ycb_dataset

YCB dataset with urdf files. Ready to be used in different simulations such as PyBullet.

Usage

Have ycb folder in your project directory.

For convenience, the locations of urdf file can be found with __init__.py of ycb directly. An example is

from ycb import YCB
cracker_box_urdf = YCB["cracker_box"]

You need to have pathlib installed.

pip install pathlib

Procedure

Here is the procedure of how this repository is made.

• The dataset is downloaded from the YCB dataset website with script download_ycb_dataset.py provided by ycb-tools. Only textured models with Google 16K are used.
• The urdf files are generated with scripts build_object_library.py and object_urdf.py provided by object2urdf.
• The mass of each object is manually set based on the provided mass form. The inertia of all objects are simply diagonal matrix with value 1e-3.
• All the texture images are compressed to 80% JPG with image size unchanged with image_compressor.py. This should improve simulation performance and save space without affecting the vision pipeline.

Models not included

This repository only includes models with Google 16K. The other models, shown as below, are not included.

• models only with rgbd, not google_16k:
  • 001
  • 041
  • 049
  • 072-f
  • 072-h
  • 072-i
  • 072-j
  • 072-k
  • 073-h
  • 073-i
  • 073-j
  • 073-k
  • 073-l
  • 073-m
• low-quality models:
  • 023
  • 039
  • 046
  • 047
  • 063-c
  • 063-d
  • 063-e
  • 063-f
  • 072-g