Tactile Dynamic Behaviour Prediction Based on Robot Action

Kiyanoush Nazari, Willow Mandill, Marc Hanheide, Amir Ghalamzan Esfahani

Abstract: Tactile sensing provides essential information about the state of the world for the robotic system to perform a successful and robust manipulation task. Integrating and exploiting tactile sensation enables the robotic systems to perform wider variety of manipulation tasks in unstructured environments relative to pure vision based systems. While slip detection and grip force control have been the focus of many research works, investigation of tactile dynamic behaviour based on robot actions is not yet sufficiently explored. This analysis can provide a tactile plant which can be used for both control methods and slip prediction using tactile signals. In this letter, we present a data driven approach to find an efficient tactile dynamic model with different tactile data representations. Having evaluated the performance of the trained models, it is shown that the tactile action conditional behaviour can be predicted in a sufficiently long time horizon in future for doing robot motion control.