SOFT ROBOTICS

Volume 4, Number 3, 2017

Mary Ann Liebert, Inc.
DOI: 10.1089/soro.2016.0080

ORIGINAL ARTICLE

Toward Self-Growing Soft Robots Inspired by Plant

Roots and Based on Additive Manufacturing Technologies
Ali Sadeghi, Alessio Mondini, and Barbara Mazzolai

Abstract

In this article, we present a novel class of robots that are able to move by growing and building their own structure. In
particular, taking inspiration by the growing abilities of plant roots, we designed and developed a plant root-like
robot that creates its body through an additive manufacturing process. Each robotic root includes a tubular body, a
growing head, and a sensorized tip that commands the robot behaviors. The growing head is a customized three-
dimensional (3D) printer-like system that builds the tubular body of the root in the format of circular layers by fusing
and depositing a thermoplastic material (i.e., polylactic acid [PLA] filament) at the tip level, thus obtaining
movement by growing. A differential deposition of the material can create an asymmetry that results in curvature of
the built structure, providing the possibility of root bending to follow or escape from a stimulus or to reach a desired
point in space. Taking advantage of these characteristics, the robotic roots are able to move inside a medium by
growing their body. In this article, we describe the design of the growing robot together with the modeling of the
deposition process and the description of the implemented growing movement strategy. Experiments were per-
formed in air and in an artificial medium to verify the functionalities and to evaluate the robot performance. The
results showed that the robotic root, with a diameter of 50 mm, grows with a speed of up to 4 mm/min, overcoming
medium pressure of up to 37 kPa (i.e., it is able to lift up to 6 kg) and bending with a minimum radius of 100 mm.

Keywords: growing robots, 3D printing, additive manufacturing, plant-inspired robots, soft robots

Introduction more safely with humans.4,5 Structures made of soft matter

with variable stiffness properties6,7 give rise to continuum

T he emergence of bioinspired approaches represents an

innovative way to rethink robot design since new strat-
egies, new patterns of movement, and new sensing and ac-
soft robots that, differently from the former rigid-link robots,
are able to undergo elastic deformation at any point in their
structures for producing structural movements.8–10
tuation abilities can be proposed to build innovative robotic The use of soft materials and continuum deformations
systems.1 This bioinspired evolution of robotics also supports paves the way for scientific and technological challenges,
the effectiveness of soft bodies, as living organisms exploit which include developing control solutions that are more
soft tissues and compliant structures to move effectively in difficult in computational terms, defining new design rules,
complex natural environments.1,2 and also developing and adopting innovative manufacturing
Soft robotics, which focuses on generating devices based technologies, including combinations of advanced fabrica-
on compliant and deformable materials in the interaction with tion techniques (e.g., micromolding,11 soft lithography,12 and
the environment,3 represents a recent disruptive technology multimaterial-embedded three-dimensional (3D) printing13).
and an approach to develop new classes of robots expected to The complexity of developing bioinspired soft robots is in-
act in natural, unstructured environments (e.g., locomotion in creased by the need for mimicking biological system cap-
uncertain terrains, manipulation of unknown objects, and abilities in being energetically efficient, in changing their
accomplishment of non-predetermined tasks) and to interact morphology, in adapting their body and functionality in their

Center for Micro-BioRobotics (CMBR), Istituto Italiano di Tecnologia (IIT), Pontedera, Italy
ª Ali Sadeghi et al. 2017; Published by Mary Ann Liebert, Inc. This article is available under the Creative Commons License CC-BY-NC
(http://creativecommons.org/licenses/by-nc/4.0). This license permits non-commercial use, distribution and reproduction in any medium,
provided the original work is properly cited. Permission only needs to be obtained for commercial use and can be done via RightsLink.

211
212 SADEGHI ET AL.

