Sensor design and model-based tactile feature recognition

This paper 1 presents the design of a flexible tactile sensor and a model-based approach for the pose estimation and surface reconstruction of objects in a gripper. We show that the proposed sensor composite can be easily attached to almost all object shapes, while still achieving a high spatial sensor resolution and a high force sensitivity. Since machine learning algorithms require a large data base and do not offer the scalability of training data, the approach that we prefer here uses model-based feature classification. In order to improve the accuracy of our approach, we investigated fundamental sensor properties and applied sustainable correction methods to the data processing. Finally, the sensor's operability and the evaluation results have been verified in a pick-and-place application for two different grippers.