Modeling Novel Soft Mechanosensors based on Air-Flow Measurements

Stefan Escaida Navarro (1) , Olivier Goury (1) , Gang Zheng (1) , Thor Morales Bieze (1) , Christian Duriez (1)

1 DEFROST - Deformable Robots Simulation Team

In this paper, we introduce a new pneumatic mechanosensor dedicated to Soft Robotics and propose a generic method to reconstruct the magnitude of a contact-force acting on it. Changes in cavity volumes inside a soft silicon pad are measured by air-flow sensors. The resulting mechanosensor is characterized by its high sensitivity, repeatability, dynamic range and accurate localization capability in 2D. Using a regression found by machine learning techniques we can predict the contact location and force magnitude accurately when the force magnitudes are within the range of the training data. To be able to provide a more general model, a novel approach based on a Finite Element Method (FEM) is introduced. We formulate an optimization problem, which yields the contact load that best explains the observed changes in cavity volumes. This method makes no assumptions on the force range, the shape of the soft pad or the shape of its cavities. The prediction of the force also results in a model for the deformation of the soft pad. We characterize our sensor and evaluate two designs, a soft pad and a kidney-shaped sensor, in different scenarios.

Mots clés: Learning for Soft Robots; Force and Tactile Sensing; Soft Sensors and Actuators; Modeling

Domaines: Robotique [cs.RO]; Automatique / Robotique

Stefan Escaida Navarro, Olivier Goury, Gang Zheng, Thor Morales Bieze, Christian Duriez. Modeling Novel Soft Mechanosensors based on Air-Flow Measurements. IEEE Robotics and Automation Letters, 2019, 4 (4), pp.4338 - 4345. ⟨10.1109/LRA.2019.2932604⟩. ⟨hal-02239080⟩