lwelzel / grip-net

Real-time frictional Safety Margin Estimation with imaging RGB soft tactile fingertips

grip-net

Real-time frictional Safety Margin Estimation with imaging RGB soft tactile fingertips. Implements a basic fully connected DNN with SOTA performance compared to CNN on training data, and which can outperform SOTA CNN in generalization. The DNN architecture and hyperparameters have not been fine-tuned. The input to the DNN are extracted dot cordinates and radii from preprocessed images. The preprocessing was done by Jingwen Tang, here. The data courtesy of Jingwen Tang, here.

Pre-trained models are available as pytorch lightning checkpoints, and pytorch state dictionaries. Normalized data is available as .csv.

Description

Nice figures are from R. Scharff et al. 2022, "Rapid manufacturing of color-based hemispherical soft tactile fingertips." DOI:10.1109/RoboSoft54090.2022.9762136

Performance:

grip-net's Issues

data file

Thank you for the nice documentation! Could you remove the link to the .csv file please？