Introduction

Autonomy is the ability of an object to perform a task without human supervision or intervention. Such a decision-making process not only requires awareness of the environment, but also the ability to process information, evaluate the relative value of options, and execute an option within the environment. When this cycle of observe–orient–decide–act is unsupervised and recurring, it is autonomous. How much “autonomy” a system needs reflects the complexity of tasks, and what it can (and cannot do) within its environment. A car is considered autonomous if it can assist your highway (environment) drive to work (objective). Similarly, if the objective of a metal bar is to lengthen as the environment heats up to turn off current to a heating element (as is common in old thermostats) it is also autonomous. We would likely, however, not consider it as autonomous as the car. This classification inconsistency is quite severe both within various materials disciplines (e.g., autonomous materials, responsive materials, adaptive materials, etc.), and even more so between engineering disciplines (e.g., between robotics and materials science). An exciting outcome of creating a framework for classifying the degree of autonomy and its relationship to characteristics of materials would be a greater understanding of present capabilities and consistent metrics to guide the evolution and interdependencies of these disciplines.

Continuing the illustrative comparison between an autonomous vehicle and a metal bar, at least two major differences in the origin of the perceived “autonomy” can be discussed. First is in the complexity of the response and the number of subtasks performed. An autonomous vehicle uses thousands of sensors of different types, hundreds of actuators, and significant computing power to navigate traffic using numerous responses—accelerating, maintaining speed, steering, etc. On the other hand, the metal bar absorbs heat from its surroundings and the crystal lattice expands—a single deterministic relationship between input and output. Due to the extensive and diverse environmental input, the car responds “adaptively” to the environment by choosing among many possible responses. In contrast, the bar responds “predictively,” in that there is only one response to heat. In other words, there is a discrepancy in the scope of the potential response space between the car and bar.

The second difference is the spatial and temporal complexity of their architectures. Our first example of a car is a system composed of various interacting (and interdependent) components. The materials and devices that comprise the components are not only optimized for the components’ function, but they are also optimized to interface with each other and provide multiplicative benefits. Their spatial and temporal arrangement is architected to deliver system-level performance, ranging from fuel economy to driver safety and passenger experience. In other words, this hierarchical complexity emerges from a system of spatially and temporally distinct components with different functions sharing information and energy. The metal bar made of steel, for example, has carbon interstices in a lattice of iron. These constitute anisotropic grains with a distribution of orientations with commensurate and incommensurate grain boundaries. The macroscopic performance of the bar derives from the relatively simple arrangement of these similar constituents. In other words, the steel bar is a material, whereas the car is a system composed of distinct subsystems, components, devices, and materials.

The complexity of the response landscape and the architecture of the system are positively related—the ability to choose among many responses scales with a hierarchy of functional components. Complex autonomy arises from interactions between many components, and hence, the autonomous system must scale positively with the number, size, and structure of its functional constituents. Also, to affect its environment, the system must be of a comparable size to its surroundings. Thus, we propose that autonomy in materials can be framed by structural hierarchy (i.e., number, size, structure). This organization also provides connectivity to the classic materials design interdependence of structure processing–properties, where processing techniques establish structure and, therefore, the resulting suite of materials properties.

Autonomous material systems framework

This correlation between complexity of autonomous behavior and of the architecture provides an approach to frame and guide the development of autonomous material systems (AMSs). Such an approach borrows from system engineering concepts and brings to mind parallels with microelectronic devices and single-celled organisms. For the sake of the following discussion, it is useful to be precise with the use of various terms. Herein, a material is defined as a collection of atomic species with a nominal composition, local structure, and a hierarchical morphology reflecting differences in local orientation, order, and composition. A composite, or system of materials, is a construction of materials, such as a fiber-reinforced composite, designed to deliver a deterministic property suite. Finally, an emergent material system is an architecture of heretofore defined materials or composites, of which property set, which includes adaptivity to multiple external stressors, is specifically designed to emerge from the interactions and interdependencies of the deterministic and discrete response of the different materials or composite constituents in a functional component. The latter, or assemblies thereof, are analogous to a microelectronic device or a single-celled organism.

