Tactile sensing has recently been used in robotics for object identification, grasping, and material identification. Although human tactile sensing is multimodal, existing tactile material recognition approaches use vibration information only. Moreover, material identification through tactile sensing can be solved as an continuous process, yet state of the art approaches use a batch approach where readings are taken for at least one second. This work proposes a recursive multimodal (vibration and thermal) tactile material identification approach. Using the frequency response of the vibration induced by the material and a set of thermal features, we show that it is possible to accurately identify materials in less than half a second. We conducted an exhaustive comparison of our approach with commonly used vibration descriptors and machine learning algorithms for material identification such as k-Nearest Neighbour, Artificial Neural Network and Support Vector Machines. Experimental results show that our approach identifies materials faster than existing techniques and increase the classification accuracy when multiple sensor modalities are used.
- Robotic tactile sensing
- Multimodal classification
- Recursive material classification
- Supervised learning