4. Adding new objects to the grasping pipeline

• Copy the object meshes

  Copy the meshes to the grasping_pipeline/models/{DATASET_NAME} directory. The meshes should either be in .stl or .ply format and must be scaled to mm (i.e. 1 ‘unit’ of the mesh should equal 1 mm). The name of each mesh should be the same as the names that the object detector and pose estimator return.

• Copy the grasp annotations

  Copy the grasp annotations to the grasping_pipeline/grasps/{DATASET_NAME} directory. The grasp annotations should be saved as a ‘.npy’ file and should contain a numpy array of shape (N, 1, 16), where N is the number of grasps and each row contains the flattened 4x4 transformation matrix. These annotations can be created with this blender script.

• Run the create_model_info.py script

  Update the create_model_info.py script with the new dataset name and run it. The script will show you two different bounding boxes and you have to choose which one fits better.

• Update the config/placement_areas.yaml config file (optional)

  Add the new objects to the config file and define the placement area for each objects. This is only necessary if you want the robot to be able to place the objects.

• Update the grasping_pipeline/config/config.yaml file

  Update the dataset entry in the grasping_pipeline/config/config.yaml file with the new dataset name. Alternatively, you can update the grasping_pipeline/dataset rosparam-parameter.