The relationship between materials science and robotics, automation and artificial intelligence (AI ), is very much two way: Materials science as applied to robotics & AI Vs. Robotics & AI (and also Machine Learning (ML)), as applied to materials science. This section discusses the former.
Robotics – based on the idea of constructing remotely controlled or autonomous mechanical analogues of humans or animals, manipulating objects, controlling processes, or gathering gathering data is always conjoined with the parallel idea of AI, based on the idea of mimicking human or animal intelligence, with which to better control the robots. Automation, is the replacement of human labour by such machines within manufacturing, and within hostile or hazardous environments.
Robotic systems are widely used within a number of industries – principally in mass manufacturing plants (think car/automobile manufacture, electronic assembly), but also in medical surgery (think keyhole surgery, eye surgery, brain surgery ), and in surveying (think deep sea mining and Mars ‘explorers’), inspection (think nuclear plant inspection, pipeline inspection, etc), search and rescue (S&R) operations (think earthquakes, military reconnaissance, etc). Robots may be tethered (as in surgical and manufacturing applications), or free roaming (as in maritime, aerial, extraterrestrial or military settings). With the advent of autonomous vehicles (self driving cars), and similar applications, we will doubtless all encounter robots of one sort or another, first hand.
Several areas of challenge exist for robotic systems (partially depending on their application area), and several of these objectives tend to be contravariant (eg: power Vs weight) :
- cost
- precision and accuracy
- speed, power
- dexterity and sensitivity
- size, weight
- endurance – for self-powered, untethered operation
- reliability
Developments in robotics are closely dependent on available computing technology (particularly for the AI/ML component) (see section on computing and communications technology), and also the energy/power component, for untethered operations ( see section on energy). Other areas of dependency are cost and weight, especially for structural components, which may benefit from advances in materials traditionally aimed at automotive and aerospace, for example ( see sections on land transport, aerospace).
Materials requirements that might be deemed more particular to the robotics and automation industry than others, include:
- actuator technology – particularly, high speed, high precision, high reliability, over a range of power bands
- sensor technology – not only for gross positioning (by way of accurate position transducers), and artificial vision (object and scene recognition), but also increasingly in the area of haptic/tactile systems, equilibrioceptive/inertial systems, and proprioceptive feedback systems – for example enabling robots to handle delicate and tiny objects safely and/or orient themselves accurately in rapidly changing environment or circumstances.
- computation technology (see separate section: electronics, telecommunications, computing, and also the sections below)
- communications technology (see separate section: electronics, telecommunications, computing , and also the sections below}
- energy transport/transfer technology (see also separate section: energy , manufacturing, and also the sections below}
Traditionally these functionalities would be integrated and assembled as separate subsystems with the robot. Latterly however, advances in multifunctional composite materials, have given rise to a new approach that combines all (or many) of these functionalities within a single material – so called ‘robotic materials’, from which are made ‘material robots’. In many ways this approach is very similar to the original concept of robots as imagined by Karel Čapek back in 1920, and offers many advantages, not only in operation, but also in the manufacturing and assembly of the robots themselves.
Three highly desirable properties of ‘robotic materials’ have been identified:
- functionality independent of size, i.e., performance is unchanged when cut in half (up to a reasonable discretization)
- self-similarity and bulk reconfigurability, i.e., consisting of homogeneous elements that can be arranged in either an amorphous or a discrete, grid-like fashion, and
- robustness, i.e. the material does not lose its capabilities should failure of any constituent elements occur.
In the realm of AI, there has been considerable ongoing debate as to whether the field, intrinsically requires new materials solutions in order to hit the high benchmarks set – for example of the computational power, and flexibility of the human (or animal) brain, relative to its size, weight and power consumption. One important area of study on this domain is the field of neuromorphic computing/AI – which seeks to match or even better the performance of biological brains using synthetic analogues of biological neurons. Although still in its infancy, promising results have been obtained using spintronic, memristor and ionic exchange technology, based on now nano materials, including graphene nano wires and carbon nanotubes.
Another approach, at least to simpler computational tasks, is morphological computing, where system elements or element types, primarily purposed for one set of tasks, are repurposed to another, namely computation. Morphological computation can for example, be implemented using microfluidics, or alternatively by exploiting physical systems with high-dimensional non-linear dynamics, such as feedback loops around sensor and actuator clusters. Distributed computing, computing at this scale, whether it be discreet or amorphous in implementation, is as it as the somewhat larger scale of the Internet of Things (IoT), a heavily researched field, where we should expect to see rapid advances within short time frames.
Generally however, challenges remain, particularly in the fast and data intensive computation that is required of AI and machine learning systems in use use with such material robots – for example Fourier transforms, convolutional neural networks and transformers; much current research is directed at implementation of algorithms within the ‘neural’ resource we are able to implant in materials directly (see examples below). The method of Transfer Learning, where learning of a new task or pattern can be based on previously acquired learning (of pre-cursive patterns and tasks) is one area of active research that has the potential to be adapted here.
