Hindawi Publishing Corporation

ISRN Robotics
Volume 2013, Article ID 726506, 19 pages
http://dx.doi.org/10.5402/2013/726506

Review Article
Continuous Backbone ‘‘Continuum’’ Robot Manipulators

Ian D. Walker
Department of Electrical and Computer Engineering, Clemson University, Clemson, SC 29634, USA

Correspondence should be addressed to Ian D. Walker; [email protected]

Received 8 May 2013; Accepted 4 June 2013

Academic Editors: J. Archibald and J. B. Koeneman

Copyright © 2013 Ian D. Walker. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This paper describes and discusses the history and state of the art of continuous backbone robot manipulators. Also known as
continuum manipulators, these robots, which resemble biological trunks and tentacles, offer capabilities beyond the scope of
traditional rigid-link manipulators. They are able to adapt their shape to navigate through complex environments and grasp a wide
variety of payloads using their compliant backbones. In this paper, we review the current state of knowledge in the field, focusing
particularly on kinematic and dynamic models for continuum robots. We discuss the relationships of these robots and their models
to their counterparts in conventional rigid-link robots. Ongoing research and future developments in the field are discussed.

1. Introduction discuss in this paper, is to create a robot with a continuous

form or backbone. These robots, termed continuum robots,
Robotics as a field is still in its formative stages. Designers can be viewed as being “invertebrate” robots, as compared
of robots are continuing to explore the range of possibilities with the “vertebrate” design of conventional rigid-link robots.
for robot structures which can sense and perceive, navigate Continuum robots can bend (and often extend/contract
and locomote, as well as grasp and manipulate. The creation and sometimes twist) at any point along their structure.
of programmable manipulators can be traced to be very This provides them with capabilities beyond the scope of
beginning of robotics as a discipline [1]. To date, robot their rigid-link counterparts. An example (in the following
manipulators remain the core product of the field, being paragraph) serves to illustrate this.
productively and profitably deployed in industrial settings In Figure 1, a continuum robot gently contacts and adapts
worldwide. its shape to wrap around a fragile object (a glass lampshade).
However, when moving outside the highly structured Subsequent to this image, the robot lifted and performed
world of industry, especially the factory floor, traditional “whole arm manipulation” [2] of the object, without causing
rigid-link manipulators have been less successful. Their rigid- breakage or damage. This was achieved with no prior plan-
link structure (while excellent for precise positioning of their ning or knowledge of the object and without any sensing
end effector) tends to be the cause of unwanted collisions of the contact between robot and object. This would be an
when not in workcell environments specially engineered to extremely difficult operation for a conventional rigid-link
maintain open spaces for their movements. Their inability robot manipulator. Active force sensing would be required to
to grasp objects other than at their end effector significantly present a sufficiently compliant interface to avoid breakage
restricts their manipulation capabilities beyond those of of the object. Further, note that even grasping such an
objects preengineered to fit their end effectors. Consequently, object would require significant a priori specialized tooling or
in real-world environments and situations not prechore- fixturing, as the object geometry is inherently unsuitable for
ographed, it is generally nontrivial, and often not possible, to grasping with a parallel jaw gripper. However, the continuum
deploy rigid-link manipulators. robot was able to use its compliant, actively controlled
Robot manipulators do not, however, have to be formed continuous structure to perform the operation easily, quickly,
from rigid-links. An alternative design possibility, which we and without a priori knowledge or task engineering.
2 ISRN Robotics

group of contributions, coined the term “continuum robot”

in 1999 [15].
In parallel with the above developments in hardware,
the 1990s saw progress and innovation in continuum robot
modeling, particularly in kinematics. In a series of papers
[16–18], “bottom-up” continuum robot kinematics models
were developed. These works began with the constraints of
physical continuum robot backbones and formulated back-
bone kinematics from them. It can be seen (this is reviewed
in detail in Section 3) that the models resulting from these
“bottom-up” approaches can be formed from the initially
published “top-down” theory of [4, 5]. The new models, being
Figure 1: A continuum robot grasping and manipulating a fragile matched directly to hardware constraints, enabled model-
(glass) object, using its compliant backbone to adapt to the object’s based implementations and thus real-time computer control
shape. of continuum robot shapes.
The late 1990s and the 2000s have seen a significant
increase in the number of researchers in continuum robots
The above example of compliant whole arm grasping and a corresponding increase in the number and breadth
demonstrates one novel and appealing feature offered by con- of published papers on the topic. In this paper, we review
tinuum robots, compared to conventional rigid-link robots. and discuss the current state of the art in the field of
Another key feature (and dual to the previous one above) is continuum robots. Summaries of early developments in the
the potential to use the backbone to keep away from, rather field have appeared [19, 20]. In this paper, we summarize
than contract and grasp, complex environmental geometries. the early work and update for the many recent developments
In this way, continuum backbones (note that the term back- in this rapidly emerging field. We begin in Section 2 by
bone is used for continuum robots despite their invertebrate discussing the underlying design principles of continuum
nature) can “twist and turn” to negotiate very tight spaces, robots and classify continuum robot designs to date into three
thus penetrating areas where conventional robots would be main design types. Section 3 provides an overview of the
unable to enter or would get their links stuck in. This novel kinematics of continuum robots. The dynamics and control
potential and the attempt to realize that potential in useful of continuum robots are reviewed and discussed in Section 4.
hardware remain at the core of the field of continuum robots. Ongoing research in the field is summarized in Section 5.
Historically, interest in continuous backbone robots Conclusions follow in Section 6.
began as early as the 1960s. As for the first continuum robot,
Anderson and Horn’s Tensor Arm [3] (discussed further in
Section 2) appears to have been the first prototype reported in 2. Design Principles
the literature. Intended for underwater applications, the pro-
totype did not, however, progress from the laboratory stage. The defining feature of a continuum robot is a contin-
It was quickly realized that, while early prototypes could be uously curving core structure, or backbone, whose shape
made to achieve a wide range of shapes, the relationship can be actuated in some way. An almost universal addi-
between the shapes and inputs was highly complex—much tional property is that the backbone is compliant; that is, it
more than for rigid-link robots and certainly challenging for yields smoothly to externally applied loads. Together, these
implementations using the computing environments of the properties enable the physical capabilities which motivate
time. continuum robots: to adapt the backbone shape to maneuver
There followed a lull in activities in the area until the late the robot within more complex environments and to conform
1980s and early 1990s, when two key developments occurred. to grasp a wider class of objects than feasible with rigid-link
Inspired by the need to approximate hyperredundant (rigid- robots.
link) robot backbones, Chirikjian published the first works The design space available to achieve the above properties
[4, 5] providing general continuous backbone kinematics is very large. For example, the backbone core does not
and dynamics for robots. Inspired by the example of snake even have to be continuous. Snakes present the external
locomotion, Hirose [6] published innovative results on the appearance of having a continuous structure, but are verte-
evolution of continuous backbone shapes. (Hirose also pro- brates, with an internal segmented backbone comprised of
duced continuum robot prototypes [6]). Together, the above (many very small) rigid-links. Robots with segmented rigid-
efforts are the seminal work establishing the field. link interior backbones presenting a continuous external
Further efforts in continuous backbone robot hardware form have been developed [8, 21–23]. These are sometimes
in the 1990s centered in two directions: (1) extension of the termed “continuum-style” robots. However, such designs
original tendon-driven design concepts of Anderson/Horn are rare, and almost all designers have sought to create
and Hirose aimed at practical implementation [7, 8] and truly continuous backbone structures. The most significant
applications [9–11] and (2) the new innovation of creating exceptions are the “snake-arms” [24] of OC Robotics [25], the
backbones with pneumatically actuated chambers [12–14]. only continuum-style robot currently commercially available.
Robinson and Davies, codevelopers of some of the second These robot arms, as the name suggests, are composed of
ISRN Robotics 3

serially connected modular rigid-link sections. While not

truly continuum, with enough modules, the form resembles
a continuous backbone. They have been deployed in nuclear
reactors and inside airframes, among other applications
[25].
Design of robot structures in the absence of rigid ele-
ments is an unfamiliar process for most robotics design-
ers. However, several fundamental design principles can be
identified by a careful study of biological “tongues, trunks,
and tentacles” [26–28]. In particular, the group of structures
Figure 2: The Tendril, a tendon-driven continuum robot.
termed “muscular hydrostats” [29], which includes octopus
arms, elephant trunks, squid tentacles, and mammalian
tongues, has provided a rich vein of insight for continuum
robot designers. Animals do not have to be the only source of
inspiration; the vines and tendrils of plants [30] are a source The first published example of a continuum robot, the
of inspiration also [31, 32]. “Tensor Arm” [3], is a good example of a tendon-based
Muscular hydrostats are structures comprised almost design. Tendons, routed through spacer elements, were used
entirely of their own actuators (muscle), with some additional to effect bending of the core backbone element in several
fluid and connective tissue. They can typically bend and “sections.” The termination points of sets of tendons along the
twist and often extend to some extent at any point along backbone define the sections; see Figure 2.
their structure. The muscles are arranged in oriented arrays One choice for the core backbone element is a com-
(longitudinal, transverse, and oblique) in a way that enables pressible spring. An early spring-backbone-based design was
both the motive force and structural support for bending, produced by Hirose [6]. The more recent long thin “Tendril”
extension, and torsion to be provided by the muscles [29]. continuum robot by NASA Johnson Space Center [31] is also
Some initial work [33] aimed to mimic (albeit at a much based on a spring backbone (Figure 2). The spring backbone
less detailed scale) the muscular hydrostat design concept in provides natural compliance. However, this also makes, the
continuum robots, using various artificial muscle technolo- designs difficult to control, as control effort intended for
gies. However, practical continuum robots require not only backbone bending is lost in compression. The same is true
significant bending but also high force generation, and the for tendon-actuated pneumatic backbones (Figure 3).
state of the art in artificial muscle technology (both at the time A simple solution to the problem of uncontrolled com-
of publication of [33] and at the time of writing this paper) pression is to use a flexible incompressible rod as the
was not capable of satisfying these needs at scales suitable for backbone element [7]; see Figure 4. This approach has several
continuum robots. If, at some future time, artificial muscle advantages including a slender low-profile backbone and
technologies advance sufficiently, the possibilities for design more predictable behavior. The disadvantage of course is
and operation of continuum robots could be revolutionized that this approach precludes the incorporation of backbone
[34]. Note that issues of packaging, power consumption, extension. However, the incompressible backbone concept
wiring, and so forth would remain major challenges. has proved a popular and successful design, with numerous
In the absence of technologies which can easily mimic the implementations based on it [36–39].
key biological inspirations, designers have followed several Tendon-based continuum designs share the following
alternative paths. The basic requirements are to produce general features: (1) the backbone shape resolves into a
active bending and ideally also some extension and local finite series of “sections” whose end points are defined by
torsion, of continuous backbone structure which possesses the tendon termination points along the backbone; (2) the
some helpful (predesigned) internal energy properties. Three forces achievable with the device are relatively high (tendons
alternative fundamental design strategies have emerged. Each generate relatively high forces); (3) some method must be
strategy and notable continuum robots constructed to date found to prevent slack [17] and backlash [36] in the tendons;
using it are summarized in the following subsections. and (4) the design requires a relatively bulky actuator (motor)
package at the base of the robot. With respect to (3), most
implementations either actively actuate all tendons [40] or
2.1. Tendon-Based Designs. Perhaps the most direct approach use a single actuator to actuate antagonistic tendon pairs,
to bending a continuous structure is the use of remotely with a spring mechanism to take up the slack [17]. With
actuated tendons. Given a backbone which, in the absence respect to (4), the location of the actuator package outside the
of external loads, consistently attains a given shape (typically backbone has led to the tendon design being categorized as an
a straight line, though this is not strictly necessary [35]), “extrinsically actuated” continuum robot design [20].
tendons can be used to deviate it from that shape via bending. Tendon-actuated continuum robots have been designed
Tendons are routed along the backbone and terminated in for space operations [31] and, in particular, medical pro-
groups at selected points down it. Forces applied to the cedures [40, 41]. A spring-based tendon-driven backbone
tendons at the base produce torques at the termination continuum robot was developed for sinus surgery in [42]
points, resulting in bending. The design is quite simple and and another developed for ACL surgery in [43]. A system for
(relatively) straightforward to realize in hardware. laryngeal surgery was developed in [44]. A “robot octopus”
4 ISRN Robotics

