Scholarlink

<p>Robot skill learning is an important field in robotics technology, involving multiple complex processes such as data collection, experience replay, feature extraction, and strategy learning. Although deep reinforcement learning (DRL) has become a powerful method that combines deep learning with reinforcement learning, the application of DRL to robot skill learning in complex scenarios often faces challenges such as low sampling efficiency and slow learning speed. This paper introduces the robot skill deep reinforcement learning algorithm based on analogical reasoning and Self-Attention mechanism (ARA-DRL). First, it leverages the attention mechanism to focus on essential information in the environment. Second, it employs analogical reasoning to enhance the agent’s decision-making capabilities, ultimately improving robot skill learning performance. We conducted experiments in multiple environments in MuJoCo, and the results showed that ARA-DRL outperforms traditional DRL methods, especially in complex and dynamic environments, significantly enhancing the learning efficiency of robots. In addition, we conducted physical experiments to verify the effectiveness of ARA-DRL in real-world environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ARA-DRL: robot skill deep reinforcement learning based on analogical reasoning and self-attention

  • 期刊论文

  • 发表日期：

    2025年8月31日

查看全文
  • Yunchuan Zhang,
  • Muqun Gao,
  • Wenbai Chen,
  • Peiliang Wu

摘要

Robot skill learning is an important field in robotics technology, involving multiple complex processes such as data collection, experience replay, feature extraction, and strategy learning. Although deep reinforcement learning (DRL) has become a powerful method that combines deep learning with reinforcement learning, the application of DRL to robot skill learning in complex scenarios often faces challenges such as low sampling efficiency and slow learning speed. This paper introduces the robot skill deep reinforcement learning algorithm based on analogical reasoning and Self-Attention mechanism (ARA-DRL). First, it leverages the attention mechanism to focus on essential information in the environment. Second, it employs analogical reasoning to enhance the agent’s decision-making capabilities, ultimately improving robot skill learning performance. We conducted experiments in multiple environments in MuJoCo, and the results showed that ARA-DRL outperforms traditional DRL methods, especially in complex and dynamic environments, significantly enhancing the learning efficiency of robots. In addition, we conducted physical experiments to verify the effectiveness of ARA-DRL in real-world environments.