A Biohybrid Self-Dispersing Miniature Machine Using Wild Oat Fruit Awns for Reforestation and Precision Agriculture

Abstract

Advances in bioinspired and biohybrid robotics are enabling the creation of multifunctional systems able to explore complex unstructured environments. Inspired by Avena fruits, a biohybrid miniaturized autonomous machine (HybriBot) composed of a biomimetic biodegradable capsule as cargo delivery system and natural humidity-driven sister awns as biological motors is reported. Microcomputed tomography, molding via two-photon polymerization and casting of natural awns into biodegradable materials is employed to fabricate multiple HybriBots capable of exploring various soil and navigating soil irregularities, such as holes and cracks. These machines replicate the dispersal movements and biomechanical performances of natural fruits, achieving comparable capsule drag forces up to ≈0.38 N and awns torque up to ≈100 mN mm⁻¹. They are functionalized with fertilizer and are successfully utilized to germinate selected diaspores. HybriBots function as self-dispersed systems with applications in reforestation and precision agriculture.

1 Introduction

By taking inspiration from living organisms, scientists have prototyped different machines capable of adapting to complex 3D surfaces.^[1] These machines are traditionally fabricated with soft or rigid synthetic materials. However, the past 10 years have seen a rapid expansion of the transdisciplinary field of biohybrid robotics, where synthetic materials are combined with biological components.^[2] Several biohybrid systems have been developed by using microorganisms (e.g., bacteria, fungi, algae), tissues (e.g., cardiomyocyte, skeletal muscles, invertebrate tissues), or the entire animal body as cyborgs (e.g., insects, jellyfish) (e.g., a recent overview is reported).^[2] When miniaturized, bioinspired, and/or biohybrid machines can allow to precisely access confined spaces in real-world environments.^[3] Recently, miniaturized machines inspired by plants have shown a growing interest for applications in soft- and microrobotics,^[4] space exploration and manipulation,^[5] and environmental protection.^[6]

Within the plant kingdom, small-scale plant fruits and seeds have evolved several dispersal mechanisms to improve germination and complete their life cycle.^[7] Although they lack an active metabolism, self-burying fruits and seeds can create a motion by reacting to changes in the environmental humidity to efficiently dig into the soil before germination.^{[7, 8]} Especially, wild oats like Avena fruits are one of the most widely distributed grass species due to their very effective seed dispersal mechanisms by combined torsion/bending and repetitive snapping movements of two interacting sister awns which allows them to colonize several habitats in Asia, Europe, and Africa.^[8] Although some examples of self-dispersing fruit-inspired machines have been reported (e.g., by mimicking the Geraniaceae family),^{[6, 9]} artificial systems that mimic the unique dispersal mechanism of Avena fruits are still to be reported. Moreover, to the best of our knowledge, biohybrid systems with embedded fruit body parts have not been exploited so far, although they can lead to a new category of biohybrid machines where entire plant body parts, such as the dead tissues of fruits, could be used as biological actuators to autonomously explore and interact with complex real-world unstructured scenarios.

Here, we propose a new class of miniaturized biohybrid machines inspired by Avena fruits (hereafter called Hybribots) embedding natural diaspores into a biomimetic capsule-like cargo delivery system by using fruit interacting sister awns as motors. We designed and prototyped high-resolution miniaturized biohybrid machines able to explore and interact with soil roughness and crevices by mimicking and integrating natural Avena fruit structures. To reach this goal, we combined a 3D reconstruction technique via X-ray microcomputed tomography (µCT) with a micromanufacturing process via two-photon lithography, molding and casting of biodegradable flour-based materials. From a biomechanical viewpoint, we characterized capsule stiffness and drag force, as well as awn contact force and related torque, by also comparing natural fruits and biohybrid machines, and we introduced a mathematical model describing awns dynamics (including flexibility and contacts). Finally, we functionalized the biohybrid machines with fertilizer to carry different payloads and facilitate the germination of selected diaspores, as seeds and fruits. Our resulting system can inspire a new class of biohybrid machines and paves the way for sustainable delivery systems with potential applications, such as reforestation^[6] and precision agriculture.^[6]

2 Results and Discussion

2.1 Inspired by Avena Fruits, a New Biohybrid Machine

The fruits of the wild oat Avena sterilis L. were selected as a biological model for this study due to their unique seed dispersal mechanism (Figure 1a; and Figure S1, Supporting Information). The fruit body consists of a passive capsule and two hygroscopically responsive awns (Figure S1c–g, Supporting Information). We carried out a morphometric analysis of the capsule and sister awns of natural Avena fruits (Table S1, Supporting Information). Starting from the capsule, we noted that Avena sterilis L. fruits are fully covered by directionally aligned hairs, which become thinner and sharper approaching the tip of the capsule reaching a tip diameter of 42 ± 5 µm (Figure S1d,e, Supporting Information). Instead, the two sister awns are made of two functional parts: a twisting base beam and a bending tail (Figure S1c, Supporting Information).^[10] More specifically, the awns are composed of dead tissues and, when passing from dry to wet conditions, the base beam's cells tilt in their alignment around the awn's long axis (Figure S1f,g; and Movie S1, Supporting Information), whereas tail bending is driven by the structured connection at the knee point joining base beam and tail. The resulting movement causes entanglement/contact between the sister awns and with the surrounding environment, thus finally enabling fruit locomotion also through multiple jumps on the ground and penetration into the soil.^[8]

Figure 1

From natural to biohybrid machines inspired by Avena fruits. a) Natural fruits of wild oat Avena sterilis L. Scale bar, 1 cm. b) Sketch of the biohybrid approach used in this work. c) Biohybrid machines that mimic Avena fruits (HybriBots). d,e) Serial snapshots at different time frames (time: 0, 48, 108, and 240 seconds) of d) natural fruits and e) biohybrid machines placed on artificial clay soil, while entering soil crevices.

Therefore, we investigated the internal structure of the awns by differentiating between the twisting base beam and bending tail (Figures S2 and S3, Supporting Information). The base beam has two furrows which separate three wings (two adaxial and one abaxial) bonded by a connective tissue. Its internal structure is formed by cells with thick secondary cell walls, with symmetric or asymmetric deposition, that differentiate for their cellulose microfibril angle (MFA), as described previously.^[8] With the polychromatic dye, toluidine blue, we could identify sclerenchyma-like cells, stained in green or blue, which confer important mechanical functions to the structure besides actuation (Figure S2a, Supporting Information). The epidermal and subepidermal cell layers and the wing tissue cells show a similar dark blue staining (Figure S2b,c, Supporting Information), while the connective tissue a dark green one (Figure S2d, Supporting Information), suggesting slight differences in cell wall composition, besides in MFA.^[11] This structure is constant along the twisting base, with structural changes visible in the region connecting the base with the bending tail. In the tail region, we observed lignified cells with wider cell lumen with no thick secondary cell wall, and thus the loss of sclerenchyma-like-cells properties and, therefore, their mechanical and actuating behavior (Figure S3a–c, Supporting Information). The furrows contained many chlorenchyma cells (Figure S3d, Supporting Information), a characteristic also found in the awns of other grasses, such as barley and wheat.^{[7, 12]} As observed by SEM imaging, directionally aligned silica hairs are located on the awn surface along the base beam and tail facilitating the awns entanglement, resulting in energy accumulation and subsequent release of the interlocked awns (Figure S4 and Movies S4 and S5, Supporting Information).