Figure 5: Concentric tube continuum robot concept.

shape variables. Disadvantages include the need for an

external actuator package and the lack of inherent support
for actively controlled bending.
Concentric tube continuum robots have found a niche
application in the medical field, where their small profile
and high compliance are well suited for minimally invasive
procedures [49, 51, 52]. In this context, they are smaller-scale
Figure 3: Air-Octor Continuum robot. This two-section, six- and lower-force devices than their counterparts constructed
degree-of-freedom robot bends a pneumatically inflated core tube via the other two designs and are sometimes termed “active
using six tendons. cannulas” [53]. For example, in [52], sampling-based motion
planning techniques are used to design a concentric tube
robot specifically for the task of navigation through the
human lung. In [48], an MRI-compatible, piezoelectrically
actuated concentric tube robot is designed for neurosurgery
and percutaneous interventions.

2.3. Locally Actuated Backbone Designs. The third design

type differs from the previous two by including the actua-
Figure 4: Tendon-based continuum robot with incompressible
tors directly in the backbone. Indeed, this type of “locally
flexible rod core.
actuated” continuum robot typically forms the backbone
from its actuators. In this regard, the design is closest to
the biological continuum structures which often motivate
with six cable-actuated limbs has been demonstrated under- continuum robots. This also gives rise to the categorization
water in [45]. of the design as “intrinsically actuated” [20].
Typical locally actuated designs form the backbone from
2.2. Concentric Tube Designs. A second form of extrinsically pneumatic “McKibben” muscles [54, 55], though numerous
actuated continuum robot (and the most recent to emerge) versions using shape memory alloys [56] have also been
is based on a backbone formed by concentric compliant built. The strategy is to form the backbone from indepen-
tubes. The tubes are free to move (translate and rotate) with dently actuated sections. Each section is constructed from
respect to each other (subject to hardware limits) with the (typically) three independently actuated muscles, connected
translations and rotations actuated at the base of the robot. together along their length. The muscles can be “extenders”
The net effect is similar to some telescopes: the structure (increased length as a function of increased pressure) [54,
can extend and contract by translational sliding of the tubes 57] or “contractors” (decreased length as a function of
longitudinally (modulo the length of the tubes, the smaller higher pressure) [54, 58]. See Figure 6 for an example of the
diameter tubes becoming the most distal), and the structure “Octarm” series of locally pneumatically actuated continuum
can achieve local rotation by rotational sliding of the tubes; robots.
see Figure 5. When pressure is evenly increased or decreased in all
The concentric tube design thus directly achieves both three actuators of a straight section, the section length
extension and torsion [46]. However, it does not inherently increases or decreases. When differing pressures are applied
provide for backbone bending. The simplest approach to this to the actuators, the section bends into a segment with
issue is to use precurved compliant tubes [47, 48]. This, approximately constant curvature. The plane of the curve is
when combined with the directly controllable extension and determined by the three pressures. In general, the section
torsion, provides some useful variation in backbone shapes. extends, contracts, and changes its curvature and plane of
Another approach is to use tendons to bend the tubes [49, 50]. curvature as a function of the three applied pressures. The
However, this significantly increases the complexity of the net shape of the backbone is thus a serially connected set
design [32]. of approximately constant curvature segments (with the end
Advantages of the concentric tube design include the tangents coinciding).
inherently clean and thin design (assuming the design with The locally actuated continuum robot design has been
no tendons to bend the tubes) and the fact that the actuator the subject of much research and numerous realizations in
values (unlike with tendons) directly correspond to backbone the recent years. In particular, the high-profile “Octarm”
ISRN Robotics 5

advantage of being able to actively regulate the internal

stiffness of the backbone. However, these designs, in common
with spring-based backbones, suffer from the problem of
uncontrolled longitudinal compliance along the backbone.
Also, pneumatics offer a limited range of possible backbone
stiffnesses.
Alternative approaches to variable (controllable) stiffness
backbones have considered magnetorheological and elec-
trorheological fluid-based actuation [70, 71]. In these mate-
rials, magnetic and electric fields can be used to change the
properties from fluid to stiff or solid. This allows backbones
built from, or strongly biased by, such materials to feature
tunable stiffness.
More recently, the idea of using “jamming” of suitable
media (such as sand or coffee grains) has been exploited for
Figure 6: Intrinsically actuated three-section, nine-degree-of- novel variable stiffness continuum robot design [70, 72, 73].
freedom pneumatic “Octarm” continuum manipulator. The underlying notion is to pack the media in a closed
chamber to bring and vary the internal pressure to “loosen”
the media in a fluid state, or “jam” it into a solid state. For
example, in [73], granular media are used as the jamming
[33, 57, 59, 60] and “European Octopus” [45, 54, 61, 62] element. Initially packed into a chamber under a vacuum,
projects featured continuum robots based on this design. the grains jam and unjam the chamber as a function of
Other realizations of the design include the “Bionic Assistant” applied pressure. It is demonstrated in [73] that, when three
[35, 63], which closely resembles the trunk of an elephant such chambers are combined in parallel with a McKibben air
[28]. Shape memory alloy actuation has been used to steer muscle, a tunable stiffness section element can be produced.
an active cannula for medical procedures, and in [64, 65], A prototype tendon-driven continuum robot using granular
dual shape memory alloy-based backbones are used in a jamming for variable stiffness is demonstrated in [72].
system designed for single port access surgery. Additionally, a In [70], the jamming elements are multiple surface layers,
locally actuated system for endoscopic stitching intended for interleaved in various ways. Negative pressure is used to
surgical obesity treatment is presented in [66], and a design bring the layers together and exploit friction to create tunable
for colonoscopic insertion is described in [67]. stiffness via “layer jamming.” A tubular continuum backbone
Actuator selection for intrinsically actuated continuum is built (locally actuated using shape memory alloy wires
robots can be from any available type of artificial muscle. This as core actuators) and shown to exhibit significant range
could include muscles based on engineered polymers, such as of backbone stiffness. This approach to augmenting core
elastomers, for example. An extensive study of the potential of continuum robot designs with new innovations to enhance
these types of actuators is presented in [34]. However, at this performance shows significant promise for producing the
time, only the pneumatic or hydraulic actuator technologies next generation of continuum manipulators.
feature the combination of bending and force generation
capabilities for continuum robots at the human scale or 2.5. Common Property: Constant Curvature. Notice that,
larger. independent of underlying physical structure, a common
Locally actuated continuum robot designs have the property exhibited by virtually all continuum robots ([35,
key advantage of inherently providing the backbone with 63] being notable exceptions) is that the resulting backbone
extension, bending, and torsion (actually, bending in two approximates a serially connected set of constant curvature
dimensions). This is a feature not directly provided by sections. This arises due to the following: (1) all three previous
either tendons or concentric tubes, as discussed in the design types create a series of serially connected sections;
preceding subsections. Disadvantages of locally actuated (2) internal potential energy in each section is uniformly
designs include relatively low force generation capabilities distributed (unactuated, each section is straight, or bent
(for pneumatically actuated designs at least), fairly complex at a fixed configuration); and thus, within each section,
tube routing/valving, and the need for external pressure internal forces act to drive the unactuated (passive) degrees of
regulation equipment and a compressor. freedom to equalize in value along the section. This produces
internal section bending of constant curvature at any given
2.4. Variable Backbone Stiffness. An interesting choice for moment.
continuum backbones is to use pneumatically actuated tubes. Therefore, in practice, achievable continuum backbone
The KSI Tentacle Manipulator [11] and the Air-Octor robot shapes are (fairly close approximations to) sequentially con-
([68, 69] Figure 3) were each based on tendon-actuated nected segments of circles in three dimensions (with the
extensible pneumatic chambers. This design allows tendon tangents to successive section end points aligned and the
actuation of both bending and extension. Notice, how- arc lengths of the segments corresponding to robot section
ever, that the pneumatically actuated tube design adds the lengths). While the “constant curvature” property is affected
6 ISRN Robotics

by external loading (some sag typically exists due to gravity t(s)

or grasped objects), it remains a good first approximation to
backbone shape and has been strongly exploited in kinematic
x(0) ||x(s)|| x(s) 𝜃
models, as discussed in the next section. 2𝜃
𝜃