A big advantage of all these approaches, as that by relegating functionality to lie within material structure itself, we are further able (or forced) to further abstract functionality and the applications design process, thus making applications design that much easier to achieve (eg: via the use of ML itself), and allowing a far greater degree of autonomy to be manifest within the robots themselves. A further requirement of such materials is that they easily bond with each other, at both a structural and functional level. In addition, the vastly expanded scales and densities of sensor and actuator instantiation in such materials requires similarly vast sampling rates, communications bandwidth, computing resource, and motive power distribution; in however theory, the nature of the materials themselves may allow us to deal with these constraints more easily also, by carrying out much of the work on a cellular or network basis within the material itself.
One significant outstanding challenge with robotic materials approaches, is in the distribution of power to the vast network of sensors actuators and computing chips that they entail. Whilst Energy scavenging technology might be possible in some applications, for there is the right sort of (free) environmental energy to scavenge (eg: uniformly distributed flexing movements for piezo transducers), and whilst wireless power transfer (WPT) using purposely designed microwave transmitters or free RF energy, might be suitable in some restricted circumstances, they by no means offer a solution to more general applications. Microfluidic or chemical power systems are another approach suiting certain situations, but not all. One part of any more generalised solution to power distribution, has to be designed in redundancy and the resilience that affords, should any area of the robotic material become damaged or inoperative or starved of power. Much additional research and development is still needed to adequately address this particular challenge.
Current state of the art in such robotic materials has still a great distance to go before finding widespread applications. Examples of experimental and laboratory demonstration systems include:
- active camouflage systems, based on intelligent ‘swarms’ of synthetic chromatophores, set in a flexible rubber ‘skin’ – mimicking the camouflaging ability of the octopus ( see also section on: military sector)
- a texture detecting synthetic skin, again using an intelligent network of sensors in a rubber substrate, that can (locally) characterise the type of surface (smooth, rough, etc) by profiling the vibrations that occur when stroking or touching it.
- shape changing beams that both calculate and then assume the required stiffness, based on whole surface measurement of stress and strain as picked up by an embedded network of intelligent sensors.
- a modular, gesture detecting smart fabric, that can detect, characterise and communicate gestures of touch made upon its surface, even when many modules are combined into larger areas.
- modular building blocks, that like the synthetic fabric above, can recognise complex shapes, gestured across their surface, no matter the arrangement of the blocks and no matter there maybe small gaps in the arrangement of blocks, (ie: extensible, reconfigurable and fault tolerant).
- a smart tire, capable of sensing the variability of the immediately upcoming terrain – providing protection against skidding as well as against running over unexpected objects and surfaces – pretty much essential sensory data for the safety of autonomous vehicles
- a robotic skin capable of tactile gesture recognition as well as near-field object avoidance (see also section below on robots working with humans, and safety in real working environments).
In order to advance robotic materials technology efficiently, significant collaboration is required between materials scientists, and for example, sensor network specialists, distributed algorithm specialists, polymer scientists, and specialists in soft robotics and multi-material manufacturing techniques. It is also true to say, that the very fact of having such functionality manifest in a (range of) materials, makes their uptake into real world products that much easier.
Whilst material robots, robotic materials and soft robotics are all very much emerging technologies, the here and now is still very much ‘alive’ with robotic deployments using more conventional technologies in more conventional settings. For example, heavy duty tethered robots in automobile manufacturing and electrical and light assembly robots in the electronics, computing and telecommunications sector.
It seems highly likely that some technological convergence will ultimately occur. That is, conventional robotics will increasingly be upgraded, with new features and new functionalities – especially those depending on readily now available AI and ML solutions. These will likely include, improved tactile senses, better ‘safety awareness’, since close proximity to human beings must increase with numbers and varieties of deployment settings; greater flexibility towards different tasks, and above all lower cost. At the same time, we are likely to see the emergence of similar features, plus many more, from a new breed of soft robots and material robots, aimed in the first instance at particularly suitable settings.
One notable early example of such cross-over technology is likely to be in actuator technology. Current robots most often use geared electric stepper motors or hydraulic actuators, alongside compatible power, communications and control systems. For some applications however, (or some parts of the robot – eg: the fingers), such technology might readily be supplanted with for example ‘synthetic muscle’ either based on electrically driven polymer materials, fluidic systems, piezo materials, or memory alloys, particularly where the deployment of such approaches also confer additional benefits (eg: improved dexterity or sensitivity).
Robotics, AI and automation represent a particularly exciting arena for new materials and techniques. Potentially it can help free humans of drudgery and danger and allow us to explore places, and extract resources, never before possible. On the flip side of course there are a wealth of economic, social and ethical issues to contend with too, barely over the horizon, in addition to the many technical hurdles that remain to be overcome.