Simulation and analysis of neuromorphic tactile data for object interaction speed detection

Abstract

The sense of touch provides an intimate connection to our surroundings and is a critical component for navigating the environment and manipulating objects. To better capture and convey touch information to prosthetic arm users or autonomous robots, it is useful to leverage biological representations of tactile signals for rapid and efficient processing. We developed a software pipeline that simulates and analyzes biomimetic tactile signals for developing better techniques and approaches for robotic systems that rely on the sense of touch. Touch sensor signals were simulated with a virtual hand model (MuJoCo HAPTIX) and were converted to a biological representation of tactile afferent spiking activity using existing neuron models (TouchSim). We simulated virtual hand object interaction, at three different movement speeds, and found that biomimetic tactile signals yield better results (91%) compared to using traditional force signals (61%) at classifying the different hand speeds. This software toolkit enables researchers to quickly generate, capture, and analyze biologically realistic tactile behavior for creating smarter and more effective robotic systems.