Evaluation, selection and validation of force reconstruction models for vision-based tactile sensors

Vision-based tactile sensors can provide robots with high-resolution distributed force applied to robots' fingertips. They reconstruct the force distribution from the measured deformation of their elastic bodies based on force reconstruction models. The force reconstruction models are various mathematical expressions to approximate the deformation-force relationship of the elastic bodies. The force reconstruction models determine the force reconstruction accuracy, modeling cost, and computational consumption of the sensor. Therefore, selecting the force reconstruction model is an important issue in the studies of vision-based tactile sensors. This study focuses on the evaluation and selection of the force reconstruction models. A method for evaluating the nonlinearity phase of the elastic body is proposed. Then, three representative force reconstruction models are evaluated, validated, and compared theoretically and experimentally under various load conditions. This study provides quantitative and theoretical guidance for selecting the force reconstruction models and makes the model selection rigorous.