Hierarchical Reinforcement Learning with Intrinsic Motivation for Complex Robotic Manipulation Tasks

This paper presents a novel approach to hierarchical reinforcement learning (HRL) for robotic manipulation tasks, leveraging intrinsic motivation to improve exploration and learning efficiency in complex, multi-stage scenarios. We propose a hierarchical architecture where high-level policies learn to decompose tasks into subgoals, while low-level policies execute primitive actions to achieve these subgoals. Intrinsic motivation, in the form of curiosity-driven exploration, is incorporated at both levels to encourage the discovery of novel and potentially useful behaviors. We evaluate our approach on a simulated robotic arm manipulation task, demonstrating significant improvements in learning speed and task success rate compared to flat reinforcement learning and traditional HRL methods.