The brain is the fundamental organ of the central nervous system. As the main driver of human cognitive functions, it governs different processes by sending and receiving chemical and electrical signals. Some of its messages remain within the brain, while others are relayed across the body. To do this, it relies on a network of around 85 billion neurons. These neurons, its true “building blocks”, are the basic cellular units of nervous tissue, transmitting information from one point to another in the human body and enabling coordination. The precision of these interactions – and of this intricate “chain of command” – is now inspiring one of the most pioneering experiments in robotics: what specialists call “neurocellular automata”.

“Simply put, it is a computing paradigm inspired by the neuronal structures of the human brain, composed of clusters of neurons that communicate with each other and ‘inform their neighbours’ clusters on how to act, while also determining how the robot should operate locally”, clarifies Kasper Støy.

A professor of robotics and a member of the “Robotic Evolution, and Artificial Life” (REAL) research group at the IT University of Copenhagen, Støy also serves as the scientific coordinator of MOZART, an EU initiative developing reconfigurable surfaces embedded with soft sensors and controlled by AI-powered learning tools for handling fragile and delicate objects. Among the project’s partners is the Italian Institute of Technology (IIT) in Genoa, where Lucia Beccai leads the Soft BioRobotics Perception research line. “We are developing a system to integrate these technologies into a movable platform. One of the main applications developed within the project is the so-called AUTOMAT, an autonomous manipulation system integrated with ‘morphing modular mats". These morphing modular mats are a kind of ‘sensing skin’ equipped with sensors that you lay on this movable platform, allowing it to ‘feel’ the objects placed on it, locate their exact position, and ‘manipulate’ them according to their characteristics and weight”, she notes.

To describe it, Støy uses the image of a next-generation conveyor belt, moving in 3D. “Compared to the ones that deliver luggage in airports, what we are trying to do is add a third dimension to the motion that can be imparted to the transported objects. This means they can be not only moved from point A to point B, but also flipped and rotated. The surface moves autonomously, manipulating them gently and adapting to their shape”, he explains. Far from sci-fi stereotypes, this highly sophisticated solution appears as a surface of about one square meter composed of around ten modules, each of which functions as a robot in itself. “People should imagine a soft membrane supported by these modules that can lift this top surface and change its shape depending on what is placed on it and how it needs to be manipulated”, adds Støy. “In our system, there is not just one robot but several of these modules positioned next to each other – like a checkerboard of small robots working together. If one side lifts and another does not, the product falls off. This is why they all have to communicate and remain coordinated. And this is where neural cellular automata come in. As in the human brain, there can be a neural structure controlling each element on the checkerboard while also communicating with the neural networks that control the surrounding elements”, he remarks.

What further enhances this system’s performance is the fact that it can be trained through machine learning to adapt its behaviour based on the weight, softness, fragility and other characteristics of the objects placed on it. “Manipulating an object depends on multiple parameters – such as mass, friction, and so on – making it extremely challenging to manually code how to perform each action”, MOZART’s scientific coordinator specifies. “Therefore, instead of hard-coding behaviour, we give the object to the robot and let it try many possible combinations, learning through exposure to these variables. And this knowledge is then encoded in the neural network of the neural cellular automata”. The result is a system that, according to the project’s ambitions, could enable automation in industrial sectors where it is still hardly implemented. “The prices of many products available on the market today are only possible because their production chains are automated. By introducing this technology into sectors where automation is currently not – or only partially – possible, we could, for instance, reduce the price of many food products to make it affordable to a broader group of consumers”, he argues.

Current applications being explored are particularly in the food-packaging sector, for products such as tuna and poultry. Beccai – who at IIT in Genoa also acts as principal investigator of the Soft Biorobotics Perception research unit – elaborates: “Packaging processes are very demanding, and in some countries, even the slightest damage to the products can make them unfit for sale. This is where soft robotics can prove to be precious: while a rigid manipulator like a metal gripper might harm the food, embedded smart sensors allow more natural movements and extremely gentle, safe interactions”. According to Støy, positive impacts on workers’ health and conditions would also be significant. “Imagine a cooked tuna. Because it is completely loose, it is very hard to manipulate for processing and packaging. This is work for which there is currently no automation. So, there are people who spend their days lifting and skinning fish weighing between four and eight kilos, in a very demanding and exhausting routine. And that’s where our technology can help, too”, he stresses.

After three years mainly dedicated to research and now in its final year, MOZART is entering the practical phase of assembling the various technologies developed. The first practical demonstration is expected in the coming months, and “the resolution of the AUTOMAT manipulation surface has been designed to handle ‘chicken-type objects’”, reports Støy. “However, once we demonstrate the concept, nothing will prevent us from scaling it up or down for much larger or much smaller surfaces and objects”.

The potential applications of gentle, soft-robotic handling are numerous, and although “their level of biocompatibility must obviously be improved, these technologies can be applied in many fields – such as assisting elderly or disabled persons in clinical environments, helping nurses delicately move patients from one bed to another, or manipulating tissues and cells that might be dangerous for humans”, concludes Beccai.