Grasping in cluttered and tight scenes is a necessary skill for intelligent robotics to achieve more general application. Such universal robotics can use their perception abilities to visually identify grasps from a stack of objects. However, most existing grasping detection methods based on deep learning just focus on estimating grasping pose with single-layer features. In this paper, we present a novel grasp detection algorithm termed as multi-object grasping detection network, which can utilize hierarchical features to learn object detector and grasping pose estimator simultaneously. The network is mainly composed of two branches: 1) Object detection branch which is based on the single shot multibox detection approach to discriminate object categories and locate object positions by bounding boxes; 2) Grasping pose estimation branch where hierarchical features are fused together to predict grasping position and orientation. To improve grasping detection performance, attention mechanism is employed in hierarchical feature fusion. For evaluating the proposed model, we build a multi-object grasping dataset where every image contains numerous different graspable objects. The extensive experiments demonstrate that the multi-object grasping detection method achieves the state-of-the-art performance on both object detection and grasping pose estimation.

Domain adaptation aims to generalize the classification model from a source domain to a different but related target domain. Recent studies have revealed the benefit of deep convolutional features trained on a large dataset (e.g. ImageNet) in alleviating domain discrepancy. However, literatures show that the transferability of features decreases as (i) the difference between the source and target domains increases, or (ii) the layers are toward the top layers. Therefore, even with deep features, domain adaptation remains necessary. In this paper, we propose a novel unsupervised domain adaptation (UDA) model for deep neural networks, which is learned with the labeled source samples and the unlabeled target ones simultaneously. For target samples without labels, pseudo labels are assigned to them according to their maximum classification scores during training of the UDA model. However, due to the domain discrepancy, label noise generally is inevitable, which degrades the performance of the domain adaptation model. Thus, to effectively utilize the target samples, three specific robust deep softmax regression (RDSR) functions are performed for them with high, medium and low classification confidence respectively. Extensive experiments show that our method yields the state-of-the-art results, demonstrating the effectiveness of the robust deep softmax regression classifier in UDA.

No abstract