Bin Picking Dataset Generation Environment

The scripts uses Pybullet and Open3D to create virtual clutter bin scenario dataset (rbg, depth, segmentation, point cloud) for AI training

Dependencies

• Open3D
• Pybullet

Steps to run and collect data

All 3 scripts mentioned below will use the predefined setting to generate dataset and folders.
You might need to change the generation parameters in data_generation_setting.json

1. Launch Pybullet to drop items

   python 1_pybullet_create_n_collect.py

   This step will:

   • create pybullet virtual scene with a bin
   • drop N x item(s)
   • record the item poses
   • record the image (rgb,depth,segmatation)
   • clear the scene and repeat with N+1 until maximum limit

2. Use Open3d to reconstruct the full scene

   python 2_open3d_reconstruct_pcd_scene.py

   This step will:

   • reconstruct the full scene base on the item poses record in step 1
   • compute the center scores based on the object model for each points
   • record and save the point cloud data with center score and segment index

3. Filter reconstructed data based on camera FoV

   python 3_generate_pcd_fov_points.py

   This step will:

   • Filter the full scene point cloud data ((remove blocked points ) based on camera pose
   • save the filtered data into .H5 format

Example on the captured data

About Object Model

Model preparation:

1. Get your model in .obj format
2. Convert .obj with refine collision model ( change the path folder inside the script)

   run util/convert_obj_convexhull.py

3. Follow other model template to create a .urdf model

Note: make sure the object generate from .urdf is having the same orientation as the orientation from .obj

Acknowledgemets

Thanks for YaJun for providing the initial code for this development.

The Gear and Ring model included in this repo is from IPA Dataset learn more

The dataset is generate and recorded base on the steps mentioned in IPA Dataset.