Object Classification with Tactile Sensing
Inspired by the remarkable capabilities of human touch, researchers have been developing artificial tactile sensing systems to imbue machines with similar capabilities. These systems aim to enable machines to interact with the environment in a more intuitive and sophisticated manner, opening up new possibilities for various fields, including robotics, prosthetics, healthcare, and human-machine interfaces.
We use data collected from the DIGIT optical tactile senor to perform object classification of different US coins: Pennys, Nickels, Dimes, and Quarters. We generate both real and simulated data to train neural networks for performing this classification. We achieve up to a 96.81% in test accuracy in identifying US coins with our transfer learning approaches, and 25% training from a 2 layer CNN from scratch. We utilize transfer learning to achieve these results and compare the results from several different prebuilt models.
I worked as a part of a 3-person team to complete this project.