This is a set of scripts based on the functions in the file hupl.py for the incremental learning of human-like target postures for the humanoid robot ARoS. A human learning-by-doing cycle is mimicked by the introduction of a Variable-kernel Similarity Metric (VSM) that provides a measure of distance between different situations of a workspace. The scripts training_vsm.py and predicting_vsm.py are used by the Motion Manager to train on collected optimal data and predict human-like arm configurations on novel situations, respectively. The description of the leaner and some important results have been published in G. Gulletta, W. Erlhagen and E. Bicho, "Continual Learning of Human-like Arm Postures," 2021 IEEE International Conference on Development and Learning (ICDL), 2021, pp. 1-6, doi: 10.1109/ICDL49984.2021.9515565.
The purpose of this document is to provide a brief introduction of the scripts, while more technical details are described in the Wiki pages.
cd /home/${USER}
git clone https://github.com/zohannn/HUPL.git