3. Kinematics
t(0) s t(0)
In order to coordinate the movements available in continuum
robots, kinematic models, which capture the relationship Figure 7: Geometry of constant curvature in the plane.
between configuration (backbone shape) variables and both
task (e.g., tip) and actuator (e.g., tendon or muscle length),
variables need to be established. Such models form the basis using the conventional D-H approach, for the virtual robot.
for motion planning and control algorithms and are the This approach, first used in [8], has been used numerous
critical step between prototype development and practical times subsequently [18, 36, 39]. The details (D-H table and
implementation of continuum robot hardware. associated homogeneous transformation matrix) are given in
Since continuum robots can change their shape at any Table 1 and (1).
point along their structure, their models necessarily differ Consider
significantly from those of conventional rigid-link robots,
where configuration changes can occur only at a finite num- cos (𝜃1 + 𝜃3 ) − sin (𝜃1 + 𝜃3 ) 0 −𝑑2 sin 𝜃1
[ ]
ber of fixed locations along their structure (the joints between [ sin (𝜃1 + 𝜃3 ) cos (𝜃1 + 𝜃3 ) 0 𝑑2 cos 𝜃1 ]
[ ]
the rigid-links). For rigid-link robots, the well-established [𝐻3 ] = [
0
[
] . (1)
]
Denavit-Hartenberg (D-H) convention [1] provides a general [ 0 0 1 0 ]
underlying framework for the development of kinematic (and [ ]
0 0 0 1
dynamic) models. The D-H convention establishes a local [ ]
coordinate frame fixed in each of the (finite number of) links
and develops the overall kinematics via a sequential series of Useful continuum robot kinematics can now be devel-
frame-to-frame steps, as a function of the (finite) number of oped by noting, as well as substituting in the virtual robot
joint angles [1]. kinematics, relationships between the joint variables for
For continuum robots, the fact that the local shape varies the virtual robot and corresponding configuration space
continuously along the backbone needs to be reflected in variables for the continuous curve. Specifically (see Figure 7),
kinematic models. Two alternative approaches have emerged. 1
The first takes a “bottom-up” strategy, building a continuum 𝜃1 = 𝜃3 = 𝜃, 𝑑2 = ‖𝑥 (𝑠)‖ , 𝑘=( ). (2)
radius
model via exploiting the D-H approach to fit a “virtual” rigid-
link robot to the backbone. The second uses a “top-down” We have (utilizing the underlying geometry)
philosophy, explicitly treating the backbone as a continuous
curve, in order to formulate the models. We review each (2𝜃) (𝜃1 + 𝜃3 )
𝑠 = 𝑟 (2𝜃) = = . (3)
approach and demonstrate how ultimately they lead to the 𝑘 𝑘
same models in the following two subsections. So
(𝜃1 + 𝜃3 ) = 𝑠𝑘. (4)
3.1. Continuum Kinematics via Virtual Rigid-Link Kinematics.
The first (and the most inspired by hardware) approach to Also
continuum robot kinematics strongly exploits the constant
‖𝑥 (𝑠)‖ 𝑑2 sin 𝜃
curvature sections feature possessed by almost all continuum = = 𝑟 sin 𝜃 = . (5)
robots to date. This approach, which first appeared in [8], is 2 2 𝑘
based on the observation that the evolution from one end to So
the other of a constant curvature curve can be represented,
2 sin 𝜃
in the plane of that curve, via three discrete transformations: 𝑑2 = . (6)
(1) a rotation to “point” the tangent at the curve beginning to 𝑘
the curve end point; (2) a translation along the newly aligned Substituting (4) and (6) into the model (1) and simplifying
direction (from curve beginning to end); and (3) a second gives
rotation (of same amount as the first) to realign with the
tangent at the curve’s end; see Figure 7. 1
cos (𝑠𝑘) − sin (𝑠𝑘) 0 ( ) {cos (𝑠𝑘) − 1}
Given this observation, in the plane, a “virtual” three- [ 𝑘 ]
[ 1 ]
joint rigid-link manipulator, with identical (i.e., coupled) [ sin (𝑠𝑘) cos (𝑠𝑘) 0 ( ) sin (𝑠𝑘) ]
[ ]
rotations as its first and third joints and a prismatic joint in [𝐻30 ] = [ 𝑘 ] . (7)
[ ]
the middle, can be used to model the kinematic transforma- [ 0 0 1 0 ]
[ ]
tion along any constant curvature backbone. Consequently, [ ]
it is possible to find the corresponding kinematic model, 0 0 0 1
[ ]
ISRN Robotics 7

Table 1 Table 2
Link 𝜃 𝑑 𝑎 𝛼 Link 𝜃 𝑑 𝑎 𝛼
2 ∗ 0 0 −90 1 ∗ 0 0 90
3 0 ∗ 0 90 2 ∗ 0 0 −90
3 ∗ 0 0 0 3 0 ∗ 0 90
4 ∗ 0 0 −90
z4

z3 a suitable coordinate system down the backbone. The position

at 𝜎 along the backbone is found as (Φ is frame orientation)

𝜃4
x4 𝜎 1
d3 0
p (𝜎, 𝑡) = ∫ 0 Φ (𝜂, 𝑡) [0] 𝑑𝜂. (8)
0
z2 x3
[0]
𝜃1 z1
The above equation reflects the model developed by
z0 Mochiyama and Suzuki [75–77], wherein the 𝑥-axis of the
𝜃2 local coordinate system is aligned with the tangent down the
backbone. An equivalent formulation developed earlier—the
x0 x2 first kinematic analysis for continuum robot backbones—was
x1
developed by Chirikjian [4, 5]. In that formulation, the 𝑦-axis
Figure 8: Three-dimensional constant curvature section geometry was aligned with the backbone tangent.
obtained via rotation about initial tangent. For the planar constant curvature section in Figure 7, the
orientation matrix is given by
0
Φ (𝜎, 𝑡) = [𝑅𝑧orientation ]
The model (7) describes the forward kinematic relation-
ship (3 by 3 orientation, top left of (7), and 3 by 1 translation, 𝜎 𝜎
top right) between continuum curve shape (arc length and [cos (∫0 𝑘 𝑑𝜂) − sin (∫0 𝑘 𝑑𝜂) 0]
curvature) and task space. Note that the relationship is not, as [ 𝜎 𝜎 ] (9)
[ ]
for the rigid-link case, restricted to transformations from end = [ sin (∫ 𝑘 𝑑𝜂) cos (∫ 𝑘 𝑑𝜂) 0]
[ ]
to end of the “link” (“section” here). By making the arc length [ 0 0 ]
𝑠 arbitrary, the expression (7) models the transformation 0 0 1
[ ]
from curve shape to any task-space point along the backbone.
Thus, the D-H algorithm for discrete jointed robots has been (𝑘 is the curvature). Utilizing (9) in (8) and performing the
used to create a truly continuum section kinematic model integration, we obtain
in (7). Planar multisection kinematic models can be easily
1
created by chaining together (multiplying the homogeneous ( ) {cos (𝑘𝜎) − 1}
[ 𝑘 ]
transformation matrices) the models for the individual sec- [ 1 ]
tions [17].
0
p (𝜎, 𝑡) = [
[ ( 𝑘 ) sin (𝑘𝜎) ] .
] (10)
The kinematics of spatial constant curvature curves can [ ]
0
similarly be modeled by the addition of an extra pair of [ ]
(again identical, coupled) rotations to each end of the planar
version to create a 3D virtual rigid-link robot. In a similar Noting that
fashion, continuum kinematics are found by substitution 𝜎 𝜎

[cos (∫0 𝑘 𝑑𝜂) − sin (∫0 𝑘 𝑑𝜂) 0]

of appropriate geometric relationships; see Figure 8 and
Table 2. The 3D virtual robot is formally a 5-joint robot, [ 𝜎 𝜎 ]
0 [ ]
with 2D rotational joints at each end of a prismatic joint. Φ (𝜎, 𝑡) = [ sin (∫ 𝑘 𝑑𝜂) cos (∫ 𝑘 𝑑𝜂) 0]
[ ]
However, the robot has only three independent degrees of [ 0 0 ]
freedom, as the two 2D rotational joints are coupled. This 0 0 1 (11)
[ ]
agrees with the intuitive number of degrees of freedom of a
constant curvature curve in space, that is, arc length, radius cos (𝑘𝜎) − sin (𝑘𝜎) 0
of curvature, and orientation of curve plane in space. Mul- = [ sin (𝑘𝜎) cos (𝑘𝜎) 0] ,
tisection 3D kinematic models can be created by chaining [ 0 0 1]
together individual section models as before [17].
and recalling that
3.2. “Direct” Continuum Kinematics Approach. An alterna- 0 0p 0
Φ (𝜎, 𝑡) (𝜎, 𝑡)
tive (and more mathematically direct) approach explicitly [𝐻30 ] = [ ]. (12)
treats the continuum backbone as a curve in space and “floats” 0 1
8 ISRN Robotics

We obtain (noting the “0” in the previous equation is a 1 use the “natural basis set” or “box functions” (left column in
by 3 vector) Figure 9):

