Neuromorphic Twins for Networked Control and Decision-Making

TL;DR

This paper explores the potential of analog neuromorphic hardware for remote control and decision-making in networked control systems, proposing a theoretical framework that extends digital twin concepts to neuromorphic computing.

Contribution

It introduces a theoretical approach to neuromorphic twins, analyzing their computability for control tasks within the Blum-Shub-Smale framework, contrasting with digital hardware limitations.

Findings

Analog neuromorphic hardware can theoretically perform control tasks beyond digital limits.

The paper establishes a computability framework for neuromorphic twins in control systems.

It highlights the potential of neuromorphic computing for future distributed control applications.

Abstract

We consider the problem of remotely tracking the state of and unstable linear time-invariant plant by means of data transmitted through a noisy communication channel from an algorithmic point of view. Assuming the dynamics of the plant are known, does there exist an algorithm that accepts a description of the channel's characteristics as input, and returns 'Yes' if the transmission capabilities permit the remote tracking of the plant's state, 'No' otherwise? Does there exist an algorithm that, in case of a positive answer, computes a suitable encoder/decoder-pair for the channel? Questions of this kind are becoming increasingly important with regards to future communication technologies that aim to solve control engineering tasks in a distributed manner. In particular, they play an essential role in digital twinning, an emerging information processing approach originally considered in the context of Industry 4.0. Yet, the abovementioned questions have been answered in the negative with respect to algorithms that can be implemented on idealized digital hardware, i.e., Turing machines. In this article, we investigate the remote state estimation problem in view of the Blum-Shub-Smale computability framework. In the broadest sense, the latter can be interpreted as a model for idealized analog computation. Especially in the context of neuromorphic computing, analog hardware has experienced a revival in the past view years. Hence, the contribution of this work may serve as a motivation for a theory of neuromorphic twins as a counterpart to digital twins for analog hardware.