Multiobjective evolution for deep learning and its robotic applications

Abstract

In numerous industrial applications where robot object recognition and grasping are the primary concern as the most effective and reliable object sorting policy. Deep Learning approaches have produced promising results in object recognition and robot gasping, its performance does not have any influence from handcrafted features. In this paper, we propose a multiobjective deep belief neural network (DBNN) method. It employs a multiobjective evolutionary algorithm integrated with DBNN [10] training technique subject to accuracy and network time as two conflicting objectives. We evaluate the proposed method on the real-time object recognition and robot grasping tasks. Experimental results demonstrate that the proposed method outperforms on the assign tasks.