Virtual Training for a Real Application: Accurate Object-Robot Relative Localization Without Calibration

• Published in: International journal of computer vision Vol. 126; no. 9; pp. 1045 - 1060
• Main Authors: Loing, Vianney, Marlet, Renaud, Aubry, Mathieu
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.09.2018; Springer; Springer Nature B.V; Springer Verlag
• Subjects: Artificial Intelligence; Artificial neural networks; Computer Imaging; Computer Science; Computer Vision and Pattern Recognition; Construction sites; Data acquisition; Image Processing and Computer Vision; Localization; Neural networks; Pattern Recognition; Pattern Recognition and Graphics; Robots; Three dimensional models; Training; Vision; Convolutional Neural Networks; Virtual training; Pose estimation; Robotics; Relative localization; Synthetic data; Relative localisation
• ISSN: 0920-5691; 1573-1405
• DOI: 10.1007/s11263-018-1102-6

Localizing an object accurately with respect to a robot is a key step for autonomous robotic manipulation. In this work, we propose to tackle this task knowing only 3D models of the robot and object in the particular case where the scene is viewed from uncalibrated cameras—a situation which would be typical in an uncontrolled environment, e.g., on a construction site. We demonstrate that this localization can be performed very accurately, with millimetric errors, without using a single real image for training, a strong advantage since acquiring representative training data is a long and expensive process. Our approach relies on a classification Convolutional Neural Network trained using hundreds of thousands of synthetically rendered scenes with randomized parameters. To evaluate our approach quantitatively and make it comparable to alternative approaches, we build a new rich dataset of real robot images with accurately localized blocks.