1 𝑛
cos (𝑠𝑘) − sin (𝑠𝑘) 0 ( ) {cos (𝑠𝑘) − 1} 𝑘 (𝑠) = ∑𝜇𝑖 𝜙𝑖𝑏 (𝑠)
[ 𝑘 ] (16)
[ 1 ]
[ sin (𝑠𝑘) cos (𝑠𝑘) 0 ( ) sin (𝑠𝑘) ]
𝑖=1
[ ]
[𝐻30 ] = [ 𝑘 ] . (13)
[ ] and the “Haar” basis set (right column in Figure 9):
[ 0 0 1 0 ]
[ ]
[ ]
𝑛
0 0 0 1
[ ] 𝑘 (𝑠) = ∑𝜇𝑖 𝜙𝑖𝑤 (𝑠) . (17)
𝑖=1
Note that the resulting model in (13) is identical to the
one obtained in (7). A similar parallel calculation can be The natural basis set comprises “box functions,” with
performed to generate the 3D kinematic model. Note that the (fixed) support, chosen to match the backbone region of
model suffers from singularities (division by zero) when the (nonextensible, planar) sections, and with the box mag-
curvature is zero or when any section is straight. nitudes corresponding directly to the curvature of those
Consequently, we have established a baseline kinematic sections. The effect of using this basis set is illustrated in
model relating backbone shape to task-space variables, inde- Figure 9 (column 2) to successively (top to bottom) form
pendent of the approach used to find it. a specific shape for a four-section planar backbone. Note
that the natural basis set features fixed spatial resolution at
3.3. Modal Approaches. The previous approaches establish variable locations along the backbone.
kinematics which directly model the nominal shapes the The (first four elements of the) Haar basis set (both
robots can obtain. However, they are fairly complex to for- sets are orthogonal sets in the sense of Wavelet bases) is
mulate and manipulate. An alternative strategy is to “build” shown in the right hand column, top to bottom, of Figure 9.
backbone shapes via a finite number of relatively simple These functions feature variable spatial resolution, at variable
modal functions. Originally introduced to robotics in [78] locations. However, for this particular example, it can be seen
(in which the resulting shapes were used to plan movements that they can also be combined (top to bottom, third column
for hyperredundant rigid-link robots), the modal approach of Figure 9) to produce the same shape as with the natural
builds the kinematics using as backbone curvature: basis set.
Note that the natural basis “box” functions in Figure 9
𝑛
directly synthesize the kinematics for four constant curvature
𝑘 (𝑠) = ∑𝜇𝑖 𝜙𝑖 (𝑠) , (14)
𝑖=1
sections. This is an example of exact modal function model-
ing. However, the box function approach is not generalized
where 𝜇𝑖 are coefficients, and 𝜙𝑖 (𝑠) are the modal functions. to extensible sections or to 3D backbones; so this is a
The modal functions act as “basis” functions for the backbone (convenient) special case.
curvature, with the coefficients selected to tune their combi- The key advantage of the modal function approach is
nation to approximate (or, in some special cases, match) the that the robot shape can be parameterized by a finite set
robot backbone shape [16, 78]. The coefficients become the of user-selected functions with convenient properties. The
“configuration” of the robot. number of “modes” used can be user-selected, for example,
The (number and type of) modal functions can be to constrain the computational complexity of the resulting
selected from a wide range of possibilities [16, 78, 79]. For model. Modal models can be synthesized to eliminate the
example, in [78], Chirikjian and Burdick used the following singularities inherent in the models of Sections 3.1 and 3.2
model: (when sections straighten and curvature is zero) [79].
𝑛 The main disadvantage of modal function-based
𝑘 (𝑠) = ∑𝜇𝑖 cos (𝑖𝜋𝑠) + 𝛽𝑖 sin (𝑖𝜋𝑠) . (15) approaches is that they are inherently approximations: most
𝑖=1 modal function sets which are convenient to manipulate
require many (typically, infinite) numbers of them to achieve
In this model, the configuration becomes 𝑞 = a given backbone shape. Most importantly, the set of shapes
[𝜇1 (𝑡), 𝛽1 (𝑡), . . . , 𝛽𝑖 (𝑡)𝜇𝑖 (𝑡), . . . , 𝜇𝑛 (𝑡), 𝛽𝑛 (𝑡)]. The use of available in the model is inherently restricted by the range
the classical trigonometric basis functions appears a natural of (linear combinations of) the modal basis functions. For
choice. However, an infinite number of trigonometric modes many modal function sets, even if an infinite number is
are needed to model an arbitrary backbone shape. Also, this used, it is not possible to model an arbitrary backbone shape.
basis set provides no spatial resolution (each basis function Therefore, some physically achievable backbone shapes may
affects the shape of the whole robot); so it is not possible to not be included in the modal models. This particularly is a
“tune” a given region of the backbone. problem for 3D backbones [16]. Therefore, the use of modal
It is sometimes possible to select modal functions to function approaches typically needs to be augmented with
achieve spatial resolution. In [16], (the first few elements of) additional analysis of and compensation for the inherent
two alternative sets of Wavelet basis functions are used. These mismatch between models and hardware.
ISRN Robotics 9

4 1 1 4

𝜙1w (s)
0 0.5 0.5 0

𝜙1b (s)

−4 0 0 −4
0 1 0 0.5 −0.5 0 0 1
4 1 1 4

𝜙2b (s)
0 0.5 0.5 0 𝜙2w (s)

−4 0 0 −4
0 1 −0.5 0 0.5 −0.5 0 0.5 0 1

4 1 1 4

𝜙3w (s)
𝜙3b (s)
0 0.5 0.5 0

−4 0 0 −4
0 1 −0.5 0 0.5 −0.5 0 0.5 0 1

4 1 1 4

𝜙4b (s)
𝜙4w (s)
0 0.5 0.5 0

−4 0 0 −4
0 1 −0.5 0 0.5 −0.5 0 0.5 0 1

Figure 9: Construction of four-section planar continuum backbone shape via alternative modal basis sets. Left column: first four elements
(top to bottom) of natural basis set. Right column: first four elements (top to bottom) of Haar basis set. Second column: backbone shape
generation (top to bottom), adding successive natural basis functions. Third column: backbone shape generation (top to bottom), adding
successive Haar basis functions.

3.4. Kinematic Transformations Incorporating Actuator Vari- robots, the relationship between the key variables defining
ables. Kinematic backbone models are very useful and pro- backbone configuration (extension, curvature, and torsion)
vide key insight into the possible configurations and behavior and actuated variables is considerably more complex.
of continuum robots. However, for practical implementation For example, consider tendon-based continuum robots.
of continuum robot realizations, even at the kinematic Here, the only actuated (and, often, sensed) variables are
level, further modeling is required. In the case of rigid-link tendon lengths, and it is desirable in practice to be able
robots, the variables underlying their kinematic models (joint to convert these tendon lengths to backbone configuration
angles and displacements) correspond closely to actuator (extension, curvature, torsion). This issue has been addressed
geometry. Conversion between configuration and actuator in [18], and in the following, we summarize the solution for
variables (typically involving gear/belt/drive reduction ratios) the case of a single (extensible) section with circular cross-
is relatively straightforward [1]. However, for continuum section, actuated by three tendons, spaced at 120 degrees
10 ISRN Robotics

apart around the section perimeter. A good example of The aforementioned in models are critical in imple-
such a section is the “Air-Octor” continuum robot [68, 69], mentation of tendon-based continuum robots. They can be
illustrated in Figure 10. easily modified for locally actuated designs, where tendon
For the continuum section in Figure 10 (the kinematics lengths are replaced by internal actuator lengths. Finally,
are similar, if not identical, for the sections of almost all note that the concentric tube realization of continuum robots
tendon-driven continuum robots), the forward problem is presents a distinct advantage in the context of kinematic
to find the section shape from the three tendon lengths 𝑙1−3 . transformations, as the local extension and torsion (of each
The shape is given by the three configuration space variables: tube section) are directly actuated, and thus no kinematic
section length 𝑠, curvature 𝑘𝜙 , and angle of curvature (from transformation between these configuration and actuation
the 𝑥-axis, measured about the 𝑧-axis of a coordinate frame variables is required.
with its 𝑧-axis aligned with the base tangent of the section)
(see Figure 11). 3.5. Extensions, Inverse and Velocity Kinematics. The kine-
The key to developing the needed transformations is to matics formulations in Sections 3.1–3.4 represent the core
exploit the geometric constraints of the design. The three theory underlying continuum robots and have been the
tendons are routed through a series of 𝑛 − 1 intermediate subject of much of the theoretical research activity in the
connection points before being terminated at the end of the area thus far. Extensions to this core theory have been
section. During actuation, this causes the tendons to form developed and there remains active research in the area.
𝑛 straight line segments within the section. Straightforward In particular, efforts to remove the restriction to constant
geometrical analysis can be used to show that the length ℎ𝑐 of curvature sections have been made.
a (imaginary) tendon running directly through the center of More accurate kinematic models which explicitly include
a single such segment of the section is given by the effects of external (particularly gravitational) loading have
been explored in the literature. The theory of Cosserat Rods
𝑙3 + 𝑙2 − 2𝑙1
ℎ𝑐 = , (18) has proved particularly helpful in enabling researchers to
6𝑛 develop “geometrically exact” nonconstant curvature kine-
where the shortest tendon length is 𝑙1 , and 𝑛 is the number of matic models [37, 60, 83]. However, the models are com-
segments in the section. This length can be used to analyze putationally complex [60] and harder to implement than
the “side-on” geometry (Figure 12). the constant curvature models. Consequently, the constant
Use of the geometric information in Figure 12 and a curvature models remain predominant in continuum robot
projection onto the (𝑧, 𝜙) plane results [18] in expressions implementations.
for the curvature and angle of curvature in terms of tendon Inverse models (including those for both the configura-
lengths tion-workspace transformations of Sections 3.1–3.3 and
the actuator-configuration relationships of Section 3.4) have
√𝑙1 2 + 𝑙2 2 + 𝑙3 2 − 𝑙1 𝑙2 − 𝑙2 𝑙3 − 𝑙1 𝑙3 been developed in [74] (for nonconstant curvature models),
𝑘𝜙 = 2 , [79, 82].
𝑑 (𝑙1 + 𝑙2 + 𝑙3 ) Various approaches to exploiting the kinematic relation-
(19)
ships to formulate and exploit Jacobians which reflect the
√3 𝑙3 + 𝑙2 − 2𝑙1
𝜙 = tan−1 ( ), velocity-level kinematic relationships have been proposed
3 𝑙2 − 𝑙3 [16, 74, 84]. Conventional techniques for formulating Jaco-
where 𝑑 is the radius of the section cross-section, and 𝑟1 bians for rigid-link robots can be applied to the “virtual”
is the inverse of its curvature. Finally, after some further rigid-link manipulator of Section 3.1 to derive a “continuum”
geometrical analysis, it can be shown that [18]: Jacobian. More directly, any of the kinematic relationships
(modal or direct) can be differentiated to find the appropriate
𝑛𝑑 (𝑙1 + 𝑙2 + 𝑙3 ) Jacobian. A Jacobian (pseudo-) inverse can then be used
𝑠= to iteratively solve configuration space rates given desired
√𝑙1 + 𝑙2 2 + 𝑙3 2 − 𝑙1 𝑙2 − 𝑙2 𝑙3 − 𝑙1 𝑙3
2
tip rates. For example, the modal function approach for the
(20) planar four-degree-of-freedom example in Section 3.3 can be
√𝑙1 2 + 𝑙2 2 + 𝑙3 2 − 𝑙1 𝑙2 − 𝑙2 𝑙3 − 𝑙1 𝑙3 used to derive
−1
⋅ sin ( ). 𝑑𝜇
3𝑛𝑑 𝑑𝑥 (𝑠)
= [𝐽 (𝜇, 𝑠)] , (21)
𝑑𝑡 𝑑𝑡
Equations (19) and (20) constitute the needed forward thus formulating the Jacobian and its relationship in terms
map between actuator and shape variables for tendon-driven of the modal function coefficients and their time derivatives
extensible sections. Further work on kinematic transfor- [16]. Inverting this relationship yields model coefficient (and
mations can be found in [18, 80, 81]. For example, in hence shape) rates which can be numerically integrated to
[80], a mechanics-based model for transforming between provide shape trajectories corresponding to the input tip
beam configuration and tendon displacements is formulated. rates. See Figure 13 for an example trajectory generated for
Inverse maps can also be formulated [81, 82]. It is fairly the example robot of Section 3.3 using this approach.
straightforward to chain together multiple section models to Research into continuum robot kinematics continues.
model a multisection robot; see [18] for details. For example, an alternative (screw theoretic) approach to
ISRN Robotics 11

Cable guide

120∘

Central hose

Outer hose

Figure 10: The single-section “Air-Octor.” Internal hose is pressurized; external hose is actuated (two independent degrees of freedom
bending, plus one of extension) by three remotely actuated tendons, terminated at the end of the section and spaced 120 degrees apart around
its perimeter.

z
y l3 + l2 − 2l1
hc =
6

x l1
𝜙
2n
k
r1

Figure 11: Definition of section configuration space variables for

curvature and angle of curvature.

the development of both kinematics and Jacobian formula-

tion, using the “product of exponentials” approach, is given
recently in [39]. The approach is shown to produce the same
Figure 12: View from side of one segment of section.
results as using the models described previously.

