We introduce the Omnipush dataset, which includes RGB-D images as raw inputs
as well as a large variety of objects to test generalization.
In particular we provide 250 pushes for each of 250 objects, all recorded with RGB-D and a high precision tracking system.
The objects are constructed so as to systematically explore key factors that affect pushing
--the shape of the object and its mass distribution-- which have not been broadly explored in previous datasets.
We also propose baselines for three key learning-based problems: dynamic modeling, state estimation and RGB video prediction.
Paper: Maria Bauza, Ferran Alet, Yen-Chen Lin, Tomás Lozano-Pérez, Leslie P. Kaelbling, Phillip Isola, and Alberto Rodriguez. Omnipush: accurate, diverse, real-world dataset of pushing dynamics with RGBD images. (PDF).
Contact: Maria Bauza (email@example.com)
In the first version of this dataset, we record each straight push in a different file. We archived the files based on the object shape. We provide different sets of data:
Inside the folder "omnipush_dataset_simplified" you will find the data for each of the 250 objects considered. For each object, we provide 6 different files depending on whether they contain input or output information, and the number of dimensions considered to represent the pushes.
The surface considered is ABS. You can get the same material from McMaster (US vendor):
We have used 250 objects to record this dataset. In the picture below, you can see some examples of the shapes considered for these objects.
As you can see, each object is build by magnetically attaching different sides to a central square piece. Additonally, some side can carry extra weight which allows us to obtain object with non-uniform mass distributions.
We provide a rendering script in Python to visualize different shapes visualization code .
Example of how data is collected (low resolution video):
This dataset also allows to study the effect of the mass ditribution on the motion of objects.