As our discussion so far suggests, designing systems that are more fully plant-like accords with soft robotics and a broader embodied perspective. One reason for this emphasis on soft bodies and smart morphology is that plant intelligence lacks the sort of organization and architecture modeled by symbolic, language like, or more explicitly deliberative architectures (Newell and Simon, 1976; Pylyshyn, 1984). Research in plant intelligence, for instance, indicates the distributed nature of control, where adaptive responsibility is shared between local responses, internal long-distance signaling mechanisms, the material properties of organs, and the dynamics of body-environment interactions (recalling the “principle of ecological balance,” Pfeifer and Scheier, 1999). More fully plant-like robots will exploit similar means for adaptive behavior through principles of the smart embodiment such as sensorimotor coupling with soft bodies, and decentralized control (Linson and Calvo, 2020; Calvo and Trewavas, 2021).

Enactivism stresses the role of an adaptive embodiment for intelligence and behavior and, thus, coheres with other soft and embodied perspectives, but additionally centers the role of “autonomy” and “adaptivity” (Froese and Ziemke, 2009), based on the conviction of strong continuity between life and mind (Varela et al., 1991/2017; Thompson, 2007). As with all organisms, such adaptive autonomy plausibly plays an important role in plant intelligence and behavior, as we shall see. We contend, therefore, that an enactive perspective on plant bioinspiration serves as a heuristic for drawing attention to the contribution of soft materials and morphology to plant intelligence as well as ask us to consider the role of adaptive autonomy. On the flipside, plant bioinspiration offers enactivism a case study for exploring the possibility of engineering more fully agential systems.

Enactivism refers to a family of theories that share historical roots and central tenets, but either diverge in significant ways or otherwise stress different aspects of cognition (Ward et al., 2017). For present purposes, the important aspect of enactivism, as we intend it, is that it emphasizes not only: (1) agent-environment coupling and the importance of bodily morphology for intelligent action, in keeping with other embodied approaches, but also the role of (2) autonomy (Varela et al., 1991/2017; Thompson, 2007). Autonomy is here defined as a kind of recursive process of production, in which a system is constituted by a network of processes that recursively depend on each other to generate the processes themselves, and constitute the system as a unity individuated from its environment. To quote Thompson, “an autonomous system is a self-determining system, as distinguished from a system determined from the outside or a heteronomous system” (Thompson, 2007, p. 37). For brevity, we focus on basic metabolic or autopoietic autonomy (Ruiz-Mirazo and Moreno, 2004), i.e., the capacity of a system to reproduce and maintain itself physically. However, enactivists often recognize other forms of autonomy (e.g., neurological, immunological, sensorimotor). Robotics and plant research may benefit from attending to these other forms of autonomy, which find a parallel in the plant kingdom. For instance, in addition to “phytoneural” (Calvo et al., 2017) and sensorimotor behavior, we would do well to examine research in plant immunology (Jones and Dangl, 2006; Li et al., 2020).

Complementing the basic idea of autonomous constitution is the idea that a truly autonomous system is “precarious” — it must actively work to ensure its continued existence. This links autonomy with adaptivity (Di Paolo, 2005; see also De Jesus, 2018). Contemporary enactivism places great emphasis on adaptivity—the capacity of the system to actively modify its relationship to the environment in a manner that facilitates its persistence (Di Paolo, 2005; Di Paolo and Thompson, 2014). Marrying autonomy with adaptivity, we get “adaptive autonomy” (Barandiaran, 2002, 2004; Barandiaran and Moreno, 2008; Thompson and Stapleton, 2009), i.e., the notion of a system that regulates its interactions with the world, thereby managing its conditions for viability (the conditions under which it persists as a distinct system). This creates a kind of interdependence between the interaction of a system and its environment and the persistence of that system; actions of a system and its constitution are intertwined.

Although autonomy for enactivists is, strictly speaking, an all or nothing phenomenon—with living systems as the only known instance of an unequivocally autonomous system—we can still think of robots as more or less engineered in relation to enactivist principles. This is because the design of such systems may more or less emphasize autonomy as an important ideal and guiding heuristic (in addition to the importance of morphology and body-environment coupling, shared with other embodied perspectives). Three considerations are worth bearing in mind here. The first is that even embodied robots that are typically thought of as autonomous because they can operate independently for certain durations do not necessarily meet all the requirements for full autonomy in the enactivist sense (Froese and Ziemke, 2009). The second is that even if one falls short of designing a fully autonomous system, autonomy can still function as a model criterion. Finally, a focus on autonomy will produce different results depending on whether research of an individual is animal- or plant-inspired; autonomous growbots may meet different criteria from “locobots” because of their architectural and morphological idiosyncrasies (for a related discussion on the specificity of “organismoid embodiment,” see Vernon, 2010).

Autonomy (as well as adaptivity) is argued to be a crucial determiner of genuine agency. We can unpack agency, from an enactivist perspective, in terms of an autonomous organization that adaptively manages its coupling to the environment and, thus, contributes to sustaining itself (Barandiaran et al., 2009). A more exact definition of “basic autonomy” (which slightly diverges from the traditional formulation in terms of autopoiesis) is provided by Ruiz-Mirazo and Moreno: “the capacity of a system to manage the flow of matter and energy through it, so that it can, at the same time, regulate, modify, and control: (i) internal self-constructive processes and (ii) processes of exchange with the environment. Thus, the system must be able to generate and regenerate all the constraints—including part of its boundary conditions—that define it as such, together with its own particular way of interacting with the environment” (Ruiz-Mirazo and Moreno, 2004, p. 240. Original emphasis).

