Our brain as an example

As a young child, she mainly played with Playmobil, Krauhausen explains somewhat regretfully. “It’s only now that I’ve discovered how much fun Lego really is.”

At the beginning of her PhD track, she found herself diving into a giant container of building blocks to assemble a robot designed to help her with her experiments. Krauhausen wanted to see if she could teach this robot to navigate a maze.

But the most remarkable part is inside the robot itself: it is controlled by a neuromorphic circuit. “In the lab, we developed artificial synapses – flexible connections that enable the individual neurons in our brain to communicate with each other. We make these synapses with organic polymers that the human body can tolerate. Because eventually, we want to start using them for applications within the body: a smart prosthesis or a sensor capable of adapting inside your body.”

Super-efficient

For some time now, researchers have been studying the functioning of our brain as an example for the development of a new generation of computer systems.

Since we are increasingly using smart devices and artificial intelligence, we also need new, and above all, more energy-efficient computer technology, Krauhausen explains. “And our brains are very good at that. They use a mere 20 watts per hour for very complex tasks, something even the most ingenious supercomputer is nowhere near achieving. Our brain is a great source of inspiration.”

Whereas traditional computer chips consist of silicon and metal, TU/e research group Neuromorphic Engineering is working on an organic system based on water and salt, just like our brains. There, synapses can transmit electrical signals thanks to an ingenious system of salts and salt-gated channels.

The electronic circuits used by Krauhausen also have such a salt channel. “A very small one, though, think about 100 to 200 micrometers. So that meant a lot of microscopy work in the cleanroom. Within our group, we work with several researchers on different aspects of the same artificial neuronal system. This ranges from developing new organic materials and researching new algorithms to fine-tuning the device’s mechanical behavior.”

Autonomous robot

The focus of Krauhausen’s research – a collaborative effort between TU/e and the Max Planck Institute of Polymer Research – is on the actual application of the artificial synapses developed in the lab. She reaches for her dissertation in which she has fleshed out three examples in detail; this was partly done in Germany. “I started by training a single artificial synapse, in the Lego robot. It had to learn how to find the exit of the maze by itself."

"I showed the robot pieces of brightly colored paper: turn left when you see this signal, right when that signal is absent. In the end, the robot was able to find its way through several mazes all by itself. These are first steps for applications such as a rescue or space robot; any situation where a robot needs to be able to act autonomously and adapt to a new environment, but where a large battery isn’t practical.”

Hot and cold

In a follow-up study, Krauhausen made the transition to a device with two artificial synapses, a robotic arm. This arm has a 3D-printed hand, made of a very soft and flexible material. A potential interactive prosthesis, Krauhausen points out for a future application. She used multiple sensor types to allow the robotic hand to feel the difference between hot and cold.

In the first few training sessions, the hand was supposed to pick up a cup. It worked, Krauhausen recounts enthusiastically. “After that, the experiments became more exciting. We spent a long time training the hand to distinguish between cold and hot surfaces. It was a real victory when, in an experiment, the hand avoided the cup of hot tea – ouch, dangerous – but instead, grabbed the cold cup.”

Her findings mark a significant technological advancement in intelligent robotics. And this is being recognized, as it has just been announced that they will soon be published in leading scientific journal Nature Communications, an elated Krauhausen informs us.

In the third study, Krauhausen tried to scale up the system even further by simulating a network of artificial synapses capable of predicting blood glucose levels.

“The current system is as follows: a patch on your skin communicates with your smartphone or the cloud to monitor blood sugar levels allowing the patient to manually make a next prediction. But thanks to the use of artificial synapses, the computer calculations for glucose prediction could also be done locally on the skin. We were able to demonstrate that the algorithms executed by our artificial neural network were just as effective as with the traditional software network.”

New architecture

Krauhausen’s work successfully demonstrates the practical applications of organic artificial synapses outside the lab for the first time. But in order to realize actual products in the future, a lot of development is still needed, especially in the area of algorithms, she emphasizes.

“Of course, follow-up steps need to be taken in terms of the device itself. But that aside, this is really an entirely new form of computing. We have to let go of the traditional system architecture, and many new algorithms are needed. So it’s a highly multidisciplinary project, which is why we’re very happy to be part of the ICMS (Institute of Complex Molecular Structures, Ed.). It requires the collaboration of a wide range of researchers, from chemical engineers, mechanical engineers, material experts and soft robotics engineers to mathematicians and IT specialists. We had to learn to speak each other’s languages and find a common ground to keep sight of the big picture. But in the end, many bridges were built”. She laughs. “Including ones made of Lego.” 

Source: Cursor/Nicole Testerink