Neuromorphic BrailleNet: accurate and generalizable braille reading beyond single characters through event-based optical tactile sensing

Abstract Replicating fluid, human-like Braille reading requires artificial tactile systems that can maintain high recognition accuracy across varying contact depths and scanning speeds. To achieve this, we present an end-to-end neuromorphic pipeline utilizing the Evetac optical tactile sensor to directly decode dynamic contact events during continuous, real-time sliding motion. We combine spatiotemporal segmentation of characters with a lightweight ResNet-based classifier to achieve high-accuracy character recognition across varying indentation depths and scanning speeds. Our system achieves near-perfect accuracy (≥98%) at 0.6 -1.5 mm indentation depth and generalizes across multiple Braille board layouts and scanning speeds. On a physical Braille board featuring daily-living vocabulary, our system achieves above 90% word-level accuracy, demonstrating high accuracy and robustness to noise in a real-world application. These results position neuromorphic tactile sensing as a scalable, low-latency solution for robotic Braille reading, with implications for tactile perception in assistive and robotic applications.Journal vv (yyyy) aaaaaaAuthor et al artificial tactile systems that combine compliant, skin-like sensors with continuous sliding motions and temporally aware processing (e.g. SNNs), and sometimes even aim to replicate SA/RA-like afferent function [11]. However, despite all these advances, human-like tactile robustness still remains elusive, emphasizing the need for accurate, high-speed sensors and robust tactile perception pipelines.