Improved GQ-CNN: Deep Learning Model for Planning Robust Grasps

• DOI: 10.48550/arxiv.1802.05992
• Publication Date: 2018-01-01
• Authors: Maciej Jaskowski; Jakub Świątkowski; Michał Zając; Maciej Klimek; Jarek Potiuk; Piotr Rybicki; Piotr Polatowski; Przemyslaw Walczyk; Kacper Nowicki; Marek Cygan
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Computer Science - Robotics; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Statistics - Machine Learning; 0202 electrical engineering, electronic engineering, information engineering; Machine Learning (stat.ML); 02 engineering and technology; Robotics (cs.RO); Machine Learning (cs.LG)

Recent developments in the field of robot grasping have shown great improvements grasp success rates when dealing with unknown objects. In this work we improve on one most promising approaches, Grasp Quality Convolutional Neural Network (GQ-CNN) trained DexNet 2.0 dataset. We propose a new architecture for GQ-CNN and describe practical that increase model validation accuracy from 92.2% to 95.8% 85.9% 88.0% respectively image-wise object-wise training splits.