Aiming at the stability and accuracy of visual robot grasping target,a reinforcement learning strategy focusing on exploration method was proposed.The deterministic strategy gradient algorithm with deep concern was adopted.The information of pre-exploration area was selected by using the region of interest recommendation network,and the information was calculated by the adaptive exploration method to adjust the strategy with the change of the target.According to the distance between the end effector and the center of the pre-exploration area,a hierarchical reward function was defined to reduce the miscellaneous information brought by the sparse reward matrix.The training was carried out in Bullet3 environment.The experiments show that the proposed strategy can overcome the problems of poor stability and low convergence efficiency that may occur in the training process,can produce robust control against noise interference,and has a high success rate of grasping.
参考文献
相似文献
引证文献
引用本文
明鑫,卢丹萍,陈中.一种视觉机器人抓取控制策略算法研究[J].机床与液压,2023,51(11):65-71. MING Xin, LU Danping, CHEN Zhong. Research on a Visual Robot Grasping Control Strategy Algorithm[J]. Machine Tool & Hydraulics,2023,51(11):65-71