There are numerous perspectives that the reader should consult on complexity and autonomy in materials.1,2,3,4,5,6,7 Many of these other approaches are complimentary to those discussed herein, albeit are challenged when required to establish correlations across numerous fields between the degree of autonomy and requisite complexity of architecture. These models include variations on effective medium theory,1,2,3 animate materials,4 biomimetic hierarchy,6 and feedback loops,7 and “residual performance methodology,” where performance gains are assessed from hypothetical superpositions of two complementary material systems.5 We also clarify that the autonomous materials’ systems discussed here are not autonomous robots being used to discover new materials.8 For more details on what could potentially define what is and is not a material, the metamaterials’ community has expended effort on this matter and should be consulted by the interested reader.9,10

Antoine de Saint-Exupéry, the French writer best known as the author of The Little Prince, wrote in Wind, Sand and Stars that “Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away.” In our paradigm, the elegant solution results in the minimum architectural complexity (number of materials and arrangement), C, for the highest level of autonomy, A—an architecture-autonomy optimization front for material systems. It is instructive to begin this framework with a biological cell as the archetype of an AMS. In the biological cell, the hierarchy of components starts from the atomic level where a limited set of elements (C, H, O, N) forms macromolecular chains or surfactants that, through their secondary structure and assembly, create functional components such as enzymes, membranes, organelles, etc. The spatial and temporal arrangement of these functional components results in the cell’s autonomous response (i.e., the key characteristics of life). The total architecture spans ~5 orders of magnitude (atom to cell; 0.1 nm–10 µm). Following Saint-Exupéry’s insight, the Pareto front is where a high level of autonomy is achieved with a low level of distinct material components and complexity. With a biological archetype, the question this framework poses to the community is given: Can synthetic or abiotic–biotic hybrid concepts provide solutions that move toward the autonomy-architecture Pareto front found in nature?

To begin, one must adopt a means to assess the relative level of architectural complexity, C, and autonomy, A. There are at least two possible approaches—analogies or de novo theory. The complexities of creating a generalized theoretical model of C and A are vast. Therefore, we resort to constructs for classification to help organize observations and concepts, and even provide guidance for relative assessments of characteristics within a grouping.

For the sake of our discussion, we propose a simple C = 0–5 scale to express architectural complexity (see Figure 1). We begin by assigning atomic elements to C = 0 (gas, liquid, or solid), and compositionally defined materials to C = 1. This compositional definition includes metal alloys, ceramics, polymers, elastomers, etc.; and includes them whether their phase is single crystalline, polycrystalline, glass, liquid, 2-D, and so on. For both C = 0 and C = 1, compositional hierarchy is limited, and structural complexity is relatively straightforward. At the other extreme, we illustrate the C = 5 archetype as a microelectronic device composed of tens of billons of transistors and interconnects—for example, IBM’s 2021 2-nm chip11 or beyond-Moore future concepts. Both composition and structural hierarchy are precisely designed and manufactured across 6–8 orders of magnitude. With these boundaries, blends, mesophases, and formulations are assigned C = 2, reflective of multiple distinct components, but relatively imperfect structural and compositional hierarchy. The increased design of precise structures across multiple length scales in composites are assigned C = 3. For example, a carbon fiber-reinforced composite π-joint not only has precision-engineered orientation of woven C-fiber tows, but the individual fiber interface is compositionally tuned to optimize load transfer, and additives to the resin impart toughness and processability. Here, the macroscopic performance arises by composition and structure control at three different length scales. Finally, C = 4 reflects three-dimensional metamaterials. The precision and accuracy of periodic or aperiodic arrangement of composition and structure required to deliver unnatural properties are substantially beyond engineered composites.