An interesting consequence of the enactivist perspective is that relatively “simple” organisms (including all the higher plants) exhibit genuine agency, whereas robots capable of completing complex information-processing tasks typically do not. Even embodied robots with tight perception-action coupling, though perhaps exhibiting agent-like behavior, do not possess intrinsic agency unless such coupling arises from fulfilling one of that requirements of the system for continued survival (Barandiaran et al., 2009; Stapleton, 2016). In short, enactivism provides relevant perspectives for robotic design concerned with the genuine agency, rooted in the biological processes that are not exclusive to animals. Again, it is important to stress the contrast between the concept of autonomy outlined here and one invoked in many areas of robotics (for discussion on the varied of “autonomy” in robotics, see Smithers, 1997). For example, an “autonomous system” often refers to a robot with the mere capacity to self-manage for some extended period (arbitrarily benchmarked) without human supervision.

Take growth in plant-inspired robotics as a case study (Del Dottore et al., 2018). Enactivism provides the tools to assess the limitations of existing growbots, given its emphasis on homeostatic autonomy (Froese and Ziemke, 2009). Existing robots are capable of growth via root-like appendages, providing novel forms of movement (Sadeghi et al., 2017). Recent examples of effective robotic growth include soft pneumatic robots that achieve directed growth through the pressurization of an inverted thin-walled vessel coupled with controlled asymmetric lengthening, displaying a remarkable ability to move through constrained spaces (Hawkes et al., 2017). However, all the existing forms of plant-inspired roots depend on a prefixed store of energy and matter. Recent pressure-driven robots depend on stored material within a “base station” — a fixed spool of polyethylene tubing provides the material for pressure-driven eversion, i.e., turning inside out—and externally provided source of liquid or air pressure (Hawkes et al., 2017). From an enactivist perspective, a more genuinely autonomous robot actively seeks out and metabolizes all the material for growth in its environment and uses this process to aid its persistence as an individuated system. There are existing robots with artificial digestive systems that seek out energy sources, process them, and egest waste (Melhuish et al., 2006; Ieropoulos et al., 2010). Ecobot-II and -III convert biomass into energy using onboard microbial fuel cells with oxygen cathodes. However, these robots still require an external source to supply key materials.

Moving forward, more truly autonomous growbots—that are plant like in not only their material composition and morphology, but in their adaptive autonomy—will not only self-direct and self-manage in the manner of existing so-called “autonomous” robots (free from direct human management), but will actively seek out the requirements for fulfilling the conditions of their own persistence. This may also be relevant in examining limitations in the amount of growth and degree of control possible in existing growbots compared with plants, given their dependence on an external source (Hawkes et al., 2017).

In addition to any generic benefits afforded by an enactivist perspective—for example, see Smithers (1997) on the role of autonomy for navigating unpredictable environments and Lowe and Kiryazov (2014) on the role of autonomy for cognitive-affective processes—designing robots that are more fully plant like in their material composition, morphology, and autonomous control promises some particular advantages for soft robotics. Obviously, autonomous plant-like robots allow us to test the possibilities of what forms intelligence might assume by taking inspiration from a non-zoological branch on the tree of life. They may also allow us to better test existing theories within plant cognitive science/neurobiology, adopting a “synthetic methodology,” i.e., understanding a phenomenon by building physical systems that simulate aspects of the phenomenon (Pfeifer et al., 2008).

Robots exhibiting more plant-like bodies as well as stronger autonomy also promise practical benefits. These benefits would build upon (but potentially surpass) the advantages of existing plant-inspired robots. This includes the fact that plants display high levels of fault tolerance, with catastrophic damage less likely given the absence of system-critical centralized organs as well as the ability to acquire energy and material in proportion to the demands of growth (a function of their adaptive autonomy). In other words, plants have extensive redundancy built-in to their basic organization. Such a strategy can minimize existential risk (no single root is essential), but it also provides novel ways to reach new locations that have advantages over locomotion (e.g., navigating a hard surface by growing through small cracks). There is also the broad principle that engineering an intelligent system via many “not-so-smart” parts—via principles of swarm intelligence—is often optimal given the cost/risk involved. This is especially relevant, for instance, when designing expensive systems for space exploration (Mehling et al., 2006; Wooten and Walker, 2015; Gallentine et al., 2020).

Designing robots with reference to a more complete suite of plant features including stronger autonomy—thus, has the potential to produce relatively low-cost systems which can be deployed with little configuration and that will actively build themselves while exploring and adapting to their environment with little or no external management. This could have serious implications for space exploration, rescue operations, and medical procedures (see also Blumenschein et al., 2020). Plant-inspired robotics, thus, corroborates the dictum that embodied perspectives both offer theoretical insight into the principles of biological intelligence and are of practical value in the design of adaptive systems (Pfeifer et al., 2008).

To summarize, we suggest there are at least four (overlapping) reasons to consider the design of more holistically plant-inspired robots with strong autonomy as a guiding heuristic:

Of course, soft roboticists are already sensitive to some of these considerations, some of the time. As such, recognizing the possibility of more holistically plant-like robots partially serves as a tool to deepen and develop existing trends. Equally, if the preceding discussion is correct, too little attention has been paid to the possibility of genuinely autonomous systems, and the use of strong autonomy as a heuristic to develop more fully plant like (and other autonomous) robots, e.g., robots with more genuinely plant-like growth properties.

Our discussion has explored a two-way relationship between enactivism and the design of more plant-like robots. Enactivism helps us attend to the possibility of looking to plants and other non-zoological sources of inspiration, emphasizing the coupling of adaptive morphology with strong autonomy across the tree of life, while the practical success of plant-inspired robots reinforces a postcognitivist perspective (Heras-Escribano, 2019) on the diverse forms intelligence can take (Linson and Calvo, 2020).