lifetime, by growing, or even by a morphological environ- types of soil because they grow at the apical level of their
mental adaptation.14,15 roots by the division and generation of new cells and elongate
Hence, the challenge ahead for soft robotics is to further on absorbing water from their external environment by os-
develop the abilities for robots to grow, evolve, self-heal, de- mosis. These phenomena occur in the root meristem zone and
velop, and biodegrade, which are the ways that robots can adapt elongation region, respectively.28 Analogously to a natural
their morphology to the environment.16 In this context, growth organism, the root-like device penetrates the medium by
is a very interesting possibility that would allow the creation of imitating the new cell generation by adding layers of artificial
robots endowed with new and unpredictable abilities of material at its tip level, which results in a growth process.
movement, able to assume any shape on the basis of the task to This addition process provides a pushing action for over-
be accomplished. An example of these approaches is given by coming the soil pressure and penetrating the soil, while at the
the spider-like robot proposed in a study by Wang et al.,17 same time it also elongates the structure of the artificial root.
where improved locomotion capabilities in a 3D space are ob- The mechanism that performs the material deposition
tained by thermoplastic spinning of draglines. Innovative so- process (called the growing mechanism) has a short length
lutions to address these challenges can be obtained by studying and slides inside its self-developed tubular structure and al-
and imitating plants. Unlike the majority of animals, which ways remains at the apical part of the robotic root. The body
grow until they reach maturity, plants grow for their entire life. of the robot does not move with respect to the soil and only its
In this work, we present a new concept of a self-growing apical part performs the penetration. This localization of
robot that is inspired by the movement and adaptation cap- the frictional interactions at the tip level makes the behavior
abilities of plant roots in soil. By uncovering the fundamental and efficiency of the system independent from its length, in
aspects of the chosen biological model, we aim to develop contrast with the sloughing mechanism.26
novel robotic morphologies and controllers specifically op- Even if this system successfully validates and demonstrates
timized for burrowing. the advantages of following the plant root strategy in pene-
Soil is a poorly explored environment in robotics. Some of tration, the proposed solution will still not be able to provide
the robotic solutions developed in this direction are inspired the capability of deep penetration because the new layers of
by legless animals because of their exceptional ability to material added at the tip level are not sufficiently sticky to build
move in unstructured environments in different conditions a proper structure, which makes the system limited to deep
(rocks, desert, forests, underground, etc.) and their high sta- penetration and bending. In fact, in addition to straight pene-
bility (barycenter in proximity to the soil).18 There are al- tration, the plant roots can perform a bending task in soil for
ready robots inspired by legless animals for manipulation and following/escaping from an environmental stimulus or avoid-
movement in 3D spaces,19 medical applications,20,21 pipe ing an obstacle (tropisms29–32) by the differential elongation of
inspection,22 and soil locomotion.23,24 Although these robots the cells in the elongation region. Mimicking this feature, we
have implemented interesting solutions, the penetration of can improve the capabilities of the robotic root to explore and
deep soil remains challenging due to the peripheral frictional penetrate unknown environments.
interaction with the soil and the power transmission. In the current work, we propose a new generation of growing
Recently, plant roots have been proposed as a new bio- robots able to build their own bodies. In particular, we present a
logical model for developing advanced soil drilling robotic novel robotic system, inspired by plant roots, able to move in a
solutions. Plant roots are excellent natural diggers, and their granular medium through a layer-by-layer deposition process.
characteristics such as adaptive growth, low-energy con- These features are achieved by the integration of a customized
sumption movements, and the capability of penetrating the 3D printer inside the root. To the best of our knowledge, this is
soil at any angle are interesting from an engineering per- the first robot able to grow by changing shape and expanding its
spective.25 A first attempt at developing a mechanism in- body on the basis of the selected task and that integrates a 3D
spired by plant roots was reported in an article.26 In this work, printer-based solution. Taking advantage of these characteris-
we translated the outward flow of growing cells at the tip of tics, the robotic roots are able to move inside a medium by
the root and the low frictional interaction with the soil pro- growing their body in a straight direction or by bending. These
vided by sloughing cells to an engineering prototype that is growing roots have been integrated in a PLANTOID,33 which
able to penetrate into the soil. is a plant root-inspired robot that embeds innovative technol-
The system simulates these features by the outward move- ogies, such as plant-inspired materials in its leaves34 and plant-
ment of a soft sleeve from the internal hole of a rigid tube to its inspired roots with soft bending capabilities.33
external face. In addition, we studied and imitated the positive This article is organized as follows. First, we present the
effect of root hairs in providing anchorage and preventing robot design, modeling of locomotion, details of the depo-
upward movement instead of downward penetration by adding sition strategy and results of the experimental trials. Next, we
lateral hairs to the sleeve. The main challenge of the proposed describe the robot prototype fabrication and the related
solution is the tissue flexibility that does not prevent the control architecture and then provide details on the setups and
transmission of soil pressure to the rigid shaft during the pen- experiments used for the robot characterization. Last, we
etration. This problem becomes more significant when the discuss the limits and advantages of the current version of the
system penetrates into deeper soil, because the medium lateral robot, as well as possible future applications.
pressure to this system becomes higher. Moreover, a long and
rigid shaft that does not bend or change its orientation offers a Results
limited number of robotic applications.
Growing robot design
To overcome these limits, as given in Ref.,27 we proposed
another artificial solution that was based on the root-growing The robotic root is designed by exploiting the 3D printer
strategy. Plants are able to efficiently penetrate different and classic FDM (fused deposition modeling) approaches.
GROWING ROBOTS BY ADDITIVE MANUFACTURING 213

Each robotic root includes a tubular body, a growing head, and the sharpness of their corners provide an interaction with
and a sensorized tip that commands the robot behaviors the tubular body that prevents the rotation of the growing
(Fig. 1). The growing head is a customized 3D printer-like mechanism inside the printed tubular body and, at the same
system that builds the tubular body of the root in the format of time, allows the axial sliding of the growing system along the
circular layers by fusing and depositing the filament of a raw growth direction. The deposition process of the root body
material at the tip level. This miniature 3D printer-like ma- occurs under the rim of the deposition head. The addition of
chine includes an extruder and plotting units (Fig. 2A, B). The each new layer between the deposition head and the tubular
extruder unit (similar to FDM 3D printers) includes a gear- body pushes the growing tip forward to penetrate the soil. The
based feeder mechanism, a guiding tube, a heater as a liq- pressure generated by the soil on the deposited layers guar-
uefier, and a nozzle (Fig. 2B). The gear-based feeder provides antees the realization of a compact and solid robotic root
a gripping action to the thermoplastic filament, pulling it from structure because they stick to each other.
the reservoir and then pushing it toward the heater and nozzle. A customized tip, including miniaturized sensors (see
The guiding tube changes the direction of the filament Sadeghi et al.33 for details) and connected through a bearing
(*90) from the external gate of the feeder to the internal gate to the deposition head, collects environmental information
of the heater. The components of the extruder are mounted on and decides on the growth orientation on the basis of the
two parallel disks, called the deposition disk and rim (Fig. 2). embedded behavior (Fig. 2C).33
After the final assembly, these disks are fixed together and the
deposition process occurs by the relative rotational motion of
Robotic root movement in soil by growing
these disks with respect to the tubular body of the root. The
plotting unit is responsible for providing this relative rota- The simultaneous actions of feeding and rotational plotting
tional motion by means of a DC gear motor and an internal permit the layer-by-layer deposition of the fused material in a
gear, placed between the motor holder disk and the deposition tubular shape for the realization of the root body. The layers
head (Fig. 2A). This assembly smoothly functions as a flat can be deposited in complete circles or in sectors of circular
bearing due to two couples of raceways on the corresponding patterns. The compliancy and softness of the fused material
surfaces of the disks filled with bearing spheres. can be tuned by controlling the heater temperature and the
The internal gear, which also functions as the main body of feeding speed; this aspect, together with the independency of
the 3D printer, is interfaced with the root tubular body the extruder and the plotting units, allows controlling the
through four flexible clamping fingers installed on its cir- speed and orientation of the plotting. These features provide
cumferential sides (Fig. 2A). The flexibility of these fingers flexibility in the resulting thickness, shape, and position of

FIG. 1. A general view of a PLANTOID: a robot with several (two in the picture for clarity) growing roots for soil
exploration. Each root has its local control unit at the tip level, while another electronic unit placed inside the trunk manages
the communications between the roots and an operator. The root tips are equipped with soft touch sensors at their apex to
measure the soil pressure and detect obstacles, a gravity sensor to determine the tip orientation, and sensors for temperature,
humidity, and chemicals on their peripheral surface to monitor the surrounding environmental conditions and implement
the plant root behaviors.33
214 SADEGHI ET AL.

