# Bioinspired spiking architecture enables energy constrained touch encoding

## TL;DR

This paper introduces a bioinspired tactile system that improves touch sensing accuracy while using very little energy, making it ideal for robotics.

## Contribution

A neuromorphic tactile system combining e-skin and spiking neural networks for high-accuracy, low-power touch encoding.

## Key findings

- The system achieves 10× localization super-resolution and 32% better accuracy than state-of-the-art deep learning methods.
- The spiking neural network operates robustly on neuromorphic chips with sub-mWatt power consumption.
- The architecture generalizes well to multitouch and dynamic conditions.

## Abstract

The sense of touch is essential for safe interactions with the external world, enabling humans to rapidly detect and localize physical stimuli. Biological systems achieve these abilities through hundreds of thousands of mechanoreceptors distributed across the skin and efficiently processing vast streams of tactile information. Replicating these affordances in autonomous systems is crucial for advancing robotics. However, current tactile sensing solutions face critical challenges, including excessive wiring, high energy demnds of AI computing, and limitations in scalability and parallel processing. Here, we present a modular artificial tactile system combining a Fiber Bragg Grating-based e-skin with a spiking neural network (SNN) that mimics the early stages of the human somatosensory system. Our architecture achieves up to 10× localization super-resolution, improving localization accuracy by 32% over state-of-the-art deep learning methods and effectively generalizing to multitouch and dynamic conditions. Crucially, when implemented on a neuromorphic chip, the SNN demonstrates robustness to the constrained resolution and mismatches of analog neurons, bolstering highly parallel and sub-mWatt hardwired computation. Bioinspired connectivity is shown to functionally influence tactile processing, offering mechanistic insights in a framework that bridges physiological hypotheses, modeling, and validation in a real-world tactile scenario. These results demonstrate a scalable, energetically sustainable solution for touch perception, with immediate applications in autonomous systems requiring safe human interaction and operation in dynamic environments.

Current tactile sensing solutions face challenges in excessive wiring, high energy demands of AI computing, and limitations in scalability and parallel processing. Here, the authors present a neuromorphic tactile system combining artificial skin and spiking hardware encodes touch with high accuracy at low power.