Robotic Manipulation: Unlocking the Potential of Data Augmentation for Industrial Robots and Workers’ Safety and Health

Industrial Robots, Workers’ Safety, and Health

Saul Morales RodriguezAuthor

In robotic manipulation, there are a number of learning problems that do not have large datasets. The collection of these datasets can be time-consuming and costly, so learning from small datasets remains an open problem. Data augmentation is a common solution to the lack of data in computer vision. The process of data augmentation involves modifying existing examples to create additional training examples. The methods of computer vision are not easily adaptable to manipulation due to the different types of data and tasks. We propose a method of data augmentation for robotic manipulation. We believe that augmentations must be valid, meaningful, and diverse.