We present an accurate, real-time approach to robotic grasp detection basedon convolutional neural networks. Our network performs single-stage regressionto graspable bounding boxes without using standard sliding window or regionproposal techniques. The model outperforms state-of-the-art approaches by 14percentage points and runs at 13 frames per second on a GPU. Our network cansimultaneously perform classification so that in a single step it recognizesthe object and finds a good grasp rectangle. A modification to this modelpredicts multiple grasps per object by using a locally constrained predictionmechanism. The locally constrained model performs significantly better,especially on objects that can be grasped in a variety of ways.
展开▼
