Visual Data Simulation for Deep Learning in Robot Manipulation Tasks

This paper introduces the usage of simulated images fortraining convolutional neural networks for object recognition and local-ization in the task of random bin picking. For machine learning appli-cations, a limited amount of real world image data that can be cap-tured and labeled for training and testing purposes is a big issue. Inthis paper, we focus on the use of realistic simulation of image data fortraining convolutional neural networks to be able to estimate the poseof an object. We can systematically generate varying camera viewpointdatasets with a various pose of an object and lighting conditions. Aftersuccessful training and testing the neural network, we compare the per-formance of network trained on simulated images and images from a realcamera capturing the physical object. The usage of the simulated datacan speed up the complex and time-consuming task of gathering trainingdata as well as increase robustness of object recognition by generating abigger amount of data