wei, Wei Li, Lin Zhang, Lijun Gao, Zhidong Wang, Da Song, Huizhong |
| contents | Real-world decision-making tasks typically occur in complex and open environments, posing significant challenges to reinforcement learning (RL) agents' exploration efficiency and long-horizon planning capabilities. A promising approach is LLM-enhanced RL, which leverages the rich prior knowledge and strong planning capabilities of LLMs to guide RL agents in efficient exploration. However, existing methods mostly rely on frequent and costly LLM invocations and suffer from limited performance due to the semantic mismatch. In this paper, we introduce a Structured Goal-guided Reinforcement Learning (SGRL) method that integrates a structured goal planner and a goal-conditioned action pruner to guide RL agents toward efficient exploration. Specifically, the structured goal planner utilizes LLMs to generate a reusable, structured function for goal generation, in which goals are prioritized. Furthermore, by utilizing LLMs to determine goals' priority weights, it dynamically generates forward-looking goals to guide the agent's policy toward more promising decision-making trajectories. The goal-conditioned action pruner employs an action masking mechanism that filters out actions misaligned with the current goal, thereby constraining the RL agent to select goal-consistent policies. We evaluate the proposed method on Crafter and Craftax-Classic, and experimental results demonstrate that SGRL achieves superior performance compared to existing state-of-the-art methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_22008 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Goal-Guided Efficient Exploration via Large Language Model in Reinforcement Learning Qi, Yajie Wei, Wei Li, Lin Zhang, Lijun Gao, Zhidong Wang, Da Song, Huizhong Machine Learning Real-world decision-making tasks typically occur in complex and open environments, posing significant challenges to reinforcement learning (RL) agents' exploration efficiency and long-horizon planning capabilities. A promising approach is LLM-enhanced RL, which leverages the rich prior knowledge and strong planning capabilities of LLMs to guide RL agents in efficient exploration. However, existing methods mostly rely on frequent and costly LLM invocations and suffer from limited performance due to the semantic mismatch. In this paper, we introduce a Structured Goal-guided Reinforcement Learning (SGRL) method that integrates a structured goal planner and a goal-conditioned action pruner to guide RL agents toward efficient exploration. Specifically, the structured goal planner utilizes LLMs to generate a reusable, structured function for goal generation, in which goals are prioritized. Furthermore, by utilizing LLMs to determine goals' priority weights, it dynamically generates forward-looking goals to guide the agent's policy toward more promising decision-making trajectories. The goal-conditioned action pruner employs an action masking mechanism that filters out actions misaligned with the current goal, thereby constraining the RL agent to select goal-consistent policies. We evaluate the proposed method on Crafter and Craftax-Classic, and experimental results demonstrate that SGRL achieves superior performance compared to existing state-of-the-art methods. |
| title | Goal-Guided Efficient Exploration via Large Language Model in Reinforcement Learning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2509.22008 |