FIG. 2. CAD model of the

growing mechanism, includ-
ing the sensorized tip, (A) the
plotter unit, (B) the extruder
unit, and (C) the sensorized
tip decoupled from the grow-
ing mechanism by means of a
ball bearing, while their rela-
tive rotation can be monitored
by a magnetic encoder moun-
ted on the control unit and a
magnet at the axis of the tip.

the layers. Similar to our previous work,27 the force trans- complete deposition cycle. Therefore, the penetration depth
mitted from the anchored robotic root body to the advancing P, resulting from the deposition of a layer with length Ld,
tip can be modeled as the force applied by a screw by ex- thickness t, and width s in a tubular root shape with an ex-
ploiting the helical deposition of the filament material. ternal diameter D, is given as follows:
Nevertheless, different from the previous work in which
the filament diameter (d) was assumed constant during the t Ld
deposition process, in this case, due to the heating and the P ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Ld ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (2)
1
fusion process, after deposition, the filament section changes ðpðD  sÞÞ2 þ t2 ) ðtan aÞ2
þ 1)
to an almost rectangular shape (with a thickness t and a width
s) (Fig. 3). A rectangular deposited layer of length Ld can be where a is the helix pitch angle, which can be expressed as
obtained by a filament of raw material with an initial diameter follows:
d and length Lr.
t
 2 tan a ¼ (3)
p Lr d2 pðD  sÞ
Ld ¼ (1)
ts
A successful plotting elongation process can be performed
Considering the helical shape of the deposited layers, the when the system has enough power to overcome the internal
tip penetrates a distance equal to the layer thickness in each force Wf (generated by the clamping fingers) and the external
GROWING ROBOTS BY ADDITIVE MANUFACTURING 215

axial forces Ws (e.g., vertical force generated by soil pres-

sure). Defining l as the contribution of the friction force
between the depositing layer and deposition head, the re-
quired torque M at the deposition head is

  D  s l þ tan a
M ¼ Ws þ Wf : : (4)
2 1  l tan a

Playing with the deposition parameters (i.e., heater

temperature, feeding speed, and deposition speed), it is
possible to change the thickness of the material and obtain
different a values. Thus, the generated axial force and the
penetration speed can be tuned on the basis of the me-
dium that the robot has to penetrate. The provided flexi-
bility in the deposition process also permits the addition
of different amounts of material on the sides of the robotic
root structure and creates an asymmetry with a conse-
quent bending in the lower deposition direction (Fig. 4B,
C). This asymmetry caused by the material’s differential
deposition imitates a similar behavior observed in natural
roots that is generated by the differential division and
FIG. 3. (A) Schematic view of the growing mechanism. elongation of new cells at their apexes, causing the bend-
Red arrows represent forces acting on the system during the ing of the root.
deposition process: F is the force applied by the plotter From an exquisite engineering view, the bending can be
motor, W is the vertical resistance during penetration compared to the locomotion of a classic two-wheel mobile
(composed by external resistance WS and internal frictional robot on a planar surface, where the deposition speed on the
forces Wf), M is the torque required for both deposition and two sides of the structure is equivalent to the speed of the two
penetration processes to overcome W, t is the thickness of wheels. In particular (Fig. 4D), the instantaneous curvature of
the filament after the deposition, and D is the external di- the structure rc can be obtained as follows:
ameter of the tubular body. (B) Equilibrium of forces acting
on the deposited layer for one complete unwound turn, (A) a differential deposition of the same number of layers
where a is the angle made by the helix of the deposited k with different thicknesses h1 and h2 (Fig. 4B) or
filament with respect to a plane perpendicular to the axis of (B) a differential deposition of different numbers of
the tubular body, N is the reaction force, and l is the friction layers n and m with the same thickness h (Fig. 4C).
coefficient between deposited layer and deposition head.

FIG. 4. Possible penetration strategies that the robot can adopt to move straight or bend; (A) a symmetric deposition of
material behind the deposition head results in straight growth; (B) a bending of the robotic root caused by deposition of
different thickness material layers; (C) a bending of the robotic root caused by the addition of a variable number of layers on
a side of the root body (combination of complete cycles and sectors of cycles); (D) definition of the curvature parameters in
the cases of different numbers of deposited layers; and (E) the curvature radius of the structure as a function of the ratio
between heights of layers (case A) or a number of layers (case B) deposited on the two sides, calculated, respectively, by
means of equation (7A) or (7B).
216 SADEGHI ET AL.

Using rin for the internal structure curvature radius, rex for (n/m > 1). A value of 2 for the n/m deposition ratio was suf-
the external one, and b for the curvature angle, we obtain for ficient to obtain the minimum bending radius RM of 100 mm
the two cases in the root structure.

A B
   
D D
k  h1 ¼b  rin ¼b  rc  (5A) m  h¼b  rin ¼b  rc  (5B)
2 2
   
D D
k  h2 ¼b  rex ¼b  rc þ (6A) n  h¼b  rex ¼b  rc þ (6B)
2 2

h2 n
D h2 þ h1 D h1
þ1 D nþm D m þ1
rc ¼ ¼ rc ¼  ¼  (7B)
2 h2  h1 2 h2
(7A) 2 nm 2 n 1
1 m
h1

