learning-joint-limits

Code for the experiments in "Probabilistic Inference of Human Capabilities from Passive Observations". Accepted to IROS 2024.

Setting Up the Environment

To set up the virtual Python environment and then run it, run the following commands:

python3.8 -m venv learning_limits
source learning_limits/bin/activate

Once in the environment, install the necessary dependencies using the requirements file provided:

pip install -r requirements.txt

You should now be able to run the experiments.

Running the Experiments

3D Manipulator

In order to run the 3D manipulator code, execute the following commands while in the virtual environment:

cd 3D_Manipulator/
python3 Manipulator_3D.py

There are several flags that allow changes to parameters of the script:

• --verbose: Enable verbose mode for detailed logging. Default: false.
• --samples: Number of samples for the individual task. Default: 2000.
• --thresh: Threshold for accepting a sample. Default: 0.02.
• --cartesian_thresh: Threshold for accepting a sample with Task space trajectories. Default: 0.01.
• --kernel: KDE Kernel width parameter. Default: 0.5
• --runs: Number of experiment runs with the same goals. Default: 30.
• --goals: Number of different goals (tasks). Default: 10.

The results will be output in a pickle file. To analyse them, run the analysis.ipynb notebook.

HRI Scenario

To run the HRI scenario code, execute the following commands while in the virtual environment:

cd HRI_Scenario/
python3 Main.py

There are several flags that allow changes to parameters of the script:

• --samples: Number of samples for the individual task. Default: 5000.
• --thresh: Threshold for accepting a sample. Default: 1.5.
• --cartesian_thresh: Threshold for accepting a sample with Task space trajectories. Default: 2.5.
• --kernel: KDE Kernel width parameter. Default: 0.5
• --runs: Number of experiment runs with the same goals. Default: 1.
• --observations: Number of different observations provided to the learner. Default: 10.

The results will be output to a results folder. To analyse them, run the plotting.ipynb notebook.