Fully Convolutional Grasp Detection Network with Oriented Anchor Box

In this paper, we present a real-time approach to predict multiple grasping poses for a parallel-plate robotic gripper using RGB images. A model with oriented anchor box mechanism is proposed and a new matching strategy is used during the training process. An end-to-end fully convolutional neural network is employed in our work. The network consists of two parts: the feature extractor and multi-grasp predictor. The feature extractor is a deep convolutional neural network. The multi-grasp predictor regresses grasp rectangles from predefined oriented rectangles, called oriented anchor boxes, and classifies the rectangles into graspable and ungraspable. On the standard Cornell Grasp Dataset, our model achieves an accuracy of 97. 74% and 96. 61% on image-wise split and object-wise split respectively, and outperforms the latest state-of-the-art approach by 1. 74% on image-wise split and 0. 51% on object-wise split.

Authors

• Xinwen Zhou
• Xuguang Lan
• Hanbo Zhang
• Zhiqiang Tian
• Yang Zhang
• Nanning Zheng 0001

Keywords

• Feature extraction
• Robots
• Computational modeling
• Grippers
• Solid modeling
• Computer architecture
• Predictive models
• Convolutional Network
• Anchor Boxes
• Neural Network
• Convolutional Neural Network
• Deep Neural Network
• Deep Convolutional Neural Network
• RGB Images
• Matching Strategy
• Convolutional Layers
• Aspect Ratio
• Input Image
• Feature Maps
• Object Detection
• Data Augmentation
• Detection Results
• Grid Cells
• Intersection Over Union
• Rotation Angle
• Bounding Box
• February 2018
• Accurate Way
• One-stage Detectors
• Tactile Sensor
• Regression Loss
• RGB Data
• Regression Layer
• Predicted Bounding Box
• Deeper Network
• Rapid Way
• Classification Loss