4. Forces, Dynamics, and Control

4.1. Forces and Continuum Robots. The natural next step in a continuum robot backbone will be determined by both the
modeling of continuum robots is to include the relation- constraints of the controlled degrees of freedom and internal
ships involving forces (and moments) and backbone shape and external forces.
and shape changes. For conventional rigid-link robots, the The kinematic models of Section 3 made no explicit
kinematic variables (joint angles/displacements) completely assumptions of internal or external loading. Implicitly, the
determine the shape (configuration) of the robot. However, constant curvature assumptions in Section 3 assumed no
for continuum robots, external forces (in particular, gravity) external loading and an even distribution of internal loading
act with the configuration variables and the internal energy within each section (to even out the uncontrolled degrees
of the backbone to determine the ultimate shape of each of freedom to create the constant curvature). In practice,
section [19, 47], even in free space. In any physically realizable gravitational loads cause some “section sag,” resulting in
system, there can be only a finite number of actuated (i.e., section curvatures that are not truly constant. However, the
directly controlled) configuration degrees of freedom. The deviation from constant curvature in free space is typically
values of the remainder of the (infinite) degrees of freedom in not large, and the constant curvature models have proved
12 ISRN Robotics

1 for concentric tube robots is presented in [87]. An approach

to continuum backbone contact detection and location of
0.9 contact along the backbone is developed in [88]. Methods
Initial
Initi l pose
n tia o
0.8
based on kinematics and statics are presented, supported by
experimental results with a tendon-driven continuum back-
0.7 bone. Recent extensions [38] present algorithms (supported
by experimental results) to register and locate continuum
0.6 backbones under external contact conditions with respect to
an a priori 3D model.
x2

0.5
A novel alternative strategy is to explicitly use continuum
0.4 robots as force sensors [37]. In [37], an Extended Kalman
Filter approach is used with a tendon-driven continuum
Final pose
0.3 robot to provide tip force estimates given kinematic models
and estimates.
0.2

0.1 4.2. Continuum Robot Dynamics. The earliest published work

on continuum robot dynamics, based on a modal model,
0 was [5]. Subsequent efforts based on the well-understood [1]
−0.5 0 0.5
Lagrangian [75, 76, 89, 90] and Newton-Euler [61, 91, 92]
x1 methods have been established. In the following, we outline
Figure 13: Straight line tip trajectory for four-section planar the Lagrangian dynamics approaches [75, 76, 90], which
continuum robot, generated using Jacobian from modal kinematics. generally parallel the commonly used formulation for rigid-
link robots. However, there are several steps which are both
specialized to continuum robot scenario and which give good
insight into the nature of continuum robots.
a good practical approximation in many implementations The key initial step [75] is to model the backbone as being
[17, 59, 63]. comprised of circular cross-sectional “slices,” of infinitesimal
However, as applications become more challenging, the thickness. Each slice, at a location 𝜎 along the backbone, has
desire for more accurate models is increasing. Internal loads mass 𝑚(𝜎), inertia tensor 𝐼(𝜎), and first moment of inertia
can play a significant role in determining the shape of 𝑚(𝜎)𝑟(𝜎), where 𝑟(𝜎) is the distance from the slice geometric
concentric tube robots in particular. In [47], a quasistatic center to its center of mass. The overall strategy is to find
model for predicting the shape of concentric tube robots the kinetic and potential energy of each slice, then the total
is presented. The model in [47] takes internal forces from energies 𝐾 and 𝑃 (via integration along the backbone), and
both bending and shear effects into account. Mechanics- finally substitute 𝐿 = 𝐾 − 𝑃 into Lagrange’s equations (𝑖 =
based models which consider torsional interaction between 1, . . . , 𝑛)
the tubes of concentric tube robots are presented in [85].
Lumped-parameter models to model friction within a con- 𝑑 𝜕𝐿 𝜕𝐿
[ ]− = 𝜏𝑖 (𝜎, 𝑡) (22)
tinuum catheter device are presented in [86]. ̇
𝑑𝑡 𝜕𝜃𝑖 (𝜎, 𝑡) 𝜕𝜃𝑖 (𝜎, 𝑡)
More accurate kinematic models which explicitly include
the effects of external (particularly, gravitational) loading to find the dynamic model. In the above, (𝜃𝑖 , 𝜏𝑖 ) are variables
have been explored in the literature [37, 83]. However, these selected to correspond to the 𝑛 actuated configuration space
models are relatively complex. Consequently, the constant variables and the corresponding forces which change them.
curvature models remain predominant in continuum robot The kinetic energy of a slice (assuming the center of mass
implementations. is aligned with the geometric center of the slice) is most
External loading becomes a much more significant factor directly given by
when the loads result from environmental contact, however.
In this case (inevitable when using continuum elements 1 𝜕p (𝜎, 𝑡) 𝑇 𝜕p (𝜎, 𝑡)
for locomotion and actively sought when using them for 𝐾 (𝜎, 𝑡) = ( ) 𝑚 (𝜎) { } { }
2 𝜕𝑡 𝜕𝑡
grasping and manipulation), the inherent compliance of the
1
continuum structures allows for significant deviation from + ( ) 𝜔(𝜎, 𝑡)𝑇 𝐼 (𝜎) 𝜔 (𝜎, 𝑡) (23)
constant curvature. Notice that this deviation is inevitable— 2
the finite number of actuated degrees of freedom cannot, 1
in general, be made to control the infinite number of other ≜ ( ) V(𝜎, 𝑡)𝑇 𝑀 (𝜎, 𝑡) V (𝜎, 𝑡) ,
2
degrees of freedom in the structure, in the presence of general
and significant external forces. The deviation is also a desired where {𝜕p(𝜎, 𝑡)/𝜕𝑡, 𝜔(𝜎, 𝑡)} are the linear and angular veloci-
property, in order to allow the robot to adapt its shape to ties of the slice center at 𝜎.
environmental constraints, both for navigation and grasping. An additional step (analogous to the process for rigid-
In order to address some of these issues, new work link robot Lagrangian dynamics) is now necessary. In order
considers applied loading. A solution to the statics problem to obtain the dynamics in terms of the useful variables (𝜃𝑖 , 𝜏𝑖 ),
ISRN Robotics 13

a (Jacobian) transformation must be found to convert V(𝜎, 𝑡) for bending (spring constant 𝑘𝑏 ) and
in the above to 𝜕𝜃(𝜎, 𝑡)/𝜕𝑡. This Jacobian needs to account
for the phenomenon that the velocity at 𝜎 is a function of the 1 2
𝑃𝑒 (𝑡) = ( ) 𝑘𝑒 (𝑙 − 𝑙∗ ) (33)
velocities at all locations 𝜂 prior to it along the backbone. To 2
this end, intermediate Jacobians [𝐽1 ], [𝐽2 ] are found such that
for extension (spring constant 𝑘𝑒 ). Details of application
k (𝜎) k (𝜂) of the previous approach to compute the dynamics for
[ ] = [𝐽1 ] [ ],
𝜔 (𝜎) 𝜔 (𝜂) extensible and nonextensible backbones are given in [90].
(24) Establishing the previous closed form dynamic models
k (𝜂) 𝜕𝜃 (𝜂, 𝑡) for continuum robots has been significant in yielding insight
[ ] = [𝐽2 ] ,
𝜔 (𝜂) 𝜕𝑡 into the underlying structure of these devices. The structure
of the model is useful in synthesizing control strategies
and a desired overall Jacobian operator [𝐽(𝜎,𝑡) ] is found as [93, 94], and the dynamics enables realistic physics-based
𝜎
simulations of continuum robots. However, the computa-
𝜕𝜃 𝜕𝜃 (𝜂, 𝑡) tional complexity of the resulting model is very high, and
V (𝜎, 𝑡) = [𝐽(𝜎,𝑡) ] { } = ∫ [𝐽1 ] [𝐽2 ] 𝑑𝜂. (25)
𝜕𝑡 0 𝜕𝑡 the calculations for nonplanar backbones are nontrivial.
Real-time implementation of these models (even with the
For full details, see [75].
simplifying assumptions of simple mass distribution) is a
Thus, the kinetic energy of a slice can now be represented
daunting prospect.
as
Consequently, researchers have explored alternative
1 𝜕𝜃 𝑇 𝜕𝜃 approaches to dynamic modeling for continuum robots.
𝐾 (𝜎, 𝑡) = ( ) { } 𝐽(𝜎,𝑡) 𝑇 𝑀 (𝜎) 𝐽(𝜎,𝑡) { } . (26) Formulations based on the iterative Newton-Euler
2 𝜕𝑡 𝜕𝑡
approach have been established [61, 91, 92]. More recently,
The overall kinetic energy is therefore computationally (more) efficient lumped-parameter models,
based on linear mass-spring-damper elements, have
𝑙
appeared in the literature [95–97]. The model in [97] is tuned
𝐾 (𝑡) = ∫ 𝐾 (𝜎, 𝑡) 𝑑𝜎, (27)
0 to octopus-inspired underwater operation and includes
terms to model buoyancy and drag. These approaches
where 𝑙 is the length of the backbone. The potential energy of approximate the dynamics of (sections of) continuum
a slice is found as backbones by various combinations of linear elements,
with a view of trading off computational complexity of
𝑃𝑔 (𝜎, 𝑡) = −𝑚 (𝜎) g𝑇 (𝜎, 𝑡) p (𝜎, 𝑡) , (28)
the model against accuracy. Experiments have shown
where g is the gravitation vector at 𝜎. The overall potential good correspondence between the models and hardware,
energy given by for relatively a small number (less than twenty) of linear
elements [95].
𝑙 𝑙 Dynamic models for nonconstant curvature continuum
𝑃𝑔 (𝑡) = ∫ 𝑃𝑔 (𝜎, 𝑡) 𝑑𝜎 = − ∫ 𝑚 (𝜎) g𝑇 (𝜎, 𝑡) p (𝜎, 𝑡) . (29) robots are beginning to appear in the literature [83]. However,
0 0
at the time of this writing, there are very few examples
After forming 𝐿 = 𝐾 − 𝑃 and substituting into Lagrange’s of continuum dynamics being implemented on continuum
equations (22), the resulting dynamic model takes the form robot hardware. It is expected that, as applications expand,
particularly in areas such as medical procedures and loco-
Mq̈ + V (q, q)̇ q̇ + g (q) = 𝜏. (30) motion with continuum limbs, the availability of dynamics
The underlying structure of the dynamic model closely in real time will become increasingly necessary. Research in
matches that of rigid-link robots, apart from being contin- continuum robot dynamics continues to be a very active area.
uous in nature. For example, the inertia matrix 𝑀 can be
shown [90, 93] to be positive definite and symmetric, and it 4.3. Continuum Robot Control. Controller development for
satisfies the property (useful for control) robotic structures is a fundamental issue, and several of
the early works [98, 99] in continuum robots concentrated
𝜉𝑇 (𝑀̇ − 2𝑉) 𝜉 = 0, ∀𝜉 ∈ R𝑛 . (31) on this topic. Control of continuum structures is obviously
complicated by the inherently underactuated nature of the
The original derivation of (30) [75] assumed fixed length backbone. Additional problems arise from the (typical) non-
(i.e., nonextensible) backbones and the only potential energy collocation of actuators with configuration space variables,
in the system arose from gravity. Later work [90] generalized the complexity of the dynamic models, and the typical dearth
the models to include extensible sections, and the effects of of local sensors in the structure.
elastic internal potential energy due to both bending and For tendon-based continuum robots particularly, the
extension. The corresponding energy terms are existence of kinematic transformations between configura-
𝑙 2
tion space and actuator space variables is critical in enabling
1 1 effective control. Controllers specific to tendon-based con-
𝑃𝑏 (𝑡) = ( ) ∫ 𝑘𝑏 (𝜎) {𝜋 − ( ) 𝜎𝜅 (𝜎)} 𝑑𝜎 (32)
2 0 2 tinuum robots have been presented in [80, 81, 83]. Stiffness
14 ISRN Robotics