Ideally, case A creates a more continuous structure be- Deposition process

cause by tuning the diameter of the extruded filament it is
possible to deposit a more homogeneous helical structure. The filament extending from the nozzle should be partially
This structure can be realized by maintaining a constant liquefied. The main challenge is to obtain a material that is
feeding speed and extrusion temperature and regulating the externally sticky enough to permit the layers to attach to-
plotting speed. However, tuning precisely the plotting speed gether but internally semisolid such that it is sufficiently
is a challenging task and turning in the same direction can strong enough to overcome the soil pressure and push the tip
also create the problem of cable twisting. In our im- ahead. Indeed, if the material that comes out from the nozzle
plementation, we used case B because it can intrinsically is completely liquefied, it is squeezed around by the soil
prevent the problem of twisting of the filament of the raw pressure and no penetration can happen. On the contrary, if it
material and power wires (see more details in the Deposition is too solid, penetration can occur but no structure would be
Process section); moreover, the deposition process needs obtained.
less accuracy because the softness of the extruded filament The optimal temperature and feeding speed values were
compensates intrinsically the gaps created by the on/off experimentally defined as 180–200C and 8 mm/s, respec-
deposition strategy. tively, to obtain successful penetration and fabrication of the
In the performed experiments, straight growth was ob- growing structure. We selected strategy B for the bending, so
tained by a symmetric deposition of layers (n/m = 1) and the deposition process was planned with alternating changes
bending is obtained by depositing different numbers of layers in the orientation of the plotting cycles. This strategy prevents

FIG. 5. Schematic view of the parameters selected for the material deposition process (left); representation of the phase
sequence required for straight growth and bending (right). (A) The filament is first deposited toward the right; (B) the
feeding is stopped while the rotation of the deposition head continues for an EA of amplitude EA to remove the material in
front of the nozzle; (C) the rotation is stopped for 3–5 s more to cool down the old layer; (D–F) the process restarts in the
opposite direction for depositing the next K layers. EA, extra angle.
GROWING ROBOTS BY ADDITIVE MANUFACTURING 217

The differential deposition task includes three states:

(1) The first state starts when the deposition and feeding
motor are turned on. The filament comes out from the
extruder and is deposited on the old material, realiz-
ing the structure and at the same time pushing the tip
ahead (Fig. 5A);
(2) When the extruder goes out from the differential de-
position sector, the feeding motor is stopped, while the
plotting rotation continues for an EA of amplitude EA
before the plotting motor is stopped (Fig. 5B); and
(3) After a few seconds (3–5 s), necessary for cooling the
old layer (Fig. 5C), the plotting rotation starts in the
opposite direction together with feeding (Fig. 5D, E).
The filament rises above the old layer (Fig. 5E) and
continues to grow (feeding+deposition) for a defined
number (K) of complete layers (Fig. 5F) until the next
inversion.
FIG. 6. Results of straight growth in air under different
loads; the graphs present the thickness of the printed layers
under different loads at different temperatures. Validation of the robot movement by growing ability
The prototype was tested in three different scenarios (see the
Materials and Methods section for details). The performed ex-
the accumulation of twisting and torsion in the polylactic acid periments were particularly aimed at establishing (1) the thick-
(PLA) thermoplastic filament pulled from the reservoir and in ness of the deposited layers at different extrusion temperatures
the power wires connected to the trunk. The deposition and applied pressure conditions; (2) the bending capabilities and
strategy can be arbitrary; in the following, we consider that resulting curvature radius; and (3) the robot’s ability to penetrate
the deposition occurs by a sequence of straight growing (SG) through a soil-like environment by growing.
cycles followed by a single differential sector deposition Figure 6 shows the resulting layer thickness on maintain-
(DD), and we call the ratio between these two parameters the ing fixed feeding and plotting speeds and varying the extru-
deposition ratio (K = SG:DD), where DD is always equal to 1. sion temperature and applied pressure (impedance). The
The maximum curvature is obtained for K = 1:1, which cor- layer thickness was evaluated in each condition by growing
responds to a single complete layer deposition followed by a 10 layers with constant parameters and then (after the ex-
single sector deposition so that n/m = 2. periments) disassembling the developed structures and
With the proposed strategy (B), after finishing some cycles measuring the thickness of the layers at four points for each
of deposition in a certain direction (Fig. 5A), it is necessary to coil. As expected, a lower temperature (180C) and lower
stop the feeding of the raw material before stopping the applied pressure (1 kg weight on the tip) give a higher layer
plotting motion (Fig. 5B). This provides an empty gap in thickness (1.25 mm), with a 1.4 mm average width, while a
front of the nozzle (Fig. 5C) that allows the easy exit of the higher temperature (200C) combined with a higher pressure
material for plotting in an inverse direction (Fig. 5D–F) (6 kg) gives a lower layer thickness (0.55 mm) and 2.9 mm
(Supplementary Video S1; Supplementary Data are available average width.
online at www.liebertpub.com/soro). The same strategy is We evaluated the robotic root bending capabilities and the
used for the differential deposition during bending that re- consequent curvature radius by measuring the tip bending
quires the continuous inversion of the plotting orientation to angle and related speed using an embedded accelerometer.
obtain complete and incomplete layers. The relevant pa- Experiments were performed for different deposition ratios
rameters to accomplish the growing task are the central de- (K = 8:1, 4:1, 2:1, and 1:1), with a 1 kg axial load and 200C
position angle (h); bending direction (h + 180); amplitude of filament extrusion temperature. We performed tests 10 times
the angle at which the filament is applied differentially, for each K and recorded the final angle of each test after two
named the differential deposition sector (DD); the angle be- inverse bending sequences. Each direct sequence is com-
fore the inversion process between the stop of the feeding and posed of 1, 2, 4, or 8 cycles of straight growth, a half cycle for
the stop of the plotting, named the extra angle (EA); and the differential deposition, and a 5-s delay for cooling, while the
deposition ratio (K) (Fig. 5). same amount of straight deposition in the inverse direction, a

Table 1. Results of Bending Experiments with Different Deposition Ratios

Deposition Average bending Bending
ratio (K) n/m speed (/min) speed, (SD) hd (mm/min) rc (mm) rc (SD)
1:1 2 1.28 0.15 2.24 101.44 12.14
2:1 1.5 1.08 0.14 2.37 127.31 18.81
4:1 1.25 0.75 0.12 2.47 193.86 30.72
8:1 1.125 0.36 0.06 2.54 394.90 57.06
218 SADEGHI ET AL.