By focusing on the hygroscopic material per se, namely on its response at the basic material level, it is possible to engineer artificial systems demonstrating quite an extended repertoire of tasks/movements.^{[9, 13]} For instance, the micro/nanoscale fabrication of hygroscopic materials inspired by plants has garnered significant interest in soft robotics due to their potential applications in sensors,^[14] actuators,^{[6, 9]} sunlight powering,^[13] and architecture.^[15] In this context, the fabrication of hygroscopic materials mimicking the complex architecture and energy storage release mechanism of the sister awns of wild oats can lead to new designs for jumping robots^[8] triggered by environmental stimuli using multistable architected materials,^[16] or sustainable 3D or 4D-printed soft actuators.^[17] Such a research perspective is complementary to that one introduced through our machines, where the hygroresponsive behavior is integrated by deliberately preserving the whole embodiment of the natural awns, consistently with the pursed biohybrid approach. Indeed, producing biomimetic responsive materials that are durable, stable over time, and cost-effective on a large scale presents significant challenges. To overcome these issues, we used the natural dead tissues of fruits, such as wild oat awns, as biological actuators. We proposed a novel biohybrid approach to prototype autonomous self-dispersing fruit-like machines embedding two natural moving sister awns into a passive biomimetic microfabricated capsule (Figure 1b). In this biohybrid system, the awns induce the mobility required for the biohybrid machine to move on and into the soil, while the artificial capsule can be functionalized to promote targeted payload delivery into the soil (Figure 1b). Taking into consideration the Avena fruit morphology and material properties and our biohybrid approach, we designed and fabricated Avena-inspired autonomous self-dispersing machines (Figure 1c). The newly developed biohybrid machine is capable of autonomously moving along both flat surfaces and rough terrains in response to humidity changes, similar to natural fruits (Figures S5 and S6; Movies S2 and S3, Supporting Information). While the awns allow the motion of the fruits, the capsule has the fundamental role of anchoring the fruits into the soil crevices to facilitate its penetration.^{[7, 8]} Thus, various textured loam or clay soil were selected to test the capability of the biohybrid machines to move along and into the soil (Figure S7a, Supporting Information). Similar to the natural fruits of Avena sterilis L. (Figure 1d; and Figure S7d,f; Movies S3 and S4, Supporting Information), the hybrid machine can autonomously perform soil exploration and falls into soil crevices, such as holes and/or cracks (Figure 1e; and Figure S7g; Movie S5, Supporting Information). Moreover, the sister awns of the biohybrid machine can interlock each other, thus accumulating stress, whose release makes the machine jump and change its movement direction (Movie S5, Supporting Information), showing a behavior similar to wild oat fruits.^[8] Taking into account the natural fruit morphology and biomechanics, the biohybrid machines are made with biodegradable and edible materials, by using a scalable and low-cost micromanufacturing process. Figure S8 summarizes all steps required to design and prototype the bioinspired biohybrid machines. By µCT, we obtained a detailed 3D reconstruction of the capsule of Avena sterilis L. (Figure 2a,b; and Movie S6, Supporting Information) which was then used to design two complementary miniaturized molds (Figure 2c) for the artificial capsules and awns joint connections. The molds were directly 3D printed using two-photon lithography, allowing high resolution and reproducibility even at the tip level (Figure 2d; and Figures S9 and S10, Supporting Information). After the casting and coating processes (Figure S11, Supporting Information), we obtained miniaturized biohybrid machines, which consist of a biodegradable and edible capsule made with flour-based material and coated with natural hairs of Avena sterilis L., and two humidity-driven sister awns of Avena sterilis L. connected to the capsule (Figure 2e; and Figure S12, Supporting Information). The flexible natural hairs were well adhered to the capsule and directionally aligned toward the capsule base, as shown in Movie S7 (Supporting Information). We carried out a morphometric analysis of the artificial capsules and sister awns of biohybrid machines (Table S2, Supporting Information). The printed molds can be reused multiple times, allowing the prototyping of multiple autonomous self-dispersing Avena-like biohybrid machines at high resolution with low-cost materials (Movie S8, Supporting Information). In view of production scale-up, production costs can be minimized by either reusing molds created through two-photon lithography multiple times or by investigating simplified capsule geometries (e.g., by removing finer-level features and by applying standard computational geometry techniques) that can be manufactured using conventional 3D printing methods^[18] and possibly support robotic automatization for the hybrid machine assembly. Furthermore, the mass release of biohybrid machines in real-world scenarios can be envisioned using drones with compatible containers.^{[6, 19]} In addition, our biohybrid machines provide a competitive asset from the viewpoint of life cycle assessment, as they are sustainable, biodegradable and edible, without disturbing natural ecosystems and biodiversity.

Figure 2

Design and micromanufacturing of HybriBots. a,b) 3D reconstruction via µCT and Avizo software of natural capsule of Avena sterilis a) with and b) without hairs. Scale bars, 0.5 cm. c) 3D computer-aided design of biomimetic molds. Scale bar, 0.5 cm. D) 3D printed molds fabricated via two-photon lithography. Scale bar, 0.5 cm. e) Resulting biohybrid machines. Inset: details of hairy capsule (top) and joint of the capsule-awn (bottom). For a detailed micromanufacturing process and outputs please refer to Figures S8–S12 (Supporting Information). Scale bars, 0.5 mm.

Sustainable robotics is a promising research field where 3D/4D manufacturing techniques are combined with biodegradable materials to deploy robotic technologies by minimizing environmental impact.^[17] In future works, our biohybrid robots can also be made with different biodegradable materials with varying times of degradation or sensing capabilities (e.g., different natural- and synthetic-based biodegradable polymers,^[20] such as alginate-based hydrogels, PEGDA)^[17] via sustainable 3D or 4D printing for applications, such as targeted cargo delivery into soils to enhance crop production,^[21] or monitoring soil quality parameters.^{[14, 22]}

2.2 Biomechanical Characterizations and Modeling of Natural and Biohybrid Systems

We performed a biomechanical analysis of capsule and awns parts in both natural fruits and biohybrid machines (Figure 3).

Figure 3

Biomechanical characterization and modeling of natural fruits and HybriBots. a,b) Results of capsule drag force tests of natural and biohybrid fruit a) without and b) with rotation (at 1 rpm) using different soils as substrates, including artificial and natural soil. The artificial soil consists of glass beads of 1 mm. The natural soil consists of loam, sand, and clay. c) Natural and biohybrid fruit in dry and wet conditions after using a humidifier. d) Contact forces exerted by awns of natural and biohybrid fruits on a load cell, and e) associated torque. Three force levels were considered, namely 1/3 (F_1/3), 2/3 (F_2/3), and 3/3 (F_max) of the corresponding maximum value. f) Illustrative results of the mathematical model that describes the awn dynamics, based on experimentally recorded torsion and accounting for awns flexibility and mutual contact. A two-contact test-case is considered: for both contacts, characteristic awns configurations (left) and inter-awns contact forces (right) are shown. The relatively low force variability versus the only free model parameter (namely k in the force plots) allows the model to be predictive, and the obtained results are in good agreement with the load cell experiments.

First, we characterized the stiffness and drag forces of natural and artificial capsules (Figure 3a,b; and Figures S13–S16, Supporting Information). To investigate the stiffness of both natural and artificial capsules, we used a nanoindentation system with a miniaturized probe (Figure S13, Supporting Information), similar to methods used to characterize capsule material in Geraniaceae,^[9] or awn materials in Poaceae.^[7] Natural capsule materials of Avena sterilis show a local average storage modulus (E) of 7.7 ± 2.6 GPa similar to Avena-like capsules made with flour (E = 6.4 ± 3.3 GPa), demonstrating that biodegradable and low-cost flour-based material has a suitable stiffness to be selected as an optimal casting material in Avena-like biohybrid machines. To investigate the forces required for the natural fruits and biohybrid machines to dig into the soil, we measured the drag forces of natural and artificial capsules without and with rotation during their vertical penetration in different types of granular substrates, including artificial and natural dry soils with different densities and particle sizes (Figure 3a,b; and Figures S14 and S15, Supporting Information). In experimental tests without rotation, the drag forces in natural capsules ranged from lower force values of 0.06 ± 0.01 N in artificial glass beads to higher values of 0.37 ± 0.07 N in natural sandy soil, similar to artificial capsules where the drag forces increased from 0.08 ± 0.01 N in artificial glass beads to 0.38 ± 0.12 N in natural sandy soil (Figure 3a). Understanding the relationship between particle size, density, and drag forces is crucial to develop hybrid robots able to work on different unstructured terrains. Soil particles with smaller sizes and higher densities (e.g., sand, clay, loam) require higher drag forces to move through the substrate, while soil particles with larger sizes and lower densities (e.g., glass beads) require lower drag forces. Our biohybrid system showed a robust behavior (with comparable behavior respect to natural Avena fruit, Figure 4a,b) with all the considered, relevant soil types. Among natural soils, we obtained lower drag forces with loam soil (Figure 4a,b). For clarity, the main soil parameters and the related drag forces are summarized in Table S4 (Supporting Information). To test the drag reduction of natural and artificial capsules through rotational motion, we measured the drag forces of natural and artificial capsules in the different substrates while rotating at 1 rpm (Figure 3b). In both natural and artificial capsules, we observed a decrease of the drag forces in all substrates (Figure 3b), demonstrating that self-dispersing Avena fruits can reduce their granular drag with rotational motion, similar to mechanisms observed in Geraniaceae family.^[23] The rotational speed plays a key role in reducing the drag force between the capsule tip and the soil, facilitating drilling.^[23] Moreover, we observed a drag reduction with rotation in natural and artificial capsules not only in dry environments, but also in wet soil (Figure S16a,b, Supporting Information). A typical force-displacement curve obtained during drag force experiments is shown in Figure S16c (Supporting Information), where the forces increased while moving the capsules into substrates for a displacement of 10 mm after which the capsules were pulled away from the substrates. Overall, drag forces observed in natural and artificial capsules of Avena are about ten times higher than the forces observed in previously reported self-dispersing fruits such as Erodium and Pelargonium fruits,^{[9, 23]} likely due to their significant differences in capsule dimensions, mechanics, and shapes.

