Neuromorphic Learning in Organic Adaptive Biohybrid Systems

Introduction

Artificial intelligence has demonstrated unprecedented advances in pattern and image recognition and is widely expected to significantly increase progress in smart healthcare devices. However, it continues to rely on inefficient supercomputers operating remotely. On the other hand, relevant information for these applications mostly exists locally at the physiological level. Smart personalised bioelectronic applications can be tailored to a specific and unique case – or person – with the ability to be adapted, trained, and optimised over time.

Project Overview

In this ERC project, organic neuromorphic engineering is combined with bioelectronics to achieve a tuneable neuromorphic platform. This platform will locally monitor and modulate biosignals for the dynamic and adaptive learning control of a proof-of-principle soft robotic actuator.

Challenges and Opportunities

Due to their compliant and non-linear characteristics, soft actuators are difficult to model and thus present an ideal opportunity to demonstrate neuromorphic learning control. Organic electronic materials have been successfully implemented as building blocks in neuromorphic computing and bioelectronic applications. Particularly, mixed ionic-electronic conductors possess exceptional characteristics for use in biological environments.

Development of the Organic Neuromorphic Platform

At the interface between mechanical engineering, materials science, neuromorphic engineering, and bioelectronics, neuro-labs will develop an organic neuromorphic platform through:

1. Optimisation of organic materials and circuits
2. Integration of sensors, neuromorphic devices, and microfluidics

We will develop a closed-loop adaptive biocircuit and demonstrate local tuning and neuromorphic learning control of a soft gripper.

Future Directions

Finally, we will show optimised biocontrol of the gripper using biohybrid synapses modulated by the neurotransmitter environment, directly tuning the feedforward parameters in hardware. This will open a completely new field of adaptive neuromorphic biointerfaces and inspire a novel conceptual approach for learning control.