Robots working in unstructured environments and alongside people need to be able to sense contact information from both intentional and unintentional interactions. Soft and skin-like tactile sensors can provide a robot with physical contact information from the surrounding environment and beneficially also supply a compliant interface for conformal contact and impact protection.
Raw data from soft tactile sensors is normally nonlinear and has high dimensionality. These characteristics make calibration of soft tactile sensors very complicated, such that one often cannot convert the readings into physical values. The research objective of this project is to implement a calibration method that is applicable to a wide range of soft tactile sensors.
We aim to develop a parameter identification method to obtain the calibration model by combining model-based and data-driven approaches. Specifically, mechanical deformation of the soft body of the sensor can be estimated through solid mechanics. The transducer model, which produces raw data from mechanical deformation, is obtained by a data-driven method.
The future outcome of this ongoing project will be to provide a generalized method that could be used to calibrate a wide range of soft tactile sensors so that their outputs can be confidently interpreted in physical units such as pressure.