AI-on-skin: Enabling On-body AI Inference for Wearable Artificial Skin Interfaces

Abstract

Existing artificial skin interfaces suffer from the lack of on-skin compute that can provide fast neural network inference for time-critical application scenarios. In this paper, we propose AI-on-skin - a wearable artificial skin interface integrated with a neural network hardware accelerator that can be reconfigured across diverse neural network models and applications. AI-on-skin is designed to scale to the entire body, comprising tiny, low-power, accelerators distributed across the body. We built a prototype of AI-on-skin that covers the entire forearm (17 by 10 cm) based on off-the-shelf FPGAs. Our electronic skin based prototype can perform (a) handwriting recognition with 96% accuracy, (b) gesture recognition with 95% accuracy and (c) handwritten word recognition with 93.5% accuracy. AI-On-Skin achieves 20X and 35X speedup over off-body inference via bluetooth and on-body microcontroller based inference approach respectively. To the best of our knowledge, AI-On-Skin is the first ever wearable prototype to demonstrate skin interfaces with on-body neural network inference.