Figure 1
figure 1

Architectural complexity levels, C, shown with representative examples. From left to right: C = 0 corresponds to atomic elements (shown: argon), C = 1 corresponds to single materials (shown: a steel bar), C = 2 corresponds to blends and formulations (shown: a colloidal gel), C = 3 corresponds to composite materials (shown: a carbon fiber joint), C = 4 corresponds to 3D metamaterials (shown: a double-gyroid capacitor capable of sensing, energy storage, and load bearing,60 and C = 5 corresponds to material systems and electromechanical devices with structural hierarchies spanning 6+ orders of magnitude (shown: a microelectronic chip highlighting transistor junctions11).

Full size image

To evaluate the degree of autonomy, A, we propose using analogies with autonomous vehicles and adopt their terminology (further, we use cellular biology, colonial organisms, and the human peripheral nervous system to provide upper bounds). Among the many methodologies available in these communities, Automated Driving System (ADS) has six levels of autonomous vehicles as defined by The Society of Automotive Engineers,12 shown in Figure 2:

  • Level 0 has no automation whatsoever; the driver is fully in control, and their full concentration is warranted.

  • Level 1 has primitive driver assistance systems, such as antilock braking.

  • Level 2 has detection and alert technology, such as collision avoidance.

  • Level 3 has the vehicle capable of notifying the driver if they need to take control; the driver does not have to fully concentrate on driving.

  • Level 4 has the ability to drive autonomously in specific operational design domains where there are detailed maps available; thus, human attention is no longer required.

  • Level 5 has full automation. Here, the vehicle can operate on any road in any weather condition.

Figure 2
figure 2

Different levels of autonomy in vehicles and correlating examples of autonomy in materials systems. Top row: The six levels of autonomous vehicles as defined by The Society of Automotive Engineers. Bottom row: examples of corresponding autonomous material system. From left to right: a rock (A = 0), a piezoelectric bimorph (A = 1),13 a self-healing polymer (A = 2),23 Belousov–Zhabotinksy (BZ) hydrogels (A = 3),17 a paramecium (A = 4), and a Portuguese man o’ war (A = 5).45 For the materials shown, each level of autonomy, A, correlates to the Automated Driving System as follows: A = 0 is inert, A = 1 demonstrates feedforward actuation, A = 2 demonstrates decoupled computation, A = 3 demonstrates collocated capabilities, A = 4 demonstrates groups of collocation, and A = 5 demonstrates tight integration. Green-bordered materials have been achieved in the literature, while red-bordered are biological examples that we display to aid our discussion.

Full size image

Currently, no vehicles are capable of Level 4 or 5 automation.

With these definitions, we map observed levels of autonomy in materials, composites, and material systems to these ADS levels:

  • A = 0 has no computation, such as an inert material under most conditions.

  • A = 1 has the capability of feedforward actuation, such as in piezoelectric actuators.13

  • A = 2 has sensing and actuation, but this is a reflex action. Here, computation is decoupled. Examples include a soft gripper composed of a granular assembly that can form around any arbitrary shape14 or a prosthetic hand that uses stretchable optical waveguides for strain sensing.15

  • A = 3 has collocated sense–compute–response such as in “Smart Dust”16 or Belousov–Zhabotinksy hydrogels.17

Analogous to Levels 4 and 5 of ADS where “Human attention is not needed,” we can say that for A = 4 and A = 5 there is a “Hierarchical coordination between C components”:

  • A = 4 has groupings of these collocated systems. This level could require advanced manufacturing such as in the case of living-engineered material systems. Research in this realm has resulted in microscale “biobots,” with notable examples including the “xenobots”18 developed from amphibian heart and skin stem cells and “anthrobots”19 created from human lung cells. Versions of these biobots are capable of self-healing, self-assembling into different morphologies, demonstrating different movement classes, swarming behaviors, and even repairing separate organic structures within their environment.

