【摘要】 针对机械臂抓取检测中对物体抓取位姿检测精度低的问题，提出一种基于三维点云的机械臂抓取位姿检测方法。首先设计基于注意力机制的端到端抓取位姿检测网络SE-PointNetGPD(简称SEPN-GPD),其次针对pointnet网络在利用共享权重的多层感知器MLP处理3D点云数据时信息冗余的问题，引入通道注意力机制SENet模块，通过自适应地调整各个特征区域权重的方式，提升网络特征提取能力从而提高抓取位姿检测的准确性和可靠性，然后在YCB和BigBIRD公开数据集上进行验证。实验结果表明：SEPN-GPD抓取位姿检测方法的分类精度分别为86.2%和85.14%,且网络具有较好的模型泛化能力和较高的鲁棒性与稳定性，优于当前主流的PointNetGPD和GPD等抓取位姿检测方法。更多还原

【Abstract】 Aiming at the problem of low accuracy of object grasping position detection in robotic arm grasping detection, a 3D point cloud-based robotic arm grasping position detection method was proposed.Firstly, we used to design an end-to-end grasping position detection network SE-PointNetGPD(SEPN-GPD for short) based on the attention mechanism, and secondly, to address the problem of redundancy of information in the pointnet network when utilizing the multilayer perceptron MLP with shared weights to process the 3D point cloud data, we introduced the SENet module of the channel attention mechanism, and adaptively adjust the weights of the individual feature regions to improve the feature extraction capability of the network and thus improve the accuracy of the grasping position detection method.The SENet module was introduced to enhance the feature extraction capability of the network by adaptively adjusting the weights of each feature region to improve the accuracy and reliability of grasping position detection, which was then validated on the YCB and BigBIRD public datasets.The experimental results showed that the classification accuracies of the SEPN-GPD grasping posture detection method were 86.2% and 85.14%,respectively, and the network had a better model generalization ability and higher robustness and stability, which was better than the current mainstream grasping posture detection methods such as PointNetGPD and GPD.更多还原