A. Josin Hippolitus and R. Senthilnathan
Dual-arm grasping, robot control, CNN, deep learning
In a progressing and complex world, robot grasping and manipulation in a cluttered environment is a challenging activity. Especially when the object to be manipulated is of unknown geometry and located in a cluttered environment. In this work, a novel quick-pick CNN(QP-CNN) algorithm is implemented to identify the best grasp for a 3D object in real time. The potential impact of this research can range from improving the speed, efficiency, and accuracy in object manipulation of unknown objects in a cluttered environment assuming a model-free context. RGB-D data from the real world about the object to be manipulated is acquired and mapped to the objects. This information acts as the input for the pre-trained networks to provide input to a 7-DOF ABB YuMi dual-arm robot. The objectwise grasping accuracy of QP-CNN is 98.1% and grasp time is 2 s with 100% reliability.
Important Links:
Go Back