Table 2. Robot Main Characteristics

and Performance
Characteristics Values
Generated maximum axial force 5/6 kg
Generated maximum axial pressure 37 kPa
Minimum bending radius 100 mm
Maximum bending speed 1.28/min
One-layer deposition time 18 s
Layer thickness 0.55–1.2 mm
Growing speed 1.8–4 mm/min
Extrusion temperature 180–200C
Power consumption *14 W
Growing unit weight 105 g

FIG. 7. Result of bending experiment in air under a con- time to perform one cycle deposition (18 s with the set pa-
stant axial load compared with model prediction; the graph rameters), and tdelay is the cooling time before the inversion
presents the achievable curvatures by different settings of
the deposition ratio (K). (5 s during the experiment). Applying equation (8) to the
average measured bending speeds, we obtain the resulting
half-cycle deposition, and the same cooling time compose the curvature radius shown in Table 1. The comparison between
inverse sequence. the measured curvature radius and that resulting from the
Table 1 reports the results, which show a maximum bending model is presented in Figure 7.
speed of 1.28/min with K = 1:1 and 1.08, 0.75, and 0.36/min Finally, the growing robotic root was tested in an artificial
for K = 2:1, 4:1, and 8:1, respectively. We correlated these re- medium performing a constant bending with K = 2:1. The
sults with the proposed model, equation (7B), calculating the robot was able to realize a curve of almost 90 in 1 h and
corresponding curvature radius of the built structure. In par- 40 min. The measured angle is close to a straight line with a
ticular, the angle of the structure b can be calculated as follows: slope of 0.92/min (–0.04/min on three trials). Figure 8
shows the comparison of the bending angle between soil and
hd air experiments as a function of time.
sin b ¼ (8) The robot characteristics are summarized in Table 2.
rc

where hd is the average height of the structure deposited in the Fabrication of the robot prototype
unit time and can be expressed as
The design phase of the robotic root considered a reduction
 1
 of the tip surface area as much as allowed by the fabrication
ncycle þ 2 hl techniques and the selected components. This is required to
hd ¼   (9)
ncycle þ 12 tcycle þ tdelay decrease the pressure needed for soil penetration in primis
and to improve the system capability in avoiding obstacles
Here, ncycle is the number of complete deposited layers (8, and passing in narrow spaces as well as to reduce the filament
4, 2, and 1 in our test), hl is the thickness of the deposited reservoir capacity. The current version of the developed 3D
layer (0.87 mm from the previous experiment with a 200C printer-based growing mechanism has an external diameter
extrusion temperature and 1 kg applied pressure), tcycle is the of 50 mm and a maximum length of 63 mm (Fig. 9).
The deposition head, motor holder disk, and guiding tube
were made of Teflon (PTFE) to prevent heat transmission and
avoid a high thermal strength. The heater was made of a
machinable glass ceramic (Macor) and nickel/chrome wire.
The internal gear was made of aluminum by a lathe and wire-
cut machining. The flexible fingers were cut from a spring steel
sheet with a 0.1 mm thickness. The flat bearing was made
using bearing spheres of 2 mm diameter. The DC motors used
for the feeder and plotter are Micro Metal Gearmotors (from
Pololu Corporation) with a gear ratio of 1000:1 for the plotter
motor and 298:1 for the feeder motor. Two Hall-effect en-
coders regulate the speed of the feeding and plotting motors.
Commercial PLA filaments with a diameter of 1.7 mm were
used for creating the 3D printed root tubular body (Fig. 9).

Robot control architecture

The control unit of the robotic root (based on a PIC
FIG. 8. Bending angle comparison between growing in air 32MX340F512H microcontroller from Microchip, Inc.) is
and growing in soil with the same deposition ratio (K = 2:1). assembled on the surface of the deposition head between the
GROWING ROBOTS BY ADDITIVE MANUFACTURING 219

FIG. 9. Views of the growing robot prototype. (A) Bottom view of the growing mechanism: (A1) magnetic encoder and (A2)
bearing used for decoupling the sensorized tip. (B) Sensorized tip: (B1) magnet for tracking the relative tip rotation and (B2) tip
shaft mounted inside the ball bearing. (C) Front view of the growing mechanism with the sensorized tip: (C1) plotter motor;
(C2) feeder motor; (C3) flexible metallic fingers; (C4) extruder nozzle; (C5) filament of raw material; and (C6) power and
communication lines. (D) A view of a tubular-printed structure with the growing mechanism inside: (D1) printed tubular body
made of PLA; (D2) deposition head; and (D3) growing mechanism inside the tubular body. PLA, polylactic acid.

growing mechanism and the sensorized tip. The tip is dedi- maintains the optimal values for all the parameters influ-
cated to data acquisition and behavior implementation by encing the deposition process (i.e., feeding speed, heating
defining the growth direction on the basis of received external temperature, and plotting speed and orientation).
stimuli.33 The growth direction is then sent to the growing The general architecture of the robot is reported in Fig-
mechanism control unit, where a low-level control algorithm ure 10. A closed-loop control is used to manage the tem-
manages the 3D printer parameters. In particular, it sets and perature of the heater and the speeds of the feeder and plotting

FIG. 10. Control robot ar-

chitecture: a gateway/trunk
board is used to collect data
from the robotic roots and
exchange data with an ex-
ternal PC. The control unit is
used in the growing mecha-
nism to manage the deposi-
tion process; the electronic
board is used in the sensor-
ized tip to acquire data from
the embedded sensors.
220 SADEGHI ET AL.

motors by using a temperature sensor (NTC G1560 100K

Thermistor from TDK/EPCOS) and magnetic encoders
( Magnetic Encoder 12 CPR from Pololu, Inc.). A direct
connection of the heater to the control unit can increase the
temperature of this module. Therefore, a second temperature
sensor (TMP123 from Texas Instruments) directly assembled
on the electronic board is used to monitor the overall tem-
perature of the system and stop the process in the case of
overheating. Moreover, to receive the environmental stimuli
with the correct orientation, it is essential to avoid the rotation
of the tip in soil by decoupling the sensorized tip from the
deposition head.
For this reason, the tip is assembled on the deposition
head by a bearing that enables its independent rotational
movements. A magnetic encoder (AMS5055 from AMS
AG) mounted at the center of the embedded electronic
board (Fig. 10) tracks the relative position of the deposition
head and, in particular, of the extruder nozzle, with respect
to a reference magnet (SM-04x04-N-D from magnets4you
GmbH) integrated at the center of the sensorized tip. The
communication between the sensorized tip and the growing
mechanism is managed wirelessly by an RF module (RFD21733
from RF Digital Wireless).

