Online Compression and Reconstruction for Communication of Force-Tactile Signals

Abstract

We focus on force-tactile signals that are generated by sliding fingertips on physical surfaces. In tactile-communication applications, these signals are sent to surface-haptics devices for tactile-texture display. To ensure the high reliability and low latency for such communication, we present a method that compresses the signals into compact representations before transmission, and approximately reconstructs them with reasonably small errors before display. Specifically, the method is based on stacked auto-encoder (SAE) and gated recurrent unit (GRU). The former achieves compression and reconstruction without considering channel noise; the latter suppresses the channel noise for the received compact representations before reconstruction. The proposed method is numerically evaluated, using HapTex and TUM184 datasets. Results show that the method (i) achieves a high compression ratio with a reasonably small reconstruction error, (ii) can be efficiently implemented for online deployment, and (iii) requires no statistical assumption of force-tactile signals and channel noise.