Extracting contact parameters from tactile data using artificial neural networks

In this paper neural networks are used to recover contact parameters from tactile sensor data. Tactile sensors are typically modeled using linear elasticity. Even under strong assumptions, these models are very difficult to solve. The finite-element method (FEM) provides a more accurate and realistic alternative to construct and solve models of a tactile sensor. The solutions obtained using the FEM, however, are numerical and do not directly provide analytical relationships between the sensor output and the contact parameters. Artificial neural networks (ANN), therefore, provide an ideal method to model these relationships, whereby the stress distributions and the associated contact parameters serve as the training data. We describe in this paper our attempt at using ANN to compute the contact parameters of contact force (both tangential and normal), indenter width, and indenter position. Simulation results are presented to validate the proposed approach.