Supplementary Material - Learning efficient haptic shape exploration with a rigid tactile sensor array

**Recorded data set**<br> The file "objects.hdf5" contains the recorded data that was used in the experiments of this work. It is composed of 201 x 201 data points for each of the four different objects, leading to ~161000 data points in total. The code for generating the data set was written in ``python 2.7`` while the file itself was created using the [h5py](https://www.h5py.org/) package **File structure**<br> The file is built out of four containers with shapes (201, 201, 256): ``` objects.hdf5/ ├── object_1/ | └── (201, 201, 256) ├── object_2/ | └── (201, 201, 256) ├── object_3/ | └── (201, 201, 256) ├── object_4/ | └── (201, 201, 256) ``` The data is provided as triples that are containing the information of the location *x* and orientation *phi* of the utilized tactile sensor, together with its corresponding flattened and normalized pressure vector *p*, i.e. (*x* ,*phi* ,*p*).