Figure 4

Functionalized HybriBots delivering tomato seeds. a) Functionalized biohybrid machine with biochar to promote germination of favorable seeds. Scale bar, 1 cm. b,c) Germination of b) one or c) two tomato seeds from functionalized biohybrid machines. d,e) The germination process of functionalized biohybrid machines with embedded tomato seeds from day 1 to day 27 in the growth chamber. f) Biohybrid machine removed from the soil at day 35 with the visible growth of tomato roots out of the artificial capsule. Scale bar, 2 cm. Inset: microscopy details of root going out from flour-based capsule material of a biohybrid machine. Scale bar, 0.4 cm.

Then, we experimentally characterized awns kinematics and kinetics. Specifically, upon hydration, base beam torsion, and tail bending were recorded for both natural fruits and biohybrid machines (Figure 3c; and Movie S9, Supporting Information), with a slightly different torsional speed between the two base beams (Figure S17; and Movie S10; Supporting Information) that is consistent with previous observations (19). From dry to wet states, we obtained 6 and 5 rotations for awn1 and awn2, respectively, in about 30 min (Figure S17; and Movie S10, Supporting Information), associated with maximum rotational speeds of 9.2° and 6.9° s⁻¹ (Figure S17d–e, Supporting Information), thus higher than for other seed-inspired robots reported in literature (i.e., about 3.7° s⁻¹).^[6] Moreover, we tracked the capsule tip of hybrid machines moving on unstructured terrains (Figure S18a,b, Supporting Information), obtaining average cumulative displacement Δs of 87.6 ± 26.6 mm and an average speed of 21.9 ± 14 mm per cycle, with a mean peak of 41 ± 15.3 mm per cycle at the 4th RH cycle (Figure S18c–f, Supporting Information). Our biohybrid system is faster than other seed-inspired robots, featuring a mean speed of 0.93 ± 0.90 mm per cycle with a peak of about 6 mm per cycle.^[9] In addition, for both natural fruits and biohybrid machines, the tails exerted a contact force (on a load cell) on the order of 1 mN (Figure 3d; and Figures S19 and S20, Supporting Information), resulting in a maximum torque on the order of 100 mN mm⁻¹ (Figure 3e; and Figure S21, Supporting Information). The maximum average torque exerted by awns in natural Avena fruit was 81.9 ± 29 mN mm⁻¹, which is comparable to forces obtained with our biohybrid system, for which the max torque was 100.4 ± 16.6 mN mm⁻¹ (Figure 3e). We observe that the measured torque is about five times higher than that one previously recorded for other species, such as Erodium and Pelargonium.^{[9, 23]} Considering that torque is essential to increase the chance of a successful penetration into the soil (15), these results highlight the potential of the proposed Avena-based biohybrid machines for effectively interacting with real-world unstructured environments and achieving the desired functions.

Furthermore, in order to quantitatively investigate awns dynamics, we used the recorded base beam torsional speeds as input to a physics-based mathematical model accounting for awns flexibility and mutual contact (Figure 3f; and Movie S12, Supporting Information). The developed biomechanical model predicts an inter-awn contact force on the order of 1 mN, in good agreement with the aforementioned load-cell experiments (by also considering that inter-awn contact is more compliant, and generally more localized, than that one between a single awn and the load cell rigid plane). A simplified rigid-kinematics, contactless model can be used to preliminarily estimate the number of contacts (Movie S13, Supporting Information), as functional to more elaborate biohybrid machine characterization/design based on the full biomechanical model. Specifically, by adjusting the pose at which natural sister awns are inserted into the biomimetic capsule, it is possible to anticipate the number of contacts between the sister awns (based on their characterization), thus modulating, to some extent (and up to the intrinsic environmental variability) the system's “programmed motion.” Finally, the load-carrying capacity of the biohybrid machine was around 3.0 ÷ 3.5 gf, as shown by experiments in good agreement with model-based estimates using the aforementioned contact force and torque (Figure S22, Supporting Information). Such a load-carrying capacity (which could be used tower displace even larger, nonbiomimetic containers beyond the present scope), significantly oversteps the force needed to lift the (60 mg) Avena capsule. Arguably, this could be due to the fact that the awns actuation capacity was evolutionarily set also/mostly based on the requirements for them to interact with rough terrains, typically colonized by grasses (in particular by engaging with soil cracks during penetration) or between each other (to release enough elastic energy, upon mutual disengagement, for on-ground movement).

We finally remark that several key elements (including system architecture, biomechanical properties, and fine-scale components, such as the directional hairs on the capsule surface) play a role in the locomotion and burrowing functionalities of the system. Furthermore, factors like torque, rotational speed, and the number of contacts among interacting sister awns are key contributors to the system's operation. For completeness, an overview of the main features of Hybribot compared to other seed-inspired robots^{[6, 9]} is reported in Table S3 (Supporting Information). Overall, the proposed biohybrid system has comparable design, biomechanics, and kinematics as for natural wild oats, with promising implications for self-dispersing autonomous machines for exploration of complex multienvironment.

2.3 Biohybrid Machines for Cargo Delivery

Wild oats can survive in a wide range of environments and under different infestation levels.^[24] In light of their very effective fruit dispersal mechanisms, wild oats are reported to disturb agriculture by causing significant yield reductions and causing millions of annual revenue losses.^{[8, 24]} Here, we exploit the same effective dispersal mechanism in order to prototype functionalized biohybrid machines embedding different cargos to release selected seeds into the soil (Figure 4; and Movie S14, Supporting Information). The process to functionalize and integrate different seeds or fruits in the biohybrid machines is reported in Figure S23, Supporting Information. As shown in Figure 4a, we functionalized the artificial flour-made capsule with fertilizer, such as biochar, and embedded two seeds or fruits of different species into the capsule (e.g., tomato seeds or chicory fruits). After 18–20 days, 50% of the tomato seeds germinated in a growth chamber (Figure 4b–e; and Movie S15, Supporting Information) and roots growing from the functionalized capsule were visible by optical microscope (Figure 4f). We noted that either one or both tomato seeds can germinate from the same capsule (Figure 4b,c; and Figure S24, Supporting Information). In addition, 33.3% of chicory fruits germinated within 15–23 days (Figures S25 and S24; Movie S16, Supporting Information). At the end of the experiments, we removed the HybriBots from the soil to check the capsule, which showed obvious signs of degradation after 27–35 days (Figure 4f; and Figures S24 and S26, Supporting Information). The stability of the system depends on the application scenarios and can be tuned from 1 day to several months according to the thickness of the ethyl/cellulose coating used to preserve the capsule^[22] (i.e., exposure to long time moisture conditions requires a thickener coating, while exposure to dry conditions requires a thinner coating, please refer to the Experimental Section for more information).

