README.md

Running the 6 DOF Bin Pick and Place Demo

In the demo/ directory, download the benchmark tasks to the tasks/ directory

wget -qO- https://multiarm.cs.columbia.edu/downloads/data/benchmark.tar.xz | tar xvfJ -
mv benchmark tasks/

To run the demo for 500 experiments

python demo.py

This will create a results folder, where the results csv file will be saved, and a simulation folder, where the simulation pickles will be saved.

To evaluate the results

python evaluate_results.py --result_dir path/to/results/dir

where path/to/results/dir looks something like results_2021-01-01_18-06-35. The output should look something like

{
    "num_exps": 500,
    "num_valid_exps": 298,
    "num_success": 232,
    "avg_steps": 6206.353448275862,
    "num_plane_collision": 31,
    "num_robot_collision": 35,
    "num_timeout": 0,
    "success_rate": 0.7785234899328859
}

To visualize the simulations, install the PyBullet Blender Plugin, then import the desired simulation pickle file in the simulation folder from Blender's GUI.