Sensing Through the Body - Non-Contact Object Localisation Using Morphological Computation

Abstract

Biological systems exhibit remarkable sensing capabilities by using a widely distributed sensor network and a tight coupling between body and environment. The result are seemingly highly robust and adaptive solutions. To enable the next generation of embodied robots with similar capabilities, we need to develop novel sensing and computational technologies. In this paper, we propose an approach using proprioceptive sensing and leveraging the system-environmental interaction in soft robotics based on the principle of morphological computation, i.e. the use of morphological features for computational tasks. We exploit the body dynamics of a moving octopus-inspired robot tentacle in coordination with the dynamics of the surrounding water to predict the position of objects in its vicinity without touching them. The values of proprioceptive strain sensors, which were emulated with the help of computer vision techniques on recorded videos of the experiments, and simple linear regression on these values were sufficient to solve this computational prediction problem. We were able to demonstrate that the body of the soft tentacle could be used to “feel” the location of an object by observing its own body dynamics (strain sensors) which are responding to changes (i.e. different positions of the object) in the environment.