Our biohybrid machines have effective potential to be used to enrich soil and promote germination with potential applications in reforestation and precision agriculture. Reforestation tasks, for instance, could be addressed by using our biohybrid machines, based on a suitably identified payload depending on the following four key aspects: the quantity of trees planted; the careful selection of species; the survival rates of the planted trees; the amount of carbon stored.^[25] For example, in Australia, a successful reforestation method involves utilizing the bleeding heart shrub (Homalanthus novoguineensis) native to rainforests. The roots of this shrub enhance soil aeration, while its nutrient-rich leaves foster the growth of various plant species.^[25] In addition, seed balls have already obtained increasing interest in Africa,^[26] where favorable fertilizers are mixed with soil and seeds to promote germination in infertile soil.^[27] Our systems can be functionalized with various seeds/fruits (e.g., Homalanthus novoguineensis, known for its effectiveness as a kick-starter for reforestation projects in Australia) and agrochemicals, including fertilizers, herbicides, polymeric nanocarriers, nutrients, and fertilizers, facilitating sustainable targeted cargo delivery into soil.^[28] They can function as autonomous seed ball-like technology^[27] to promote germination, soil enrichment, and crop protection.^[28] In addition, they can be used to plant fruits or seeds to help bees by promoting germination of favorable flowers important for bees (i. e., Salcerella plants, Figure S27, Supporting Information).

3 Conclusions

This study proposed a novel biohybrid approach to develop autonomous bioinspired miniaturized machines capable to adapt and explore complex real-world unstructured environments such as soil. We designed and microfabricated biohybrid systems mimicking wild oat fruits by exploiting a biomimetic capsule design and the unique humidity-driven movement of their sister awns, which actuate the biohybrid machine. To this purpose, we combined different methods, including µCT, molding via two-photon lithography, and casting of flour-based material with embedded sister awns. Their contacts (both with each other and with the surrounding environment), possibly also leading to snapping and jumps, allow the biohybrid machine to effectively move on, and penetrate the soil. As shown in this study, awns integration in the biohybrid machine do not hamper their performance, as highlighted, e.g., by the measured forces/torques compared to natural fruits. Moreover, the developed mathematical model could be used to optimize biohybrid machines design, e.g., by setting awns relative pose (with respect to each other and to the capsule) in order to maximize the number of mutual contacts, to improve fruit movement. We incidentally observe that the model-based approach could be extended to develop even artificial systems inspired by Avena fruits, by codesigning payload containers (capsules) and driving systems (awns) through a multiobjective design optimization strategy. Considering the interplay between geometrical/material properties and environmental conditions, as well as specific features of the Avena fruit (such as, e.g., torsion-bending coupling in the awns), hygromorphic composite/anisotropic materials, shape profiling, and architected elements could be suitably integrated, also using advanced (multiscale) computational tools. However, the envisioned design problem, clearly beyond the present scope, should be based on the preliminary extraction of experimental data that are crucial for model calibration, and system scaling should be grounded on specific design targets and application environments. Regardless of the complementary research developments mentioned above, let us remark that this study offers cues for designing novel self-dispersing systems able to dig into different soil substrates. Finally, the proposed biohybrid machines were functionalized with fertilizer and seeds, and used as a tool for cargo delivery into the soil to promote the germination of selected diaspores (i.e., tomato seeds, chicory fruits). Based on the proposed approach, biohybrid machines featuring additional capsule materials and payloads can be investigated. The reported design and manufacturing methods can be extended also to other plants (i.e., seeds, fruits), and have direct implications for the future development of sustainable low-cost miniaturized machines for applications in reforestation and precision agriculture.

4 Experimental Section

Morphometric Analysis

The morphometric analysis was carried out on both natural fruits and biohybrid machines. Natural fruits of Avena sterilis were collected from a garden in Pontedera (Italy, Pisa). A morphometric analysis was used to define the geometrical details of natural and artificial fruit capsules (including diameters, length, and height) and sister awns (i.e., length of twisting base, length of tails, angles). For a detailed geometrical analysis of natural and biohybrid samples please refer to Tables S1 and S2 (Supporting Information), respectively. A digital optical microscope (Hirox KH-7700) was used to record pictures and videos of natural and artificial samples. Hairs were optionally removed with a scissor to facilitate the visualization of the capsule profile of the fruits. The analysis of fruit capsules and sister awns was performed using image processing software (ImageJ) and a caliber.

Scanning Electron Microscopy (SEM)

Natural and biohybrid samples were mounted onto aluminum stubs and sputter-coated with a 15 nm gold layer (Quorum Q150R ES, UK). Electron microscope images were obtained with an EVO LS10 scanning electron microscope (Zeiss, Germany).

Histology

Histological sections at the base and tail of the awn of Avena sterilis L. were obtained following the embedding procedure for the resin Technovit 7100. Sections 5 µm thick were cut with a manual microtome (Leica SM2010R). The sections were then stained with toluidine blue at 0.5% for 1.15 min and observed with the Nikon Eclipse Ni-U microscope.

X-Ray Microcomputed Tomography (µCT) Preparation and Scanning

To increase the contrast for µCT analyses, natural samples of Avena sterilis were infiltered with a contrasting agent 1% phosphotungstic acid (PTA) in 100% methanol, following the procedure.^[29] Before scanning, the samples were mounted and fixed on the sample holders (metal stubs) using a glue. All samples were scanned using a X-Ray Micro-Computed Tomography System (ProCon X-Ray, PXR GmbH, Germany). The scanning parameters include an acceleration voltage of 60 kV, source current of 100 µA, exposure time of 0.3 s, voxel pixel size of 8 µm, and a scan duration of about 30 min. The scanned data were reconstructed using the volumetric imaging software VG-Studio and exported to bmp format. The 2D sections were then postprocessed in Fiji/ImageJ to increase the contrast and reduce the file size before performing the 2D slices segmentation, surface generation and volume rendering.

From 2D Slices to 3D CAD Models

The 2D slices of natural samples of Avena sterilis generated from µCT scans were reconstructed using the software AVIZO (Thermo Scientific, USA; version 9.2). After segmentation, a volume rendering of 3D models was performed. The STL files of biomimetic capsules computed in AVIZO software were postprocessed in MeshLab to reduce the file size. Compressed STL files of the biomimetic capsule were imported into the 3D computer graphic software Blender, where the CAD models of two complementary molds of biomimetic capsule and two awns joint connections were designed and exported as STL files.

Microfabrication of Biohybrid Machines

The STL files of the two complementary molds were used to directly print the 3D negative molds using high resolution micromanufacturing technique via two-photon lithography (Nanoscribe Photonic Professional (GT) system, Nanoscribe GmbH). Dedicated Describe files were generated and imported in NanoWrite software. The two complementary 3D micromolds were printed in IPQ photoresist (Nanoscribe GmbH) on silicon wafer substrates using a 10X immersion objective. The IP-Q photoresist was poured onto the wafer which was then exposed to the laser beam (center wavelength of 780 nm, Toptica laser source), using 90% laser power and a scan speed of 50 mm s⁻¹ (base of the mold) to 100 mm s⁻¹ (rest of the mold). The samples were developed for 40 min in Propylene Glycol monomethyl ether acetate (PGMEA; Sigma-Aldrich), rinsed with isopropyl alcohol for 30 min and dried with air. After developing, the microprinted molds were UV-cured for 1 h. The resulting molds were used to prototype the biohybrid machines. The artificial capsule of the biohybrid machine was made with a flour-based dough. The dough was made by mixing 6 g of flour with 2 mL of water, shaped by the two complementary microprinted molds. Two natural sister awns were removed from the natural samples of Avena sterilis and mechanically inserted into the two closed complementary molds through the two joint connections-like pathways between capsule and awns. Then, the molds were removed and the biohybrid machines were obtained. To facilitate the detachment of the biohybrid machines from the molds, the molds were coated with butter before the casting with dough. After the dough dried, the capsule of the biohybrid machine was coated with 10% ethyl cellulose/ethanol solution twice times, following the procedure.^{[6, 22]} Ethyl cellulose is an eco-friendly biopolymer^[30] insoluble in water. It finds extensive use for the controlled release of fertilizers into soil.^[31] The thickness of the ethyl cellulose/ethanol coating can be adjusted by varying the concentration and the number of times the tip capsule is dipped in the solution, as outlined.^{[6, 22]} In scenarios where HybriBot is exposed to prolonged moisture, the tip capsule is dipped three times in the ethyl cellulose solution. For applications requiring the capsule to degrade within ≈24 h, dipping the capsule tip twice in the ethyl cellulose/ethanol solution suffices. In dry conditions, such as when stored in a petri dish without moisture, HybriBots remain stable for at least 5 months. After the first coating, the capsule of the biohybrid machine was coated with natural hairs of Avena sterilis by electrostating floating. The hairs were removed from the capsule using a scissor. Before drying, the natural hairs were directionally aligned toward the basal part of the capsule using tweezers.

