
This paper presents a novel method for detecting landmarks to decompose complex tasks into subtasks, leveraging first-order logic state representation and large language models (LLMs) to create interpretable rule-based policies through an inductive logic programming (ILP)-based reinforcement learning agent.


coremsg

landmark-based-task-decomposition-and-llm-augmented-symbolic-reinforcement-learning


Landmark-Based Task Decomposition and LLM-Augmented Symbolic Reinforcement Learning
