A Hierarchical Deep Reinforcement Learning Algorithm with Stochastic Policy Gradient for Robust Robotic Manipulation

Read full text Summarise with VIRAT
Authors: Ha-Eun Yoon, Ji-A Jang
Cited by: 0
References: 11

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The domain of robotic manipulation has witnessed significant advancements through the application of deep reinforcement learning, yet substantial challenges remain regarding sample efficiency, generalization, and robustness against environmental perturbations. This paper introduces a novel Hierarchical Deep Reinforcement Learning framework integrated with a Stochastic Policy Gradient mechanism designed specifically to address the high-dimensional state-action spaces inherent in multi-joint robotic control. By decomposing complex manipulation tasks into temporally extended sub-goals managed by a high-level policy, and executing primitive motor commands via a low-level controller, the proposed architecture effectively mitigates the sparse reward problem. Furthermore, the incorporation of a stochastic policy gradient enables the agent to maintain extensive exploration capabilities while ensuring robust performance in the presence of sensor noise and dynamic friction changes. We demonstrate the efficacy of this approach through rigorous simulation experiments involving complex pick-and-place and stacking tasks. The results indicate that our method significantly outperforms varying state-of the-art baselines in terms of convergence speed and success rates under adversarial conditions.

Fields of Study & Subjects

Reinforcement Learning in Robotics Robot Manipulation and Learning Adaptive Dynamic Programming Control

Publication details

Journal / source
American Journal of Artificial Intelligence and Neural Networks
Volume
7
Issue
1
Pages
13–23
Publication date
28 February 2026
Language
EN