  • A = 5 has tightly integrated tissues and complex decision-making abilities, for instance, an octopus tentacle,20 or Portuguese man o’ war. These examples are of interconnected systems of sense, computation, and response modules.

An important clarification of this analogy with the ADS hierarchy is that the operating goals of the vehicle are always vested in the human operator, who defines the destination, road type, and other criteria. In contrast, the operating goals of biological systems and, even some synthetic systems, are retained within the collective mass, energy, and interaction dynamics of the autonomous material system. The operating goals of biological systems are often indecipherable, and the source of much speculation, for an external observer. However, the ability of biological systems to set and adapt their operating goals is a hallmark of their autonomy.

Architecture complexity–autonomy space

Using these definitions of C and A, Figure 3 illustrates this Pareto front of architectural complexity–autonomy [C, A]. Examples from numerous fields can be discussed within this space,21,22 such as a steel bar [1, 1], self-healing composites [2, 2],23,24,25 BZ gels [3, 3],26,27 mother of pearl [1, 1], paramecia [3, 4], an autonomous vehicle [5, 4], Spot [5, 4],28 in addition to the examples previously used.

Figure 3
figure 3

A Pareto front of autonomy (A) versus architectural complexity (C) based on our autonomous material system guidelines. Several examples for this text (ranging from molecular elements to multicellular organisms) are depicted and placed according to their [C, A] values. Examples in red are biological in origin (from bottom to top: mother of pearl, a bacteriophage virus, a neutrophil, a paramecium, a slime mold, and a Portuguese man o’ war). Examples in green, with the exception of elemental fluorine, are exemplary of current scientific developments (from left to right: steel alloy bars, elemental fluorine, Belousov–Zhabotinsky gel composites, self-healing materials, the Spot robot by Boston Dynamics (photo credit: D. Newman, D. Colli, and D. Berenson), and a modern electric vehicle.

Full size image

A thermoresponsive hydrogel is considered (e.g., PNIPAm29), which is used in applications such as drug delivery and tissue engineering. These gels provide a reversible volume change when water previously bound to macromolecules is released at a first-order phase transition. Similar to the steel bar, one constituent and one deterministic response resides in the plot’s lower left, [1, 1], even though the macroscopic magnitude, speed, and direction of the reversible volume change can be tuned via composition or morphology. Increasing C–A occurs when the phase transition is tied to an embedded trigger, such as an atomic-level reaction–diffusion couple (e.g., oscillatory BZ reaction). The response can be modulated and engineered with higher-scale patterning of the BZ catalyst to match reaction–diffusion rates, concentration changes, and hydrogel volume change. These patterned (or composite) BZ hydrogels [3, 3] are able to develop synchrony, meaning that they achieve oscillating volume change and oxidation across two or more gel segments26,27,30 by converting chemical energy into mechanical swell–deswell motion.31 Implementing hierarchies of this chemical and mechanical communication between active nodes of Ru (an immobilized catalyst) harvests energy and mass from the environment to execute computation based on environmental signal processing,32,33 which has enabled applications in such areas as sensors, encryption, and artificial skins.30,34,35,36

Another example of increased C–A is self-healing polymer-based composites,23,24 with the latent ability to repair [2, 2]. Crack propagation exposes reactive constituents or ruptures vessels that release agents to fill the voids. This phenomenon has been used in electronics (electronic skins and sensors), construction (self-healing cementitious composites), coatings (metals to make microcapsules used in space), medicine (membranes and drug delivery systems), and textiles (superoleophobicity for use underwater).37 Finally, Spot is a quadruped mobile robot developed by Boston Dynamics that is capable of navigating almost any kind of terrain while simultaneously sensing its environment [5, 4]. Applications for Spot include hazmat missions, inspection of industrial sites, delivery of packages, exploration, and search and rescue.28,38 This example represents a classical system design integrating devices as well as materials, composites, and material systems with varying degrees of local autonomy.