Mathematical Modeling

An analytical biomechanical model was introduced for quantitatively describing the dynamics of sister awns (hereafter labeled as awn1 and awn2), based on their experimentally recorded torsion (see Figure S17, Supporting Information). Indeed, once characterized the hygromorphic behavior of individual (natural) awns, a physics-based modeling framework can be used to design the proposed biohybrid machines (e.g., for determining the relative pose of the awns for achieving mutual contact, which is functional to machine movement on/in soil). The actual capsule geometry was considered, as well as the measured awns morphometrical data and relative pose with respect to the capsule (see Table S1, Supporting Information). Each of the four involved segments (namely base beam and tail of both awn1 and awn2) were modeled as flexible rods with circular cross-section: each base beam was kinematically coupled to the corresponding tail (because tail bending, to a first approximation, is proportional to base beam torsion^[8] and contact was allowed between the two tails. Hence, each awn was individually actuated based on the recorded humidity-driven base beam torsion (Figure S17, Supporting Information), and contact interaction was accounted for between the two tails. More specifically and considering awns slenderness (whence negligible shear/extension effects compared to bending/torsion effects), each of the considered flexible segments was modeled by using the following (classical) Cosserat/Kirchoff governing equations^[32]

$$$$\begin{equation}\left\{ { \def\eqcellsep{&}\begin{array}{@{}*{2}{l}@{}} {{{\partial }_s}x = }&{R \cdot {{e}_3}}\\ {{{\partial }_s}q = }&{{{\Omega}} \cdot q}\\ {{{\partial }_s}v = }&{ - \hat{u} \cdot v + \hat{\omega } \cdot {{e}_3}}\\ {{{\partial }_s}\omega = }&{{{\partial }_t}u - \hat{u} \cdot \omega }\\ {{{\partial }_s}n = }&{\rho A\ R \cdot \left( {\hat{\omega } \cdot v + {{\partial }_t}v} \right) + f}\\ {{{\partial }_s}m = }&{\rho \ R \cdot \left( {\hat{\omega } \cdot J \cdot \omega + J \cdot {{\partial }_t}\omega } \right) - \ \widehat {{{\partial }_s}x} \cdot n} \end{array} } \right.\end{equation}$$$$(1)

where ∂_ς(·) denotes partial differentiation of (·) with respect to ς, s represents the arclength, x denotes the position (of a rod point, as hereafter tacitly understood) represented in a global Cartesian frame (for simplicity, normal font weight is used for both scalars and vectors, since the involved classical entities do not leave room for ambiguity), q is the quaternion determining the orientation of the local (i.e., body-attached) frame with respect to the global one, v is the velocity in the local frame, ω is the angular velocity in the local frame, n is the internal force in the global frame, and m is the internal torque in the global frame. Moreover, t denotes time, R  =  R(q) is the rotation matrix that maps vector representations from the local to the global frame, e₃ ≡ [0, 0, 1]^T (matrix formalism is hereafter used, for ease of readability), u represents the curvature in the local frame (whence matrix Ω  =  Ω(u) is well defined), ρ denotes density, A is the cross-sectional area, J represents the area moment of inertia tensor, f is the external force per unit length, and the operator $$\hat{ \cdot }$$ maps a given vector z  =  [a, b, c] to the skew-symmetric matrix $$\hat{z} = \ [ {0, - c,b;c,0, - a; - b,a,0} ]$$ (colon denoting row splitting). For clarity, let that Equation (1) stem to be mentioned from basic curve geometry, Equation (1) derive from related mathematical constraints, whereas Equation (1) provide the linear and angular momentum balance, respectively, for the considered rod. Finally, no external moments per unit length appear in Equation (1) because they are not relevant for the considered interawn contact.

In order to account for possible contacts between awn1 and awn2, a simple nonlinear penalty submodel, as detailed in the sequel, is introduced. Given a point $$P_i^{( 1 )}$$ on tail1 and a point $$P_j^{( 2 )}$$ on tail2, at relative distance $$d_{ij}^{( {12} )}$$, it is assumed them to be in contact, if $$\lambda _{ij}^{( {12} )} \equiv \varrho ( {r_i^{( 1 )} + r_j^{( 2 )}} ) - d_{ij}^{( {12} )} > 0$$, where $$r_i^{( 1 )}$$ and $$r_j^{( 2 )}$$ are the cross-sectional radii at $$P_i^{( 1 )}$$ and $$P_j^{( 2 )}$$, respectively, and ϱ  =  1.5 allows for some margin in contact detection (compared to a sharp limit, here given by ϱ  =  1, possibly causing instabilities during numerical integration). The force $$f_{ij}^{( {12} )}$$ exerted by tail1 on tail2, acting along the direction $$\delta _{ij}^{( {12} )}$$ that connects $$P_i^{( 1 )}$$ to $$P_j^{( 2 )}$$, was then computed as follows: $$f_{ij}^{( {12} )} = \ k\ F( {\lambda _{ij}^{( {12} )}} )\ \delta _{ij}^{( {12} )}$$, where k is a free model parameter and F(y) ≡ (max (y, 0))². It is observed that the above quadratic trend is a simple model law that simultaneously allows for smooth contact inception and stiffening upon reduction of the relative distance. Furthermore, k is the only free model parameter. Upon contact between $$P_i^{( 1 )}$$ and $$P_j^{( 2 )}$$, corresponding contributions were added to the linear force density f when numerically integrating Equation (1). Specifically, for tail1 and tail2, $$ - f_{ij}^{( {12} )}/\delta {{s}_1}$$ and $$f_{ij}^{( {12} )}/\delta {{s}_2}$$, respectively, were added, where δs₁ and δs₂ denote the corresponding size of the spatial discretization (see below).

Moreover, in order to close the considered problem, the above partial differential equations were complemented with the following constitutive relation

$$$$\begin{equation}m = R \cdot \left[ {K \cdot \left( {u - {{u}^{\mathrm{*}}}} \right) + \ \delta \ \left( {{{\partial }_t}u - \ {{\partial }_t}{{u}^{\mathrm{*}}}} \right)} \right]\end{equation}$$$$(2)

where u* is a given reference curvature (depending, in the general case, on both space and time), δ is a (positive) damping factor accounting for material internal dissipation (commensurate with available experimental data, a scalar damping was assumed and, for simplicity it was set low-enough not to unphysically overdamp awn dynamics during numerical integration, see below), and K ≡ (πr⁴/4)E[1,0, 0; 0, 1, 0; 0, 0, 1/(1 + ν)] is a stiffness matrix determining the elastic material response, based on the Young modulus E and the Poisson ratio ν. Specifically, for each of the four considered segments (i.e., base beam and tail of both awn1 and awn2), the reference curvature was assigned based on humidity-driven torsion data, as detailed in the sequel. Considering that the awn tails are not hygro-responsive (see Movie S1, tail part, Supporting Information), it was set u* = [0, 0, 0]^T  for the tails of both awn1 and awn2. Differently, for the base beam of awn1 and awn2, the swelling-driven torsion (see Movie S1, base beam part, Supporting Information) was set by

$$$$\begin{eqnarray}u_1^{\mathrm{*}}\ \left( t \right) &=& \left[ { \def\eqcellsep{&}\begin{array}{@{}*{1}{c}@{}} 0\\ 0\\ 1 \end{array} } \right]{\mathrm{\ \ }}\frac{{{{\theta }_{\infty 1}}}}{{{{L}_{b1}}}}\ \left[ {1 - \exp \left( { - \frac{t}{{{{\tau }_1}}}} \right)} \right]{\mathrm{\ and\ }}\nonumber\\ && u_2^{\mathrm{*}}\ \left( t \right) = \left[ { \def\eqcellsep{&}\begin{array}{@{}*{1}{c}@{}} 0\\ 0\\ 1 \end{array} } \right]{\mathrm{\ \ }}\frac{{{{\theta }_{\infty 2}}}}{{{{L}_{b2}}}}\ \left[ {1 - \exp \left( { - \frac{t}{{{{\tau }_2}}}} \right)} \right]\end{eqnarray}$$$$(3)