controllers for continuum robots are introduced in [46]. environments and are intended for real-time, or near real-
In [46], a Cosserat Rod model is used to calculate the tip time, implementation [103]. Extensions for moving obstacles
deflection due to applied forces, enabling implementation can be handled [105]. The emergence of low-cost and readily
of stiffness control. The approach is applied in [46] to a available sensors such as the Kinect RGB-D sensor catalyzes
concentric tube robot, but the modeling strategy (combining this type of research.
noncontact kinematics as in Section 3 with the Cosserat The existence of practical, real-time, sensor-based motion
Rod model to include applied forces) is applicable to other planning algorithms will be important in moving the tech-
continuum robot designs as well. nology from the laboratory to numerous real applications
Control strategies attempting to compensate for the [106]. Currently, most of the higher-level functions (including
complexity of the dynamics or for the uncertainty inherently planning) are delegated to humans. Most implementations to
present in implementation of dynamic models are presented this point require teleoperation of the device [107], typically
in [98–100]. Inverse dynamics controllers are described in using a joystick [108]. User interfaces for continuum robots is
[100]. In [99, 100], variable structure controllers are proposed. a relatively neglected and poorly understood topic [108]. The
A feedforward neural network approach for compensation movements of continuum elements are often counterintuitive
for the dynamics is discussed in [94]. Controllers in the task to human operators, much more than for rigid-link robots,
space are presented in [49, 100]. A controller based on a leading to confusion and slow or poor decision making in
mechanics model describing coupling between sections is many experiments. Efforts to reduce operator cognitive load
presented in [81]. via automatic motion planning and the synthesis of low
A key practical problem for continuum robot control degree of freedom “synergies” [109, 110] are expected to prove
is at which level to close the loop. Sensors are typically significant in widening the scope and effectiveness of possible
noncollocated with the backbone. Sensed quantities are operations.
usually limited to tendon lengths (sensed at the base), Recent theoretical work has investigated the structure of
or pneumatic or hydraulic pressures of artificial muscles the “self-motion” inherent in continuum structures [111]. This
(again, sensed remotely at pressure regulators, not at the self-motion (movement of the backbone while maintaining
muscles themselves). Direct internal sensing of backbone the location of the tip), present in kinematically redun-
shape is complicated by the limited space available and dant robots, has proved useful in conventional rigid-link
the current lack of suitable sensor technologies. Technolo- manipulators. Multisection continuum robots are inherently
gies based on fiber optics offer promise for low-profile, kinematically redundant, but, apart from its proposed use in
high-quality local sensing of backbone curvature. However, task-space control strategies [100], their self-motion remains
this has not yet been demonstrated in continuum robot a little studied phenomenon.
hardware. Consequently, transformations between actuator The use of continuum robot structures as limbs for
and configuration space become important for controller “legged” locomotion has recently been the subject of atten-
implementation, with the loop closed (error calculated) in tion [61, 112, 113]. Demonstration of multicontinuum-limbed
either space. locomotion has been reported both underwater [33, 45],
External sensing of continuum robot shape has also and in air [112]. Multi- (two-) continuum-limbed swimming
been demonstrated. Electromagnetic field sensors were used is analyzed with feasibility demonstrated via a simple pro-
in [36] to sense the shape of tendon-driven endoscopic totype in [114]. Multicontinuum-legged robots have been
systems. Vision has been used to infer the shape of continuum demonstrated to walk and trot [112]. Possible applications
robots in [101] (off-board cameras) and [102] (body-mounted for this novel locomotion mode include various undersea
cameras), but the effectiveness of this strategy is limited by and space applications. Other space-based applications in
issues of lighting conditions, occlusions, and so forth. Vision which continuum robots could be used effectively are as
has also been used to sense backbone tip location [81]. An “active hooks” [113] and as long, thin “tendril-like” robots
electromagnetic sensor is used in [49] to sense tip location [31, 115–117]. In the latter mode, the robots could penetrate
and enable task-space control. and explore within tight obstacle fields such as crevasses,
lava tubes, or skylights, where key scientific questions are
currently focused.
5. Ongoing Research Novel applications are emerging. Continuum robots have
been deployed recently as “active hoses” for ship-to-shore
With the field rapidly expanding and the core underlying refueling [118]. Extension to continuum robot surfaces, with
theory having been established over the last ten to fifteen application to rehabilitation therapy for poststroke patients,
years, a wider variety of issues and problems are now being has been proposed [119].
considered by researchers. New work in motion planning
for continuum robots [52, 85, 103–105] is expanding the
boundaries of the possible for the field. Motion planners 6. Conclusions
based on mechanics models for concentric tube robots are
presented in [85, 103]. Planners have been proposed which We have reviewed the state of the art in continuum manipu-
use the structure of continuum backbones to plan trajectories lators, focusing particularly on hardware design, kinematics,
which simultaneously avoid obstacles and grasp objects. and dynamics. The design of continuum robot hardware
These planners are proposed for both 2D [105] and 3D [103] is seen to have evolved into three fundamentally different
ISRN Robotics 15

directions. Two of these directions feature out-of-backbone anticipated that the next ten years will see an explosion of
“extrinsic” actuation: one via remote actuation via tendons research, both basic and applied, in the area of continuum
and the other via remotely actuated sliding of concentric robots.
tubes. The third direction forms the continuum backbone
from the actuators themselves and is hence termed “intrin- Acknowledgments
sically actuated.” Actuator selection for intrinsically actuated
continuum robots can be from any available type of artificial This research was supported in part by NASA under NRI
muscle. This could include pneumatic “McKibben” muscles, Contract NNX12AM01G and in part by the U.S. National
hydraulically actuated muscles, muscles based on novel poly- Science Foundation under Grant IIS-0904116.
mers, or shape memory alloys, for example. However, at this
time only the pneumatic or hydraulic actuator technologies
feature the combination of bending and force generation References
capabilities for continuum robots at the human scale or larger. [1] M. W. Spong, S. Hutchinson, and M. Vidyasagar, Robot Model-
At a large scale, the extrinsic tendon-driven designs have ing and Control, Wiley, 2006.
the advantage of high force capability. At the small scale, [2] J. K. Salisbury, “Whole arm manipulation,” in Proceedings of the
the concentric tendon design has advantages and is already 4th Symposium on Robotics Research, MIT Press, 1987.
showing promise in numerous medical applications. [3] V. C. Anderson and R. C. Horn, “Tensor arm manipulator
Understanding of the kinematics of continuum robots has design,” Transactions of the ASME, vol. 67, DE-57, pp. 1–12, 1967.
been the subject of much research and has now reached a [4] G. S. Chirikjian, Theory and applications of hyperredundant
mature stage, with theory matching most of the correspond- robotic mechanisms [Ph.D. thesis], Department of Applied
ing results for rigid-link robots now established. Indeed, key Mechanics, California Institute of Technology, 1992.
aspects of the core theory for rigid-link robots can be used [5] G. S. Chirikjian, “Hyper-redundant manipulator dynamics: a
to form a baseline model for continuum robot kinematics, as continuum approximation,” Advanced Robotics, vol. 9, no. 3, pp.
has been discussed herein. Alternatively, the same models can 217–243, 1994.
be derived from first principles, and the results are seen to be [6] S. Hirose, Biologically Inspired Robots, Oxford University Press,
equivalent. However, continuum robot kinematics modeling 1993.
presents issues and difficulties not present for rigid-link [7] I. A. Gravagne, C. D. Rahn, and I. D. Walker, “Large deflec-
robots, due to the inherent compliance and infinite degrees tion dynamics and control for planar continuum robots,”
of freedom present in continuum robot backbones. Models IEEE/ASME Transactions on Mechatronics, vol. 8, no. 2, pp. 299–
which take into account the effects on the kinematics from 307, 2003.
external loading such as from gravity have been established. [8] M. W. Hannan and I. D. Walker, “Analysis and experiments with
However, the additional computational complexity in these an elephant’s trunk robot,” Advanced Robotics, vol. 15, no. 8, pp.
models remains to be a barrier to implementation at this 847–858, 2001.
time. [9] R. Cieślak and A. Morecki, “Elephant trunk type elastic
Computational complexity is also a significant issue in manipulator—a tool for bulk and liquid materials transporta-
tion,” Robotica, vol. 17, no. 1, pp. 11–16, 1999.
the study and application of continuum robot dynamics, and
related subfields. As discussed herein, traditional methods for [10] G. Immega, “Tentacle-like Manipulators with adjustable ten-
sion lines,” U.S. Patent 5317952, 1992.
dynamics formulation, such as the Lagrangian and Newton-
[11] G. Immega and K. Antonelli, “KSI tentacle manipulator,” in Pro-
Euler approaches, can be extended and adapted to form
ceedings of the 1995 IEEE International Conference on Robotics
continuum robot dynamic models. These models are seen and Automation. Part 1 (of 3), pp. 3149–3154, Nagoya, Japan,
to possess the same key structural properties as for rigid- May 1995.
link robots, encouraging the development of correspond- [12] D. M. Lane, J. B. C. Davies, G. Robinson et al., “The AMADEUS
ing control strategies. However, the continuous nature of dextrous subsea hand: design, modeling, and sensor process-
the continuum backbone renders these models extremely ing,” IEEE Journal of Oceanic Engineering, vol. 24, no. 1, pp. 96–
computationally complex. Some recent work has developed 111, 1999.
less computationally complex approximate dynamic models. [13] K. Suzumori, S. Iikura, and H. Tanaka, “Development of flexible
However, at this time, few dynamics-based implementations microactuator and its applications to robotic mechanisms,” in
have been reported, and continuum robot dynamics and Proceedings of the IEEE International Conference on Robotics and
control remain active research issues. Automation, pp. 1622–1627, April 1991.
[14] J. F. Wilson, D. Li, Z. Chen, and R. T. George, “flexible robot
Taking advantage of the previous developments, research
manipulators and grippers: relatives of elephant trunks and
into continuum robots is actively expanding at the time squid tentacles,” in Robots and Biological Systems: Towards a
of writing this paper. New work in areas such as motion New Bionics? vol. 102, pp. 474–479, 1993.
planning and contact modeling is extending our underlying [15] G. Robinson and J. B. C. Davies, “Continuum robots—a state of
body of understanding and widening the scope of the field. the art,” in Proceedings of the IEEE International Conference on
Researchers are partnering with various industries to explore Robotics and Automation (ICRA ’99), pp. 2849–2854, May 1999.
continuum robot solutions to such diverse applications such [16] I. Gravagne and I. D. Walker, “Kinematics for constrained con-
as terrain-adapting continuum-limbed vehicles, ship-to-ship tinuum robots using wavelet decomposition,” in Proceedings of
refueling, and exploration of extraterrestrial surfaces. It is the 4th International Conference and Exposition/Demonstration
16 ISRN Robotics

