grasp_learning

Learning to grasp MVBBs with Soft-Hands

Creating poses

From the python folder, you can execute python grasp_boxes_batch.py db_file.csv to create an output database file db_file_simulated.csv with all successful and unsuccessful grasps, given a database db_file.csv containing box dimensions for a given box, and desired o_T_p pose which describes the palm pose in the object frame of reference. The box frame of reference is placed at the center of its axis, and is axis-aligned. The data format for the file is:

box_x_dim, box_y_dim, box_z_dim, p_x, p_y, p_z, q_x, q_y, q_z, q_w

With [p_x, p_y, p_z] the translation part of the p_T_h transform, and [q_x, q_y, q_z, q_w] the unitary quaternion representing the rotational part of p_T_h

The script simple_batch_splitter.py can be also called in the same form, python grasp_boxes_batch.py db_file.csv and will parallelize the computation over all available cores in the machine. The obtained simulated database can then be analyzed with the view_boxes_grasps.py script.

For instance, the pipeline would look something like:

cd python;
python simple_batch_splitter.py ../db/box_db_test_only_successful.csv
python view_bozes_grasps.py ../db/box_db_test_only_successful.csv