respectively, where (subscript 1 and 2 are hereafter understood,) the base beam lengths L_b1 and L_b2 were known from the measured geometry and the swelling-driven torsion angles, namely θ_∞1 and θ_∞2, as well as the swelling characteristic times, namely τ₁ and τ₂, were set based on the related measurements (Figure S17, Supporting Information). Let remark that only the humidity-driven torsion angles, i.e., $$\theta _1^*\ ( t ) = \ {{\theta }_{\infty 1}}[ {1 - \exp ( { - t/{{\tau }_1}} )} ]$$ and $$\theta _2^*\ ( t ) = \ {{\theta }_{\infty 2}}[ {1 - \exp ( { - t/{{\tau }_2}} )} ]$$, were used to set the reference curvature: the corresponding tail bending angles (i.e., the angles between the initial portion of each tail and the prolongation of the final portion of the corresponding base beam) were used for setting the problem boundary conditions, as detailed in the following paragraph.

For each awn, the aforementioned governing equations for the base beam, by considering a curvilinear abscissa s_b ∈ [0, L_b], and the governing equations for the tail, by considering a curvilinear abscissa s_t ∈ [0, L_t] were solved. The two sets of equations were coupled by the physical continuity at the knee point (identified by s_b = L_b  and s_t =  0), and by the kinematic constrain introduced by the humidity-driven bending angle ϕ*(t) between base beam and tail, still at the knee point. In particular, the bending angles $$\phi _1^*$$ and $$\phi _2^*$$ for awn1 and awn2, respectively, were given by

$$$$\begin{equation}\phi _1^{\mathrm{*}}\ \left( t \right) = \ \beta {{\theta }_{\infty 1}}\left[ { - \exp \left( { - \frac{t}{{{{\tau }_1}}}} \right)} \right]{\mathrm{\ and\ }}\phi _2^{\mathrm{*}}\ \left( t \right) = \ \beta {{\theta }_{\infty 2}}\left[ { - \exp \left( { - \frac{t}{{{{\tau }_2}}}} \right)} \right]\end{equation}$$$$(4)

with β≅4 · 10⁻² based on previous measurements of the bending angle.^[8] For each awn, the relevant boundary conditions then read as follows:

$$$$\begin{equation}\left\{ { \def\eqcellsep{&}\begin{array}{@{}*{2}{l}@{}} {x\left( {\ {{s}_b} = \ 0} \right) = }&{{{x}_0}}\\ {R\left( {{{s}_b} = 0} \right) = }&{{{R}_0}}\\ {v\left( {{{s}_b} = 0} \right) = }&0\\ {\omega \left( {{{s}_b} = 0} \right) = }&0\\ {x\left( {{{s}_t} = 0} \right) = }&{x\left( {{{s}_b} = {{L}_b}} \right)}\\ {R\left( {{{s}_t} = 0} \right) = }&{R\left( {{{s}_b} = {{L}_b}} \right) \cdot \Re \left( {{{\phi }^{\mathrm{*}}}} \right)}\\ {v\left( {{{s}_t} = 0} \right) = }&{{{\Re }^T}\left( {{{\phi }^{\mathrm{*}}}} \right) \cdot v\left( {{{s}_b} = {{L}_b}} \right)}\\ {\hat{\omega }\left( {{{s}_t} = 0} \right) = }&{{{\Re }^T}\left( {{{\phi }^{\mathrm{*}}}} \right) \cdot \hat{\omega }\left( {{{s}_b} = {{L}_b}} \right) \cdot \Re \left( {{{\phi }^{\mathrm{*}}}} \right) + \mho \left( {{{{\dot{\phi }}}^{\mathrm{*}}}} \right)}\\ {n\left( {{{s}_t} = 0} \right) = }&{n\left( {{{s}_b} = {{L}_b}} \right)}\\ {m\left( {{{s}_t} = 0} \right) = }&{m\left( {{{s}_b} = {{L}_b}} \right)}\\ {n\left( {{{s}_t} = {{L}_t}} \right) = }&0\\ {m\left( {{{s}_t} = {{L}_t}} \right) = }&0 \end{array} } \right.\end{equation}$$$$(5)

with ℜ(ϕ*) ≡ [1, 0, 0; 0, cos ϕ*, sin ϕ*; 0, −sin ϕ*, cos ϕ*], $$\mho ( {{{{\dot{\phi }}}^*}} ) \equiv [ {0,0,0;0,0,{{{\dot{\phi }}}^{\mathrm{*}}};0, - {{{\dot{\phi }}}^*},0} ]$$, and $${{\dot{\phi }}^*}$$ denoting the ordinary derivative of ϕ* versus t (in the case at hand, indeed, both ϑ* and ϕ* only depend on time). As anticipated, position x₀ and rotation R₀ were defined based on the measured awns morphometrical data: for the sake of definiteness, x₀ was chosen to lie on the frontal plane that (roughly) bisects the capsule through its wings, and R₀ was determined by the angle between the capsule axis (intersection between its frontal and sagittal planes) and the base beam direction, once assigned the angle Θ_root between the frontal plane and the plane defined by the initial pose of the awn. Overall, Θ_root must be regarded to as a fixed model parameter, corresponding to the actual degree-of-freedom that, in practical implementations, permits to rotate the initial pose of the awn around the base beam direction, without altering the awn wedging point on the fruit capsule. It should be noticed that, while Equation (5) and Equation (5), respectively describe the static wedging of the base beam and the load-free condition of the tail's tip, Equation (5) define a nontrivial pose coupling at the knee point, where time-varying bending plays a key role.

The governing equations were numerically discretized by adopting a method-of-lines approach.^[33] Specifically, the semidiscrete formulation was achieved by first discretizing in time with a BDF2 backward differentiation formula (also allowing to recast Equation (2) as an explicit relation for the semidiscrete variable corresponding to u). Spatial integration was then performed for (the semidiscrete counterpart of) Equation (1) by using a classical Runge–Kutta scheme: the unknowns ethyl was integrated by shooting on n( s_b =  0) and m( s_b =  0), and adopting the load-free condition given by Equation (5) (considered for both awn1 and awn2) as target. More in detail, the governing differential problem in nondimensional terms were recasted, and the aforementioned numerical integration on a common scientific computing environment (Matlab, MathWorks, USA) was implemented, by refining discretization up to achieving discretization-independent results. Finally, the following literature values were adopted for the involved biomechanical parameters: ρ  =  600 kg m⁻³,^[34] E  =  5.5 × 10⁹ Pa,^[7] ν  =  1/5.^[35] As remarked above, the only free model parameter is k: its range is physically bounded by the fact that, for too-low values, contact is not physically achieved (i.e., the resulting interaction force is so low that the tails end up with unphysically crossing each other), while for too-high values, an unphysical vibrating/jumping contact is achieved, as caused by kind-of impulsive reactions already during contact inception. Thus the “physical” range for k through numerical sweeps was explored, considering in particular the test-case associated with Figure 3f; and Movie S12 (Supporting Information), where two contacts occurred between the awns (with $${{\Theta}}_{{\mathrm{root}}}^{( 1 )} \cong - 3.2$$ and $${{\Theta}}_{{\mathrm{root}}}^{( 2 )} \cong - 4.1$$ rad). It turned out that physically representative behaviors were achieved for 0.8 < k < 1.4 (the nondimensional formulation was directly used to set k, so that it is consistently presented as nondimensional even in the main text) and, more remarkably, the numerical results were weakly-dependent on the specific k value in that range (Figure 3f). This, in turn, permitted to regard the model to as predictive, and the force results obtained from the model turned out to be in good agreement with related experimental force measurements.

