Shape estimation based on Kalman filtering: Towards fully soft proprioception

Abstract

An innovative methodology to realize a sensing system able to estimate the shape of a soft robot arm without hampering “softness” is presented. The system is based on a low-cost plastic optical fiber (POF) used as curvature sensor and on a simplified steady-state model, both integrated in an Adaptive Extended Kalman Filter (AEKF). Sensory feedback was obtained through accelerometers and it was used as quantitative benchmark for the AEKF. The AEKF estimation turned out to be more accurate (RMS error < 5°) than the model prediction alone and the soft sensor alone, thus supporting the proposed fully soft proprioception strategy.