Neural-driven, active, and reconfigurable mechanical metamaterials (NARMM)

The aim of the fellowship is to deliver the first robotic matter that can shape shift on command based on the instructions it receives.

Mechanical metamaterials are engineered materials with mechanical properties defined by their structure rather than their composition. These are usually composed of building blocks (or cells) tessellated in a periodic fashion, which enable countless possibilities in terms of achievable properties. One of these properties is the ability to change shape. Deployable systems, soft robotics and medical devices, all benefit from materials whose shape can be actively controlled. Despite the great advancements in the field, current designs lack the capability of (i) activating individual cells, (ii) reconfiguring their internal structure to mimic multiple shapes, and (iii) undergoing large deformations while being intrinsically safe (i.e., soft) for human interaction. Achieving all these characteristics in a single mechanical metamaterial is indeed a challenging task. NARMM will deliver (4 year) and go beyond this (additional 3 years).

The fellowship lays out an ambitious programme designed to investigate and develop robotic matter, based on mechanical metamaterials, that is active and can reconfigure on-command. To this end, I will employ a multidisciplinary strategy that involves mechanical modelling techniques, manufacturing methods, machine learning and, at a later stage, neuroscience.

The team will start by investigating manufacturing pathways to create arrays of interconnected soft cells (similar to hollow cubes) that can volumetrically expand when pressurized. Next, we will explore strategies to selectively constrain the expansion of single cells, while others will be free to inflate. These local features will create stiffer fibers and defects, which will govern the global deformation of the robotic matter.

In parallel, we will design numerical models to predict the deformation of the matter for different locations of the constraints, and create a database of solutions. We will then train a machine learning model on such databases, to unravel the relationship between the constraints map and the global deformation of the robotic matter. Once this is done, we will be able to provide a 3D target shape through an interactive device (e.g. pc, tablet) and software (e.g. Blender) to the machine learning model, which will identify the optimal constraints map and transmit it to the physical metamaterial to initiate the shape changing.

In the long term (+3 years) the team will look into interfacing the robotic matter to respond to the neural signals from human hosts.
Using non-invasive electrodes, we will collect electrical neural activity (EEG/EMG) from human volunteers while they perform different tasks. These will be classified into several commands for the robotic matter, which will deform to a target shape and produce mechanical work.

The fellowship will benefit from a strong interdisciplinary network of partners and mentors across KCL, MIT, Harvard, Imperial, among others--- the ambition is to deliver a design platform for reconfigurable, soft robotic matter that interfaces and responds to humans, and to explore manufacturing at scale and commercialisation. In the process, we will gain important knowledge about the complex mechanical behaviour of cellular systems and how to create effective constraints at the cell level to govern the global deformation of the matter.

The societal impact of NARMM will be enormous. With ~1.1 million people every year affected by stroke (of which 1% with locked-in syndrome), 50K individuals at any time affected by amyotrophic lateral sclerosis in Europe alone, and 60K people with amputation or congenital limb deficiency in the UK, the world needs innovative robotic devices to improve people's lives and support them during the daily tasks. NARMM will establish the first step along many paths, from wearable robots to shape-shifting prosthesis.