To conclude, a few remarks should be added. The introduced model describes the dynamic interaction between the sister awns, by also accounting for their flexibility, based on the humidity-driven changes of their pose (which modulate the torsional curvatures of the base beams, while also modulating the relative bending angles between base beams and tails). Thus a physically-representative modeling framework that, notwithstanding the simplifications inherently introduced by the analytical approach (including its numerical integration), could be used for both quantitatively investigating the natural Avena fruits and designing the proposed biohybrid machines was introduced. For instance, richer design spaces (e.g., allowing for a wider set of awns-capsule relative poses) can be explored by using the proposed model, also based on the fact that the number of contacts can be preliminary estimated by a simplified rigid kinematics, contactless model (Movie S13, Supporting Information). At the same time, the proposed model could be further refined by explicitly describing the humidity diffusion dynamics, thus possibly tailoring the awns response to specific environments and (controlled) swelling conditions. Further extensions might be envisaged by integrating the developed model into a dynamic description of fruit movement, at the cost of increased complexity (commensurate with model calibration capabilities, including environmental aspects), thorough specifically focused studies beyond the present scope.

Load-Carrying Capacity

First the load-carrying capacity of the biohybrid machine was estimated by means of a simple biomechanical model of the maximum load workable by a single awn, and complementary experiments. As regards the model, the geometrical centroid of the capsule (denoted by G in Figure S22a, Supporting Information) was initially determined, by applying the classical divergence theorem to the capsule surface (discretely known through the aforementioned STL file). Observing that G and the apical point V (approximately) lie on the frontal plane x = 0 (through the base beams, Figure S22b, Supporting Information), and in view of the (approximate) symmetry with respect to the sagittal plane y = 0, the planar schematics in Figure S22c,d, Supporting Information was then considered. In both schematics, awn2 fosters the rotation around $${{\hat{e}}_1}$$ (i.e., the direction through V and the knee point K₁ of awn1), whereas load $$\vec{W}$$ (orthogonal to the frontal plane and applied at G, for simplicity) contrasts such a rotation. (By symmetry, the following estimates can be consistently obtained via the alternative assumption that it is awn1 to foster rotation, along the direction passing through V and the knee point K₂ of awn2.) More in detail, in Figure S22c, Supporting Information, the action of awn2 is represented by torque $${{\vec{T}}_{\mathrm{a}}}$$ (along the capsule axis) and, by assuming $${{\vec{T}}_{\mathrm{a}}} = {{\vec{T}}_{{\mathrm{max}}}}\ $$ such that $$| {{{{\vec{T}}}_{{\mathrm{max}}}}} | \cong 100.4$$ mN mm⁻¹ (see Section 2.2), the maximum load $${{\vec{W}}_{{\mathrm{max}}}}$$ is determined by considering the null-torque condition around $${{\hat{e}}_1}$$, namely $$( {VG \times {{{\vec{W}}}_{{\mathrm{max}}}} + {{{\vec{T}}}_{{\mathrm{max}}}}} ) \cdot {{\hat{e}}_1} = \ 0$$. Differently, in Figure S22d, Supporting Information, the action of awn2 is represented by force $${{\vec{F}}_{\mathrm{a}}}$$ (orthogonal to the frontal plane), as applied at the knee point K₂. Such a schematization leverages the fact that the awn contact force was directly measured (whereas the associated torque, exploited in the previous schematization, was obtained also based on a complementary quantification of the contact force arm), yet it can generally underestimate the force arm with respect to $${{\hat{e}}_1}$$. In view of this point, $${{\vec{F}}_{\mathrm{a}}} = {{\vec{F}}_{{\mathrm{max}}}}\ $$ such that $$| {{{{\vec{F}}}_{{\mathrm{max}}}}} | \cong 6.1$$ mN (i.e., the maximum contact force measured for the biohybrid machine is assumed; Figure 3e; and Figure S21, Supporting Information) and an additional estimate for $${{\vec{W}}_{{\mathrm{max}}}}$$ via the null-torque condition $$( {VG \times {{{\vec{W}}}_{{\mathrm{max}}}} + V{{K}_2} \times {{{\vec{F}}}_{{\mathrm{max}}}}} ) \cdot {{\hat{e}}_1} = \ 0$$ was obtained. Based on the above schematizations, the maximum load workable by a single awn was estimated to be around 2.6 ÷ 4.3 gf. Furthermore, considering that the awns used in the proposed biohybrid machine are natural awns and by assuming that potential boundary effects at the wedging points (of the awns on the capsule) play a minor role, the above computations were repeated by considering the maximum torque ($$| {{{{\vec{T}}}_{{\mathrm{max}}}}} | \cong 81.9$$ mN mm⁻¹, see Section 2.2) and force ($$| {{{{\vec{F}}}_{{\mathrm{max}}}}} | \cong 5.6$$ mN; Figure 3e; and Figure S21, Supporting Information) previously characterized for the natural fruit, thus estimating the maximum load workable by a single awn to be around 2.4 ÷ 3.5 gf. Finally, the load-carrying capacity of the biohybrid machine is experimentally measured by means of calibration weights (Smoostart/DDZ, China), by incrementally loading the capsule (at steps of 0.5 gf) until it could not be raised by a single awn (hydrated through 0.01 mL of water). Based on the carried out experiments, the maximum load workable by a single awn was around 3.0 ÷ 3.5 gf (Figure S22e–g, Supporting Information), in good agreement with the above model-based estimates.

To conclude, a few remarks can be added. First, in order to perform the above estimates, the load was considered as applied at the centroid, for simplicity: based on subsequent implementations (e.g., featuring internal structuring/compartmentalization of the capsule domain), more elaborate approximations could be introduced. Second, the maximum load workable was deliberately considered by a single awn, because assuming synergistic actions by the two awns can lead, in general, to nonconservative estimates (and, reciprocally, potential awns synergy over a certain time interval may not dramatically alter the as-estimated load-carrying capacity). Third and last point, the force capacity of the awns outstrips the capsule-carrying requirement to the point that it was not detrimental, for the purpose of the estimates, to consider the “initial lift” of the capsule rather than a “sustained deployment” thereof (although necessarily simplified, due to, e.g., environment variability). Indeed, the aforementioned maximum load is considerably greater, for instance, than the (capsule weight, i.e., 60 mgf, as well as the) weight that would be theoretically reached by increasing the capsule density up to physically reasonable limits (for the sake of argument, e.g., an iron capsule with the considered morphometry would have a mass around 0.3 ÷ 0.4 g). This suggests the possibility to use the Avena awns to also displace larger, nonbiomimetic load containers (beyond the present scope). Moreover, the fact that the force capacity of the Avena awns significantly oversteps the force needed to lift the Avena capsule is consistent with the extended functionality of the awns, as further remarked in the main text.

Nanoindentation

Nanoindentation experiments were performed on natural capsules of Avena sterilis fruits and biomimetic artificial capsules made with dough using an iNano indentation system (Nanomechanics, Inc.).^[36] The system was equipped with a moving stage, metal holders, two microscopes and a sharp probe (Figure S13a, Supporting Information). The natural and artificial capsules were fixed on the metal holders using a UV-curable glue (UHU Booster). Before fixing the capsules, the hairs were removed with a scissor to avoid artefacts in the mechanical characterizations. All tests were performed using a Berkovich tip (with a Young's modulus of 1141 GPa and a Poisson's ratio of 0.07), following the protocol exploited.^[9] The material was subjected to a load function of 5 mN with a target depth of 300 nm and an indentation strain rate of 0.2% per second. Specifically, a Dynamic Nanoindentation Constant Strain-Rate Method^[37] for all natural and artificial samples to extract the storage modulus's values was used. All experiments were performed at room temperature (25°). In total, nine nanoindentation experiments were performed on three different replicates of natural or artificial capsules (n = 9 in artificial samples; n = 9 in natural samples).

