基于改进SAC算法的六轴机械臂目标抓取
DOI:
作者:
作者单位:

安徽大学人工智能学院

作者简介:

通讯作者:

中图分类号:

TP241??

基金项目:

国家自然科学(62473002,61773177);安徽自然科学(2308085MF204)


Target Grasping of Six-Axis Robotic Arm Based on Improved SAC Algorithm
Author:
Affiliation:

1.School of Artificial Intelligence,Anhui University;2.China

Fund Project:

National Natural Science Foundation of China(62473002,61773177);Anhui Provincial Natural Science Foundation(2308085MF204)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对深度强化学习(DRL)六轴机械臂目标抓取还存在训练效率低、抓取所需时间长的问题,提出一种结合优先经验采样(PER)和n步时序差分预测(n-step TD)的改进软行动者-评论家(SAC)六轴机械臂控制方法。智能体完成动作后计算动作后n步的累积折扣奖励和TD误差放入经验回放池,通过优先经验采样方法从经验回放池中采样数据对各网络进行更新。在仿真环境中设计多种目标点场景机械臂抓取目标点的比较实验,通过与经典SAC算法和SAC+PER算法的对比实验可知,在相同的训练回合数下,改进SAC算法都呈现出更快的收敛速度和更短的抓取时间。

    Abstract:

    To address the issues of low training efficiency and long grasping time in Deep Reinforcement Learning (DRL) for six-axis robotic arm target grasping, an improved Soft Actor-Critic (SAC) control method is proposed. This method combines Prioritized Experience Replay (PER) with n-step Temporal Difference (n-step TD) prediction to enhance performance. After completing an action, the intelligent agent calculates the n-step accumulated discounted reward G_t^((n)) and the TD error δ_t^((n)), which are then stored in the experience replay buffer. Data is sampled from the experience replay buffer using a priority experience sampling method to update various networks. In the simulation environment, multiple target point scenarios were designed for comparative experiments involving a robotic arm grasping target points. Comparisons with the classic Soft Actor-Critic (SAC) algorithm and SAC+PER algorithm revealed that, under the same number of training episodes, the improved SAC algorithm consistently exhibited faster convergence speed and shorter grasping time.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2024-09-29
  • 最后修改日期:2024-10-15
  • 录用日期:2024-10-28
  • 在线发布日期:
  • 出版日期:
文章二维码