Real-world Multi-object, Multi-grasp Detection
This addresses robotic manipulation for household tasks, but it is incremental as it builds on existing grasp detection methods with a novel classification approach.
The paper tackles the problem of robotic grasp detection for multiple objects in real-world scenarios, achieving 96.0% accuracy on the Cornell dataset and 88.0% grasping success rates on household objects.
A deep learning architecture is proposed to predict graspable locations for robotic manipulation. It considers situations where no, one, or multiple object(s) are seen. By defining the learning problem to be classification with null hypothesis competition instead of regression, the deep neural network with RGB-D image input predicts multiple grasp candidates for a single object or multiple objects, in a single shot. The method outperforms state-of-the-art approaches on the Cornell dataset with 96.0% and 96.1% accuracy on image-wise and object- wise splits, respectively. Evaluation on a multi-object dataset illustrates the generalization capability of the architecture. Grasping experiments achieve 96.0% grasp localization and 88.0% grasping success rates on a test set of household objects. The real-time process takes less than .25 s from image to plan.