Drag Force Tests on Natural and Artificial Capsules

In order to measure the forces required for the natural fruits and biohybrid machines to dig into soil (i.e., drag forces), a dedicated setup with a multiaxis measurement platform was used, equipped with a three-axis micrometric translation stage M-111.1DG for vertical motion (Physik Instrumente GmbH), a C-863 Mercury Servo Controller (Physik Instrumente GmbH) and a six-axis force/torque sensor (ATI Industrial Automation, Nano17; limit of resolution = 0.317 gram per force) (Figure S13, Supporting Information). The testing cycles and data acquisition, conducted at a frequency of 20 Hz with an averaging level of 200 samples, were controlled by specialized software developed in VB.NET (Microsoft Corp.). The protocols used for other fruits were followed.^{[9, 23, 38]} In all tests, natural or artificial capsules were vertically fixed to a supporting squared silicon wafer by a two-component silicon rubber (Figure S15, inset, Supporting Information) that was connected to the load cell. A cylindrical glass container (basal diameter = 9 cm, height = 5 cm) containing the selected granular substrate was placed under the capsules over the rotational support (Figure S15, Supporting Information). During the tests, the capsules were moved toward the glass container at a speed of 0.2 mm s⁻¹ for 10 mm. In the experiments aimed at measuring the drag reduction of natural and artificial capsules through their rotation, the drag forces rotating the beaker at 1 rpm were measured, similar to,^[23] while maintaining the constant speed of 0.2 mm s⁻¹. For each test run, the maximum peaks of force–displacement curves were calculated. Drag force tests were carried out with or without rotation of the natural and artificial capsules in different substrates, including one type of artificial soil and three different types of natural soils. The artificial soil consists of glass beads having a diameter of 1 mm. The natural dried soils were purchased from online or in situ shops and include: 1) loam soil (Terriccio Universale, Coop), 2) sandy soil (Sabbia fine, Clorophylla), and 3) clay soil (Terriccio Argilla, TERRA). The soil bulk density (D_b) of each substrate was calculated from the formula D_b = W_ds/V_c, where W_ds is the weight of the dried soil in the container and V_c is the volume of the container occupied by soil.^[39] The bulk density of artificial soil is 1429 g cm⁻³; loam soil is 0582 g cm⁻³; sandy soil is 1171 g cm⁻³; and clay soil is 1058 g cm⁻³. A total of six tests were performed for each type of experiments on natural and artificial samples with or without rotation in the different types of dried soil (n = 6). To compare the drag forces in dry and wet soil conditions, the drag forces of natural and artificial capsules with and without rotation in one selected soil substrate (i.e., dried and wet clay soil), using the parameters reported above were tested. Wet soil consists of clay mixed with 50 mL of water. In this case, a total of six tests are performed on natural samples (n = 6) and a total of three tests were performed on artificial samples (n = 3).

Rotation of Natural Awns

Two natural sister awns of Avena sterilis were removed and fixed to a silicon wafer substrate using a UV-responsive glue. A video camera (Canon) parallel to the plane of the samples was used to record the rotation of the awns when humidity around the awn was increased by supplying vapor with a commercial aerosol humidifier (Movie S10, Supporting Information). The rotation angle was computed by postprocessing Movie S10 (Supporting Information) at different timeframes using ImageJ. The associated rotational speed was then computed based on a fitting trend, as further detailed in Figure S17 (Supporting Information).

Torque Measurements on Awns of Natural and Biohybrid Machines

In order to measure the torques generated by the awns of natural and biohybrid samples, a dedicated setup with a 10 g sensitive load cell (Futek LSB200, Futek Advanced Sensor Technology Inc., US) was used. In the experiment, the capsules of natural and artificial samples were fixed on an external support at 2 mm from the load cell as showed in Figures S16 and S17 (Supporting Information). When humidity was increased, the awn started rotating and got in contact with the load cell. The forces were measured by blocking the rotation on the awn's tail, similar to^[38] A silicon wafer substrate was attached to the load cell to facilitate the block of awns rotation. To increase humidity and perform the experiment without interfering of both two sister awns, 0.01 mL of water was added to only one awn of natural or biohybrid samples. To extract the torque, forces exerted by rotation of the awn's tail in natural samples and in biohybrid machines were measured and the moment as: M = F*r, where F is the force exploited by the awn and r is the radius between the reference axis and the contact point of the tail was extrapolated. The axis of the natural or artificial capsules as the fixed reference axis was considered. All experiments were recorded (i.e., Movie S11, Supporting Information) and postprocessed in ImageJ to extract the radius at different force values (i.e., at 1/3 of the Fmax, 2/3 of Fmax, and 3/3 of Fmax by considering the first cycle of the rotational awn), as shown in Figure S16 (Supporting Information). The torque produced by three different awns in natural samples and in biohybrid machines (n = 3, in natural samples; n = 3, in biohybrid machines) was measured.

Exploration Tests of Biohybrid Machines in Climatic Chamber

The soil exploration behavior of Hybribots in controlled climatic test chamber (CTC256, Memmert GmbH, Germany) was investigated. Samples were subjected to humidity ramp at controlled temperature (30 °C) from 30% RH to 98% RH, while simultaneously video recording the behavior using a camera (Logitech Brio Stream, Logitech, Swiss). In total, 4 RH cycles in about 1 h 30 min were recorded. To investigate the absolute speed of Hybribots on unstructured soil environments, the recorded video was then postprocessed by using Tracker (Video analysis and modeling tool, https://physlets.org/tracker/) motion capture software.

Germination Experiments and Time-Lapse Recording

Germination experiments were conducted with functionalized fully biodegradable biohybrid machines with capsule made of flour. 10% w/w of biochar was mixed into the flour to made the functionalized biohybrid machines. Different seeds or fruits were embedded into the artificial capsules made with flour and functionalized with biochar, including two tomato seeds and two chicory fruits for each biohybrid machines, before covering the samples with hairs (as explained above in “Microfabrication of the biohybrid machines” section). In total, three biohybrid machines with embedded two tomato seeds or three biohybrid machines with embedded chicory fruits were used for the germination experiments (n = 3, where n = number of replicates). The germination experiments were performed in the growth chamber (temperature 25°, 60% humidity) using loam universal potting soil. The germination of biohybrid machines was recorded in the growth chamber using time-lapse photography at 5 min interval with a Brinno camera. The duration of germination experiments of biohybrid machines with embedded tomato seeds is 35 days, while is 30 days in case of biohybrid machines with embedded chicory fruits. As proof-of-concept, biohybrid machines embedding flowers such as Salcerella fruits were preliminarly tested. In this case, the fruits were not embedded inside the dough but directly mixed with the dough and biochar before the capsule molding (Figure S24, Supporting Information).

Statistics

All experiments were performed at least in triplicate, performing three independent experiments on different samples. Error bars in the graphs represent the standard deviation. ANOVA and Post-hoc Tukey's tests were conducted at the 99–95% significance level for multiple comparisons, as necessary. Data were elaborated in Excel and GraphPad Prism.

Acknowledgements

This work was funded by the European Union's Horizon 2020 Research and Innovation Programme (I-SEED project) under Grant Agreement No. 101017940 and by “RAISE – Robotics and AI for Socio-economic Empowerment” project and European Union — NextGenerationEU – Italy's National Recovery and Resilience Plan. I.F. was supported by a DAAD – Short-Term Research Grant for mobility by German Academic Exchange Service. T.S. and I.F. acknowledged funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy-EXC-2193/1-390951807 and access to the shared laboratory facilities of the cluster. I.F. additionally acknowledged the support of the Convergence Center for Living Multifunctional Material Systems (LiMC²) and the Cluster of Excellence Living, Adaptive and Energy-autonomous Materials Systems (livMatS) Living Multifunctional Materials Collaborative Research Seed Grant Program.

Open access funding enabled and organized by Projekt DEAL.

Conflict of Interest

I.F., T.S., E.S. and B.M. have a pending patent application related to the biohybrid machine inspired by wild oat fruits and related production process.

Author Contributions

I.F. and B.M. conceived the idea of biohybrid machine. I.F. designed the experimental work, microfabricated all samples, conducted experiments (except histology), performed data analysis, wrote initial draft, edited, and reviewed the manuscript. E.S. conceived and performed the modeling, supported data analysis regarding the mechanical characterizations of the awns, wrote modeling part, edited, and reviewed the manuscript. M.R. performed histological analysis on natural awns, wrote histological part, edited and reviewed the manuscript. T.S. provided funding and support for µCT experiments and edited and revised the manuscript, B.M. supervised the work, provided funding and edited and revised the manuscript.

Open Research

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.