Now it is considered where biology falls within the architecture complexity–autonomy space, as a baseline to where we are with autonomous materials systems. Viruses are an example of relatively low-level autonomy in biology. This assemblage of proteins and nucleotides is static until they enter an environment in which disassemblage can release the nucleotides to utilize cellular machinery to make copies. Despite incredibly effective, a virus’s “autonomy” is deterministic in its target environment. Its architectural complexity is relatively simple based on the previously discussed definitions—twofolded macromolecules with one level of scaling [2, 2].

Eukaryotic cells are generally capable of sensing, computation, and movement. Neutrophils [3, 3] (the most prevalent white blood cell in the human immune systems) demonstrate semiautonomy by sensing chemical signals and then moving toward damaged or inflamed tissue (chemotaxis), where they combat pathogens by engulfing foreign cells and releasing enzymes.39 A degree of this cellular semiautonomy stems from DNA being primarily confined to the nucleus in eukaryotic organisms, which enables the detection of foreign DNA and the subsequent deployment of enzymes to fight invaders.40

Unicellular protists similarly navigate with rudimentary sensory capabilities; green algae are able to follow sources of light (phototaxis) with simple eyespot organelles, while paramecia navigate across temperature gradients (thermotaxis) and via chemotaxis [3, 4].41 Compared to neutrophils, paramecia are generally considered more architecturally complex, possessing a wider array of specialized organelles (e.g., contractile vacuoles, micronuclei, cilia) and a structured cortex. Slime molds provide a final example of autonomy [3, 4]. Optimization techniques known as slime mold algorithm (SMA),42 along with marine predators algorithm (MPA), are used to detect and quantify damage to large-scale structures. In order to detect this damage, these optimization-based methods use vibration data to minimize the objective functions. The pairing of SMA with objective functions based on modified total modal assurance criterion (MTMAC) creates a tool that is accurate at damage identification.42 As a final example, the idea of designing and constructing synthetic networks for a particular function has been explored in prokaryotes by using transcriptional repressor systems to build an oscillating network within Escherichia coli.43 In this example, three gene repressor systems, each inhibiting expression of the next (with the third suppressing the first), form a sustained limit-cycle oscillation with period dependent on several factors.

In all of these natural examples, the requisite architectural precision across multiple length scales is less than needed to achieve functional goals in metamaterials or microelectronic devices. Further increases in autonomy can be achieved via coordination between groups of cells versus within a group of cells (Figure 4).44 Siphonophores are complex colonial animals, consisting of asexually produced bodies (zooids) that are functionally specialized for specific tasks, including feeding, swimming, and sexual reproduction. An example of a siphonophore is the Portuguese man o’ war, which is made up of specialized zooids.45 These zooids all work together to create what looks and functions like a single animal. The Portuguese man o’ war has four separate primary zooids: (1) pneumatophores, a carbon monoxide and air-filled protrusion that acts as a sail for locomotion; (2) gastrozooids, the feeding tentacles; (3) dactylozooids, prey capturing tentacles; and (4) gonozooids, for reproduction. The Portuguese man o’ war contracts muscles of the outer codon (a multilayered structure within the pneumatophore that consists of ectoderm, mesoglea, and endoderm) causing an increased pressure in the pneumatosaccus, resulting in elevation of the pneumatophore crest.45 This active locomotion is how the Portuguese man o’ war navigates to find food, and why we deem the complex collective coordination between zooids an example of A = 5 autonomy, but again with architectural precision that is less than N = 4 or 5. Furthermore, biology uses subunits that each have a degree of autonomy and arranges them such that interunit interactions self-reinforce to deliver an even high degree of autonomy.

