Research on Robot Grasping Control Technology in Dense Stacking Environments Based on Deep Learning

To obtain the optimal grasp sequence and grasp pose for robotic grasping of target objects in dense stacking environments, a stacked object grasping algorithm based on deep learning is proposed. This algorithm consists of two parts: grasp sequence reasoning and grasp pose detection. In grasp sequence reasoning, a segmentation network based on attention and boundary refinement is designed to enhance feature extraction capability and segmentation accuracy. Additionally, a grasp screening method is proposed to select the uppermost unoccluded object as the target to be grasped. In grasp pose detection, a grasp pose detection network integrating multi-scale dense residual modules is designed to preserve multi-scale features of objects while enabling detection of objects of various scales. Experimental results demonstrate that the proposed algorithm achieves high detection accuracy, enabling the detection of target objects in dense stacking environments while accurately detecting their poses.