on Robotics for Challenging Situations and Environments, pp. Technology VII, Proceedings of SPIE, pp. 303–314, Orlando, Fla,
292–298, Albuquerque, NM, USA, March 2000. USA, March 2005.
[17] M. W. Hannan and I. D. Walker, “Kinematics and the imple- [34] J. Bishop-Moser, G. Krishnan, C. Kim, and S. Kota, “Design
mentation of an elephant’s trunk manipulator and other con- of soft robotic actuators using fluid-filler fiber-reinforced elas-
tinuum style robots,” Journal of Robotic Systems, vol. 20, no. 2, tomeric enclosures in parallel combinations,” in Proceedings
pp. 45–63, 2003. of the IEEE/RSJ International Conference on Intelligent Robot
[18] B. A. Jones and I. D. Walker, “Kinematics for multisection Systems (IROS ’12), pp. 4262–4269, Vilamoura, Portugal, 2012.
continuum robots,” IEEE Transactions on Robotics, vol. 22, no. [35] A. Grzesiak, R. Becker, and A. Verl, “The bionic handling
1, pp. 43–55, 2006. assistant: a success story of additive manufacturing,” Assembly
[19] D. Trivedi, C. D. Rahn, W. M. Kier, and I. D. Walker, “Soft Automation, vol. 31, no. 4, pp. 329–333, 2011.
robotics: biological inspiration, state of the art, and future [36] B. Bardou, P. Zanne, F. Nageotte, and M. De Mathelin, “Control
research,” Applied Bionics and Biomechanics, vol. 5, no. 3, pp. of a multiple sections flexible endoscopic system,” in Proceedings
99–117, 2008. of the 23rd IEEE/RSJ 2010 International Conference on Intelligent
[20] R. J. Webster III and B. A. Jones, “Design and kinematic Robots and Systems (IROS ’10), pp. 2345–2350, Taipei, Taiwan,
modeling of constant curvature continuum robots: a review,” October 2010.
International Journal of Robotics Research, vol. 29, no. 13, pp. [37] D. C. Rucker and R. J. Webster III, “Deflection-based force
1661–1683, 2010. sensing for continuum robots: a probabilistic approach,” in Pro-
[21] T. Aoki, A. Ochiai, and S. Hirose, “Study on slime robot ceedings of the IEEE/RSJ International Conference on Intelligent
development of the mobile robot prototype model using bridle Robots and Systems: Celebrating 50 Years of Robotics (IROS ’11),
bellows,” in Proceedings of the IEEE International Conference pp. 3764–3769, San Francisco, Calif, USA, September 2011.
on Robotics and Automation, pp. 2808–2813, New Orleans, La, [38] S. Tully, A. Bajo, G. Kantor, H. Choset, and N. Simaan, “Con-
USA, May 2004. strained filtering with contact detection for the localization and
[22] H. Ohno and S. Hirose, “Design of slim slime robot and its registration of continuum robots in flexible environments,” in
gait of locomotion,” in Proceedings of the IEEE/RSJ International Proceedings of the IEEE International Conference on Robotics and
Conference on Intelligent Robots and Systems, pp. 707–715, Maui, Automation, pp. 3388–3394, St. Paul, Minn, USA, 2012.
Hawaii, USA, November 2001. [39] Q. Zhao and F. Gao, “Design and analysis of a kind of
[23] H. Tsukagoshi, A. Kitagawa, and M. Segawa, “Active hose: biomimetic continuum robot,” in Proceedings of the IEEE
an artificial elephant’s nose with maneuverability for rescue International Conference on Robotics and Biomimetics (ROBIO
operation,” in Proceedings of the IEEE International Conference ’10), pp. 1316–1320, Tianjin, China, December 2010.
on Robotics and Automation, pp. 2454–2459, Seoul, Korea, May [40] J. Shang, C. J. Payne, J. Clark et al., “Design of a multitasking
2001. robotic platform with flexible arms and articulated hand for
[24] R. Buckingham, “Snake arm robots,” Industrial Robot, vol. 29, minimally invasive surgery,” in Proceedings IEEE/RSJ Interna-
no. 3, pp. 242–245, 2002. tional Conference on Intelligent Robot Systems (IROS ’12), pp.
[25] R. Buckingham, OC Robotics, http://www.ocrobotics.com/. 1998–1993, Vilamoura, Portugal, 2012.
[26] F. W. Grasso, “Octopus sucker-arm coordination in grasping [41] G. Chen, P. M. Tu, T. R. Herve, and C. Prelle, “Design and
and manipulation,” American Malacological Bulletin, vol. 24, no. modeling of a micro-robotic manipulator for colonoscopy,” in
1-2, pp. 13–23, 2008. Proceedings of the 5th International Workshop on Research and
[27] J. L. Van Leeuwen and W. M. Kier, “Functional design of Education in Mechatronics, pp. 109–114, Annecy, France, 2005.
tentacles in squid: linking sarcomere ultrastructure to gross [42] H.-S. Yoon, S. M. Oh, J. H. Jeong et al., “Active bending
morphological dynamics,” Philosophical Transactions of the endoscope robot system for navigation through sinus area,”
Royal Society B, vol. 352, no. 1353, pp. 551–571, 1997. in Proceedings of the IEEE/RSJ International Conference on
[28] F. Martin and C. Niemitz, “How do African elephants (Lox- Intelligent Robots and Systems: Celebrating 50 Years of Robotics
odonta africana) optimize goal-directed trunk movements?” (IROS ’11), pp. 967–972, San Francisco, Calif, USA, September
in Jahresversammlung der Dt. Zool. Ges. und der Dt.Ges. f. 2011.
Parasitologie, vol. 96, Berlin, Germany, 2003. [43] H. Watanabe, K. Kanou, Y. Kobayashi, and M. G. Fujie,
[29] W. M. Kier and K. K. Smith, “Tongues, tentacles and trunks: the “Development of a “steerable drill” for ACL reconstruction to
biomechanics of movement in muscular-hydrostats,” Zoological create the arbitrary trajectory of a bone tunnel,” in Proceedings of
Journal of the Linnean Society, vol. 83, no. 4, pp. 307–324, 1985. the IEEE/RSJ International Conference on Intelligent Robots and
[30] A. Goriely and S. Neukirch, “Mechanics of climbing and Systems: Celebrating 50 Years of Robotics (IROS ’11), pp. 955–960,
attachment in twining plants,” Physical Review Letters, vol. 97, San Francisco, Calif, USA, September 2011.
no. 18, Article ID 184302, 2006. [44] N. Simaan, R. Taylor, and P. Flint, “A dexterous system for
[31] J. S. Mehling, M. A. Diftler, M. Chu, and M. Valvo, “A laryngeal surgery,” in Proceedings of the IEEE International
minimally invasive tendril robot for in-space inspection,” in Conference on Robotics and Automation, pp. 351–357, New
Proceedings of the 1st IEEE/RAS-EMBS International Conference Orleans, La, USA, May 2004.
on Biomedical Robotics and Biomechatronics (BioRob ’06), pp. [45] M. Calisti, A. Arienti, F. Renda et al., “Design and development
690–695, February 2006. of a soft robot with crawling and grasping capabilities,” in
[32] I. D. Walker, “Robot strings: long, thin continuum robots,” in Proceedings of the IEEE International Conference on Robotics and
Proceedings of the IEEE Aerospace Conference, pp. 1–12, Big Sky, Automation, pp. 4950–4955, St. Paul, Minn, USA, 2012.
Mont, USA, 2013. [46] M. Mahvash and P. E. Dupont, “Stiffness control of a continuum
[33] I. D. Walker, D. M. Dawson, T. Flash et al., “Continuum robot manipulator in contact with a soft environment,” in Proceedings
arms inspired by cephalopods,” in Unmanned Ground Vehicle of the 23rd IEEE/RSJ 2010 International Conference on Intelligent
ISRN Robotics 17