Materials and Methods

Experimental setups and methods

We tested the working parameters, such as growing speed,

minimum bending radius, thickness of layers, and maximum
pushing force, and the overall behavior of the growing robot
in two different media (i.e., air and granular soil). As gravity
may affect the deposition process, all the tests in air were
performed in an upward direction. The high-level control
algorithm (responsible to make the decision about the grow- FIG. 11. (A) An overall view of the lifting setup for
ing direction) was bypassed, and the commands were sent evaluating the root structure in air under different loads.
directly by a PC to manually select the growing direction and (B–D) Weights from 1 to 6 kg were applied to the system to
the deposition parameters. We used an external IR thermal evaluate the maximum axial force that the robot can exert.
camera (A325sc; FLIR Systems, 60 Wilsonville, OR) to
verify the effective extrusion temperature and cooling time of
the structure (Supplementary Video S1). system can overcome; and (3) evaluate the thickness of the
In the following, we describe the experiments performed deposited layers under different applied pressures and tem-
with the deposition and feeding processes set at the same peratures (Fig. 11 and Supplementary Video S2).
speed (8 mm/s), which allows the robotic root to perform a
complete deposition circle in *18 s. Maintaining deposition Robotic root bending in air. The bending capabilities of
and feeding at the same speed allows avoiding filament the system were evaluated under particular constant loads.
stretching, which results in a more uniform deposition of the During the bending, both the tip position and angle contin-
material with a consequent greater reproducibility in the root uously change, so we cannot apply a constant load as in the
body construction. case of the straight growth tests, described in the previous
section. We therefore developed a setup based on a pulley
Robotic root straight growth in air. We performed tests of (280 mm in diameter) with a rope to add weights (Fig. 12).
the robot’s straight growth in air to provide visual feedback of The robot tip is linked to the shaft of the pulley by means of a
its performance. We simulated a soil with virtual constant solid sliding bar (180 mm length) and a rotary joint. The
pressures by applying different weights to the top of the ro- rotary and sliding point of contact permits the tip to follow its
botic root (i.e., root tip facing upward) while it was growing. commanded directions without the disturbance of any lateral
Straight growth was implemented by the deposition of an loads. By adding constant loads to the rope, a constant axial
integer number of layers followed by the deposition of the load is transferred to the tip.
same number in the opposite direction (to avoid filament The effect of gravity on the deposition process can be
torsion) with weights from 1 to 6 kg on the top of the growing negligible for two main reasons: the system is lightweight
mechanism and with a heater temperature from 180 to 200C. (160 g tip included) and at the same time it is internally an-
These tests were performed to (1) visualize the robot func- chored to the tubular structure that does not permit gravity to
tionalities; (2) find the maximum axial pressure that the affect the deposition process. These tests allowed evaluating
GROWING ROBOTS BY ADDITIVE MANUFACTURING 221

FIG. 12. (A) An overall

view of the experimental set-
up for characterizing the ro-
bot bending by differential
growth. (B) A constant torque
applied to the pulley transmits
a constant axial load to the tip
during the bending. (CE) Se-
quential images of growing
and bending movements. The
robot starts bending on the
right, and after *20 min, it
bends on the left, simulating,
for example, the avoidance of
an obstacle.

the bending angle at different deposition ratios (K). The speed. The tests were performed in a cubic container of
bending angle was measured using an accelerometer inte- 500 · 1000 mm with a 300 mm depth. The root was fixed to a
grated at the tip under an axial load of 1 kg, 200C extrusion side of the soil container (Fig. 13A), and an accelerometer
temperature, and K of 8:1, 4:1, 2:1, and 1:1 (Supplementary embedded in the root tip tracked the tip bending during the
Video S3). experiment. After each experiment, the soil was removed
(Fig. 13B) to verify the effective position of the tip and the
Robotic root growing in an artificial soil. We performed shape of the built root (Supplementary Video S4).
robotic root growing tests in an artificial granular medium made
of polyoxymethylene (POM; Ultraform N2320 003; BASF,
Discussion
Ludwigshafen, Germany) plastic beads (diameter = 4 mm). We
selected POM for our trials in soil because it shows low moisture In our previous work,26,27 we partially translated the effi-
absorption (0.2% under experimental conditions at T = 25C and ciency of natural root movement, when growing in soil, to
40% relative humidity) and thus guarantees repeatable results. artificial devices. Although these seminal works introduced
The mechanical strength of this medium was previously for the first time the concept of a robot that self-creates its
tested using standard methods.27 In this medium, we measured body through a layer-by-layer deposition, the manufacturing
an average maximum force of 38.4 N (–4.3 N standard devia- process and material used did not allow the implementation
tion on nine trials) to obtain a vertical penetration to a depth of more complex behaviors, such as bending in air and soil
of 200 mm, with a tip diameter of 50 mm and a 60 mm/min following or escaping from external stimuli or avoiding

FIG. 13. Setup for testing the robotic root growth in a granular soil. (A) A container (500 · 1000 · 300 mm3) was filled
with POM, and the robot was fixed to the container wall. (B) View of the robotic root at the end of the growth process. The
exact position of the robot was evaluated by removing the soil granules. POM, polyoxymethylene.
222 SADEGHI ET AL.