Figure 4
figure 4

Autonomy through a hierarchy of collective coordination observed in the Portuguese man o’ war (Physalia physalis [P. physalis]) siphonophore. (a) A Portuguese man o’ war photographed in the ocean and (b) on the beach. (c) An illustration of the Portuguese man o’ war with zooid anatomy labeled. **There is terminological inconsistency within the literature regarding the anatomy of P. physalis, notably with respect to the tentacle-like structures that hang below the pneumatophore. Some researchers collectively refer to these as dactylozooids, whereas others distinguish the tentacle as a separate entity.45

Full size image

Overall, the framework and relative position of various material systems, organisms, and devices in Figure 3 provide a means to cross-fertilize among these communities. Equally important, however, is that it provides a means to help unify the AMS community. The fractionation of the AMS community can be shown by the results one gets when completing a Web of Science search. While many different types of materials with autonomous-like functions have emerged,46,47,48 a search at this moment reveals only ~50 examples on the topic of “autonomous materials.” The inclusion of “soft robotics,” “embodied intelligence,” and “smart systems,” in this search results in thousands of research articles, most of which include systems involving “autonomous materials,” or “adaptive materials.”49

Each of these materials’ subfields are pursuing multiple applications, including space50 and undersea exploration,21,49 efficient and multistable aerial vehicles,51,52,53 personal assistance for beginning and end of life care,54 physical and rehabilitation therapies, and many others.55,56,57,58,59 Solutions to these technological challenges will be addressed by complex systems that integrate material and composite innovations, not individual material innovations. The C–A space provides a construct on how to both classify and evaluate autonomous materials and adaptive materials in a more expansive context. The classification scheme can hopefully spur these related communities to adopt a unified language and more easily quantify the current state of the art and their objectives. For example, within the C–A framework, one could begin to imagine a thermodynamic framework for autonomy, where it can be thought of as a property of an open thermodynamic system, where energy and information are harvested from the environment, transformed, and processed to determine a response that manifests in an action upon the environment. Low-level automation reflects a closed thermodynamic process, such as thermal expansion of a metal bar. As such, the C–A construct does not aim to provide a precise definition for autonomy in materials, but a relative way to assess, measure, and design degrees of autonomy in materials, composites, and material systems, and thereby identify challenges and strategies toward assessing progress in this field.

For example, a common challenge that unites all these diverse material communities is manufacturing—not simply manufacturing of the material, but the designed integration of the various components into a functional system. Using biology as a guide for manufacturing scaling relationships, we can say that the smallest unit that could be autonomous (e.g., a eukaryotic cell) has Lcell ~ 10 μm, and is built from proteins of size Lprotein ~ 0.01 μm. The majority of biological cells, however, are not autonomous at a level of human interaction. Instead, the tissue level, L ~ 1 mm scale, is more demonstrative of utility. At this scale, animals have clusters of nerves, blood, muscles, and mechanoreceptors, which combine at a scale of L ~ 10 cm. Using our biological manufacturing analogy, we would require a single manufacturing process with precision from 10 nm to 10 cm—6 orders of magnitude. No singular manufacturing process exists today to accomplish this task. As an example, electronics rely on lithography at the nanoscale, and printing and pick-n-place techniques from the micron to macroscale. Such design necessitates compartmentalizing of function into small units that are combined into larger systems. Designing and interfacing a more diverse set of manufacturing processes are, thus, critical to use the emerging and diverse set of autonomous material, composite, and material systems available.

Conclusion

“Chance favors only the prepared mind,” said Louis Pasteur; we hope that the framework we have presented allows the reader to better understand where our technology and capabilities presently lie so they can better select research products that push the field forward efficiently. The combination of smart systems, soft robotics, additive manufacturing, autonomous materials, and many other materials science disciplines will create the next generation of autonomous material systems. In a way, the C-axis of Figure 3 is the role materials science will play. The A-axis, then, is more the role of traditional robotics, machine learning, and sensor fusion, to guide the design and use of these AMSs. The interface between these communities, the curvilinear relationship, defines the future of autonomous materials. All of these communities will need to work together and possibly merge their disciplines, to realize the vision and exciting potential of autonomous material systems.