This repository contains to code that was used to benchmark models that map tactile signal to contact force for the Xela Robotics uSkin sensors mounted on the Allegro Hand robot. It also contains some datasets for the index, middle and ring fingertip arrays and a middlefinger square arrays.
Install docker engine Run the launch_container.sh script:
sh launch_script.sh
use python3 to run individual model trainnings (m0.py to m8.py) or all models evaluation with autotest.py.
Results are recorded in the results folder.
pickle_data contains the dataset. The dataset is composed of sequences of rotated force label in the common taxel frame force_label.pkl, tactile signal called tact_data.pkl, a common time vector called time.pkl and a metadata.json file that contains 0 offset for each force axis and for each taxel axis.
Fingertip arrays contains 30 3 axis taxels. Square arrays contains 16 3 axis taxels.