obstacles. In the present work, we introduced a customized different from that obtained in air, where under the same
FDM 3D printer-based mechanism inside the robot root body. conditions, the bending speed was 1.1/min. This difference is
This manufacturing solution, coupled with the choice of us- mainly due to the effect of the lateral pressure present in soil
ing the commercially available PLA as the deposition ma- that must be overcome for the whole tip to perform bending.
terial, allowed us to obtain a robot that can build its own This results in a decrement of the effective bending speed.
structure by a fused deposition process of new layers and, at Moreover, the bending tests performed in the artificial soil
the same time, grow and bend. We demonstrated such root- showed a very high linearity with respect to that obtained in
like robot abilities both in air, by applying external loads to air (Fig. 8). This result could be due to the peripheral support
the tip, and in an artificial medium. that the soil can provide for the printed structure, while in air,
We found that a lower temperature (180) results in a higher the structure grows like a beam under an axial load, even if
layer thickness and consequently higher growing speed, while the load at the tip was axially applied.
a higher temperature (200) results in a more strongly built The velocity of the system, in addition to being governed by
structure. We found that the thickness of the root tubular the deposition parameters and environmental conditions, was
structure, in addition to being caused by the settings of the also limited by the overheating of the controller unit and me-
deposition process, is also the result of external forces applied chanical components. Even if the heater is insulated in Kapton
to the root. The larger the external force, the thinner the layer tape and encapsulated in a PTFE structure, after some minutes
thickness and wider the layer width. This means that, de- (approximately 10–15 min, depending also on the environmental
pending on the soil compactness and extrusion temperature, the conditions), the heat is transferred to other parts of the system,
system can automatically grow faster in loose soils and slower creating problems in the electronic components (when the
in harder soil while having a thicker root wall and stronger temperature exceeds 90C) and mainly in the feeding part,
structure for overcoming the soil pressure in harder soils. where the filament becomes prematurely soft and thereafter
Therefore, tuning the temperature on the basis of the soil im- cannot be pushed properly into the extruder.
pedance could be a strategy to optimize the growth speed. To avoid this problem, we adopted two solutions: (1) a
A similar swelling behavior is observed in living roots system shutdown every 10 min for approximately the same
facing with hard soils: a plant root expands its diameter in amount of time to permit the root to autonomously decrease
harder soils while its penetration speed decreases; vice versa, the whole temperature and (2) an air cooling system located
in loose soils a root grows faster with smaller diameter.35 at the top of the root, where the amount of air is regulated to
Some hypotheses explain that the radial expansion of im- maintain a root temperature on the order of 75, which is a
peded root apexes can reduce the axial stress on the root working temperature (experimentally verified) that assures
cap,36 and the root swelling produces soil tensile failure that long and stable system operation. This problem can also be
may propagate a crack ahead of the root apex. The generated overcome by using polymers with a lower melting tempera-
crack minimizes resistance to the elongating root resulting in ture shaped in the form of 3D printer filaments.
an accelerated elongation into this zone of reduced soil This work paves the way for a new generation of robots to
strength.37,38 The developed system can overcome a 6 kg grow. A distinctive aspect of this robot is its mimicking of the
axial load (which is equivalent to 0.37 atm of pressure with movement ability of plant roots by growing. In general, dif-
this root dimension). On increasing the weight, the growing ferent types of additive manufacturing solutions can be used to
mechanism begins to become unstable due to the increasing develop and grow robots, including the delivery of raw ma-
friction between the deposition head and deposited layers. terial to the growing zone (root tip), the plotting strategy, and
This capacity is not the limitation of the proposed concept the fusion methods. The robotic root tubular structure gener-
and it can be overcome using stronger DC motors, since they are ated by the growing process can be useful for the transmission
commercially available. In real soil, the system would be able to of building material(s) and energy-supplier solutions to the
penetrate till the environmental pressure is less than the maxi- growing root. This tubular structure can also provide inter-
mum pressure that the system can generate. Effective penetra- esting solutions for several applications, including transmitting
tion depends on soil types. A positive aspect is that normally the data from a sensorized tip in soil-monitoring tasks and passing
soil pressure increases till a certain depth (few tens of centi- oxygen, drugs, or food in rescue scenarios and cameras or
meters) and after that the interaction of soil granules prevents surgical tools for medical applications.
the pressure transmission from upper to lower layers.39 How-
ever, there are other technical issues that limit the current robot Acknowledgment
version in performing deep penetration (e.g., crack of PLA fil-
ament, longer power lines that cause an increasing loss of en- This study was partially founded by the PLANTOID pro-
ergy, and friction of the filament along the wall structure that ject (EU-FP7-FETOpen grant no. 29343).
increases the resisting force to the penetration).
The robot was able to successfully exhibit steering be- Author Disclosure Statement
havior by bending both in air and artificial soil. In air, for
No competing financial interests exist.
smooth curvatures (K = 8:1, 4:1, and 2:1), the model and the
experimental data fit (Fig. 7). However, with a higher K
(1:1), the effective bending is less than the predicted value. This References
is due to the mechanical constraints because this ratio over- 1. Pfeifer R, Lungarella M, Iida F. Self-organization, em-
comes the minimum bending radius of the structure (100 mm), bodiment, and biologically inspired robotics. Science 2007;
and the internal contact/friction of the growing mechanism 318:1088–1093.
components with the structure decreases the effective bending 2. Kim S, Laschi C, Trimmer B. Soft robotics: a bioinspired
of the tip. In soil, the bending speed obtained was 0.92/min evolution in robotics. Trends Biotechnol 2013;31:287–294.
GROWING ROBOTS BY ADDITIVE MANUFACTURING 223

