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

Authors: Alireza Kheirandish, Duo Xu, Faramarz Fekri
Summary: This paper presents an algorithm that uses Large Language Models (LLMs) to decompose complex reinforcement learning tasks into simpler subtasks. The method employs positive and negative trajectories to identify these subtasks and generate first-order logic rule templates, demonstrating effectiveness in task resolution.