3. Laschi C, Cianchetti M. Soft robotics: new perspectives for ro- Robotics and Automation (ICRA), 2013 IEEE International
bot bodyware and control. Front Bioeng Biotechnol 2014;2:3. Conference 2013;3457–3462.
4. Rus D, Tolley MT. Design, fabrication and control of soft 27. Sadeghi A, Tonazzini A, Popova L, Mazzolai B. A novel
robots. Nature 2015;521:467–475. growing device inspired by plant root soil penetration be-
5. Mazzolai B, Mattoli V. Robotics: generation soft. Nature haviors. PloS One 2014;9:e90139.
2016;536:400–401. 28. Baluška F, Mancuso S, Volkmann D, Barlow PW. Root
6. Wang L, Iida F. Deformation in soft-matter robotics: a apex transition zone: a signaling-response nexus in the root.
categorization and quantitative characterization. IEEE Ro- Trends Plant Sci 2010;15:402–408.
bot Autom Mag 2015;22:125–139. 29. Blancaflor EB, Masson PH. Plant gravitropism. Unraveling
7. Manti M, Cacucciolo V, Cianchetti M. Stiffening in soft the ups and downs of a complex process. Plant Physiol
robotics: a review of the state of the art. IEEE Robot Autom 2003;133:1677–1690.
Mag 2016;23:93–106. 30. Hart JW. Gravitropism. Plant tropisms and other growth
8. Robinson G, Davies JBC. Continuum robots–a state of the movements. Springer Science & Business Media, 1990;44:89.
art. In: Robotics and Automation 1999, IEEE International 31. Takahashi N, Yamazaki Y, Kobayashi A, Higashitani A,
Conference 1999;4:2849–2854. Takahashi H. Hydrotropism interacts with gravitropism by
9. Trivedi D, Rahn CD, Kier WM, Walker ID. Soft robotics: degrading amyloplasts in seedling roots of Arabidopsis and
Biological inspiration, state of the art, and future research. radish. Plant Physiol 2003;132:805.
Appl Bionics Biomech 2008;5:99–117. 32. Ding JP, Pickard BG. Modulation of mechanosensitive
10. Walker ID. Continuous backbone ‘‘continuum’’ robot ma- calcium-selective cation channels by temperature. Plant J
nipulators. ISRN Robotics 2013;1–19. 1993;3:713–720.
11. Heckele M Schomburg WK. Review on micro molding 33. Sadeghi A, Mondini A, Del Dottore E, Mattoli V, Beccai L,
of thermoplastic polymers. J Micromech Microeng 2003; Taccola S, et al. A plant-inspired robot with soft differ-
14:R1. ential bending capabilities. Bioinspir Biomim 2016;12:
12. Qin D, Xia Y, Whitesides GM. Soft lithography for micro- 015001.
and nanoscale patterning. Nat Protoc 2010;5:491–502. 34. Taccola S, Greco F, Sinibaldi E, Mondini A, Mazzolai B,
13. Wu W, DeConinck A, Lewis JA. Omnidirectional printing Mattoli V. Toward a new generation of electrically con-
of 3D microvascular networks. Adv Mater 2011;23:24. trollable hygromorphic soft actuators. Adv Mater 2015;27:
14. Wang L, Brodbeck L, Iida F. Mechanics and energetics in 1668–1675.
tool manufacture and use: a synthetic approach. J R Soc 35. Bengough AG, Bransby MF, Hans J, McKenna SJ, Roberts
Interface 2014;11:20140827. TJ, Valentine TA. Root responses to soil physical condi-
15. Laschi C, Mazzolai B. Lessons from animals and plants: tions; growth dynamics from field to cell. J Exp Bot 2006;
the symbiosis of morphological computation and soft ro- 57:437–447.
botics. IEEE Robot Autom Mag 2016;23:107–114. 36. Bengough AG, Mullins CE. Mechanical impedance to root
16. Laschi C, Mazzolai B, Cianchetti C. Soft robotics: tech- growth: a review of experimental techniques and root
nologies and systems pushing the boundaries of robot growth responses. Eur J Soil Sci 1990;41:341–358.
abilities. Sci Robotics 2016;1:eaah3690. 37. Materechera SA, Dexter AR, Alston ML. Penetration of
17. Wang L, Culha U, Iida F. A dragline-forming mobile robot very strong soils by seedling roots of different plant spe-
inspired by spiders. Bioinspir Biomim 2014;9:016006. cies. Plant Soil 1991;135:31–41.
18. Hirose S, Hiroya Y. Snake-like robots. IEEE Robot Autom 38. Popova L, Van Dusschoten D, Nagel KA, Fiorani F,
Mag 2009;16:88–98. Mazzolai B. Plant root tortuosity: an indicator of root path
19. Kotay K, Rus D. The inchworm robot: a multi-functional formation in soil with different composition and density.
system. Auton Robots 2000;8:53–69. Ann Bot 2016;118:685–698.
20. Menciassi A, Dario P. Bio-inspired solutions for locomotion 39. Huang W, Sheng D, Sloan SW, Yu HS. Finite element
in the gastrointestinal tract: background and perspectives. analysis of cone penetration in cohesionless soil. Comput
Philos Trans A Math Phys Eng Sci 2003;361:2287–2298. Geotech 2004;31:517–528.
21. Phee L, Accoto D, Menciassi A, Stefanini C, Carrozza MC,
Address correspondence to:
Dario, P. Analysis and development of locomotion devices
Ali Sadeghi
for the gastrointestinal tract. IEEE Trans Biomed Eng 2002;
49:613–616. Center for Micro-BioRobotics (CMBR)
22. Lim J, Park H, An J, Hong YS, Kim B, Yi BJ. One Istituto Italiano di Tecnologia (IIT)
pneumatic line based inchworm-like micro robot for half- Viale Rinaldo Piaggio
inch pipe inspection. Mechatronics 2008;18:315–322. 34-56025 Pontedera (PI)
23. Koh JS, Cho KJ. Omegabot: Biomimetic inchworm robot Italy
using SMA coil actuator and smart composite microstruc-
tures (SCM). In: Robotics and Biomimetics (ROBIO), 2009 E-mail: ali.sadeghi@iit.it
IEEE International Conference 2009;1154–1159.
24. Omori H, Nakamura T, Yada T. An underground explorer Barbara Mazzolai
robot based on peristaltic crawling of earthworms. Ind Center for Micro-BioRobotics (CMBR)
Robot 2009;36:358–364. Istituto Italiano di Tecnologia (IIT)
25. Mazzolai B, Beccai L, Mattoli V. Plants as model in bio- Viale Rinaldo Piaggio
mimetics and biorobotics: new perspectives. Front Bioeng 34-56025 Pontedera (PI)
Biotechnol 2014;2:2. Italy
26. Sadeghi A, Tonazzini A, Popova L, Mazzolai B. Robotic
mechanism for soil penetration inspired by plant root. In E-mail: barbara.mazzolai@iit.it