Robots and Systems (IROS ’10), pp. 863–870, Taipei, Taiwan, [60] D. Trivedi, A. Lotfi, and C. D. Rahn, “Geometrically exact
October 2010. dynamic models for soft robotic manipulators,” in Proceedings
[47] J. Lock, G. Laing, M. Mahvash, and P. E. Dupont, “Quasistatic of the IEEE/RSJ International Conference on Intelligent Robots
modeling of concentric tube robots with external loads,” in and Systems (IROS ’07), pp. 1497–1502, San Diego, Calif, USA,
Proceedings of the 23rd IEEE/RSJ 2010 International Conference November 2007.
on Intelligent Robots and Systems (IROS ’10), pp. 2325–2332, [61] R. Kang, E. Guglielmino, D. T. Branson, and D. G. Caldwell,
Taipei, Taiwan, October 2010. “Bio-Inspired crawling locomotion of a multi-arm octopus-like
[48] H. Su, D. C. Cardona, W. Shang et al., “A MRI-guided concentric continuum system,” in Proceedings of the IEEE/RSJ International
tube continuum robot with piezoelectric actuation: a feasibility Conference on Intelligent Robots and Systems (IROS ’12), pp. 145–
study,” in Proceedings IEEE International Conference on Robotics 150, Vilamoura, Portugal, 2012.
and Automation, pp. 1939–1945, St. Paul, Minn, USA, 2012. [62] C. Laschi, B. Mazzolai, V. Mattoli, M. Cianchetti, and P. Dario,
[49] R. S. Penning, J. Jung, N. J. Ferrier, and M. R. Zinn, “An eval- “Design of a biomimetic robotic octopus arm,” Bioinspiration
uation of closed-loop control options for continuum manipu- and Biomimetics, vol. 4, no. 1, Article ID 015006, 2009.
lators,” in Proceedings of the IEEE International Conference on [63] M. Rolf and J. J. Steil, “Constant curvature continuum kinemat-
Robotics and Automation, pp. 5392–5397, St. Paul, Minn, USA, ics as fast approximate model for the bionic handling assistant,”
2012. in Proceedings of the IEEE/RSJ International Conference on
[50] A. Degani, H. Choset, A. Wolf, T. Ota, and M. A. Zenati, Intelligent Robot Systems (IROS ’12), pp. 3440–3446, Vilamoura,
“Percutaneous intrapericardial interventions using a highly Portugal, 2012.
articulated robotic probe,” in Proceedings of the 1st IEEE/RAS- [64] J. Ding, K. Xu, R. E. Goldman, P. K. Allen, D. L. Fowler, and
EMBS International Conference on Biomedical Robotics and N. Simaan, “Design, simulation and evaluation of kinematic
Biomechatronics (BioRob ’06), pp. 7–12, Pisa, Italy, February alternatives for insertable robotic effectors platforms in single
2006. port access surgery,” in Proceedings of the IEEE International
Conference on Robotics and Automation (ICRA ’10), pp. 1053–
[51] E. J. Butler, R. Hammond-Oakley, S. Chawarski et al., “Robotic
1058, Anchorage, Alaska, USA, May 2010.
neuro-endoscope with concentric tube augmentation,” in Pro-
ceedings of the IEEE/RSJ International Conference on Intelligent [65] K. Xu, R. E. Goldman, J. Ding, P. K. Allen, D. L. Fowler, and
Robot Systems (IROS ’12), pp. 2941–2946, Vilamoura, Portugal, N. Simaan, “System design of an insertable robotic effector
2012. platform for Single Port Access (SPA) surgery,” in Proceedings of
the IEEE/RSJ International Conference on Intelligent Robots and
[52] L. G. Torres, R. J. Webster III, and R. Alterovitz, “Task-
Systems (IROS ’09), pp. 5546–5552, St. Louis, MO, USA, October
oriented design of concentric tube robots using mechanics-
2009.
based models,” in Proceedings of the IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS ’12), pp. [66] K. Xu, J. Zhao, J. Geiger, A. J. Shih, and M. Zheng, “Design of
4449–4455, Vilamoura, Portugal, 2012. an endoscopic stitching device for surgical obesity treatment
using a N.O.T.E.S approach,” in Proceedings of the IEEE/RSJ
[53] R. J. Webster III, J. M. Romano, and N. J. Cowan, “Kinematics International Conference on Intelligent Robots and Systems:
and calibration of active cannulas,” in Proceedings of the IEEE Celebrating 50 Years of Robotics (IROS ’11), pp. 961–966, San
International Conference on Robotics and Automation (ICRA Francisco, Calif, USA, September 2011.
’08), pp. 3888–3895, Pasadena, Calif, USA, May 2008.
[67] S. Wakimoto and K. Suzumori, “Fabrication and basic exper-
[54] E. Guglielmino, N. Tsagarakis, and D. G. Caldwell, “An octopus iments of pneumatic multi-chamber rubber tube actuator for
anatomy-inspired robotic arm,” in Proceedings of the 23rd assisting colonoscope insertion,” in Proceedings of the IEEE
IEEE/RSJ 2010 International Conference on Intelligent Robots International Conference on Robotics and Automation (ICRA
and Systems (IROS ’10), pp. 3091–3096, Taipei, Taiwan, October ’10), pp. 3260–3265, Anchorage, Alaska, USA, May 2010.
2010.
[68] B. A. Jones, W. McMahan, and I. D. Walker, “Design and analysis
[55] M. B. Pritts and C. D. Rahn, “Design of an artificial muscle of a novel pneumatic manipulator,” in Proceedings of the 3rd
continuum robot,” in Proceedings of the IEEE International IFAC Symposium on Mechatronic Systems, pp. 745–750, Sydney,
Conference on Robotics and Automation, pp. 4742–4746, New Australia, 2004.
Orleans, La, USA, May 2004. [69] W. McMahan, B. A. Jones, and I. D. Walker, “Design and
[56] E. Ayvali and J. P. Desai, “Towards a discretely actuated steerable implementation of a multi-section continuum robot: air-octor,”
cannula,” in Proceedings of the IEEE International Conference on in Proceedings of the IEEE IRS/RSJ International Conference
Robotics and Automation, pp. 1614–1619, St. Paul, Minn, USA, on Intelligent Robots and Systems (IROS ’05), pp. 3345–3352,
2012. Edmonton, Canada, August 2005.
[57] B. A. Jones, M. Csencsits, W. McMahan et al., “Grasping, manip- [70] Y. J. Kim, S. Cheng, S. Kim, and K. Iagnemma, “Design of
ulation, and exploration tasks with the OctArm continuum a tubular snake-like manipulator with stiffening capability by
manipulator,” in Proceedings of the International Conference on layer jamming,” in Proceedings of the IEEE/RSJ International
Robotics and Automation, Orlando, Fla, USA, 2006. Conference on Intelligent Robot Systems (IROS ’12), pp. 4251–
[58] A. Bartow, A. Kapadia, and I. D. Walker, “A novel continuum 4256, Vilamoura, Portugal, 2012.
trunk robot based on contractor muscles,” in Proceedings of [71] A. Sadeghi, L. Beccai, and B. Mazzolai, “Innovative soft robots
the 12th WSEAS International Conference on Signal Processing, based on electro-rheological fluids,” in Proceedings of the
Robotics, and Automation, pp. 181–186, Cambridge, UK, 2013. IEEE/RSJ International Conference on Intelligent Robot Systems
[59] W. McMahan, V. Chitrakaran, M. Csencsits et al., “Field trials (IROS ’12), pp. 4237–4242, Vilamoura, Portugal, 2012.
and testing of “OCTARM” continuum robots,” in Proceedings of [72] N. G. Cheng, M. B. Lobovsky, S. J. Keating et al., “Design and
the IEEE International Conference on Robotics and Automation analysis of a robust, low-cost, highly articulated manipulator
(ICRA ’06), pp. 2336–2341, May 2006. enabled by jamming of granular media,” in Proceedings of the
18 ISRN Robotics

IEEE International Conference on Robotics and Automation, pp. IEEE/RSJ International Conference on Intelligent Robots and
4328–4333, St. Paul, Minn, USA, 2012. Systems: Celebrating 50 Years of Robotics (IROS ’11), pp. 5139–
[73] A. Jiang, G. Xynogalas, P. Dasgupta, K. Althoefer, and T. 5146, San Francisco, Calif, USA, September 2011.
Nanayakkara, “Design of a variable stiffness flexible manipula- [87] D. C. Rucker, B. A. Jones, and R. J. Webster III, “A model for
tor with composite granular jamming and membrane coupling,” concentric tube continuum robots under applied wrenches,” in
in Proceedings IEEE/RSJ International Conference on Intelligent Proceedings of the IEEE International Conference on Robotics
Robots and Systems (IROS ’12), pp. 2922–2927, Vilamoura, and Automation (ICRA ’10), pp. 1047–1052, Anchorage, Alaska,
Portugal, 2012. USA, May 2010.
[74] M. Giorelli, F. Renda, M. Calisti, A. Arienti, G. Ferri, and [88] A. Bajo and N. Simaan, “Finding lost wrenches: using contin-
C. Laschi, “A two dimensional inverse kinetics model of a uum robots for contact detection and estimation of contact
cable-driven manipulator inspired by the octopus arm,” in location,” in Proceedings of the IEEE International Conference on
Proceedings of the IEEE International Conference on Robotics and Robotics and Automation (ICRA ’10), pp. 3666–3673, Anchor-
Automation, pp. 3819–3824, St. Paul, Minn, USA, 2012. age, Alaska, USA, May 2010.
[75] H. Mochiyama and T. Suzuki, “Dynamics modeling of a hyper- [89] I. S. Godage, D. T. Branson, E. Guglielmino, G. A. Medrano-
flexible manipulator,” in Proceedings of the 41st SICE Annual Cerda, and D. G. Caldwell, “Shape function-based kinematics
Conference, pp. 1505–1510, Osaka, Japan, 2002. and dynamics for variable-length continuum robotic arms,” in
[76] H. Mochiyama and T. Suzuki, “Kinematics and dynamics of Proceedings of the IEEE International Conference on Robotics and
a cable-like hyper-flexible manipulator,” in Proceedings of the Automation, pp. 452–457, Shanghai, China, September 2011.
IEEE International Conference on Robotics and Automation, pp. [90] E. Tatlicioglu, I. D. Walker, and D. M. Dawson, “New dynamic
3672–3677, Taipei, Taiwan, September 2003. models for planar extensible continuum robot manipulators,”
[77] H. Mochiyama, “Whole-arm impedance of a serial-chain in Proceedings of the IEEE/RSJ International Conference on
manipulator,” in Proceedings of the IEEE International Confer- Intelligent Robots and Systems (IROS ’07), pp. 1485–1490, San
ence on Robotics and Automation, pp. 2223–2228, Seoul, Korea, Diego, Calif, USA, November 2007.
May 2001. [91] R. Kang, A. Kazakidi, E. Guglielmino et al., “Dynamic model
[78] G. S. Chirikjian and J. W. Burdick, “Modal approach to of a hyper-redundant, octopus-like manipulator for underwater
hyper-redundant manipulator kinematics,” IEEE Transactions applications,” in Proceedings of the IEEE/RSJ International Con-
on Robotics and Automation, vol. 10, no. 3, pp. 343–354, 1994. ference on Intelligent Robots and Systems: Celebrating 50 Years of
[79] I. S. Godage, E. Guglielmino, D. T. Branson, G. A. Medrano- Robotics (IROS ’11), pp. 4054–4059, San Francisco, Calif, USA,
Cerda, and D. G. Caldwell, “Novel modal approach for kine- September 2011.
matics of multisection continuum arms,” in Proceedings of the [92] W. Khalil, G. Gallot, O. Ibrahim, and F. Boyer, “Dynamic
IEEE/RSJ International Conference on Intelligent Robots and modeling of a 3-D serial eel-like robot,” in Proceedings of the
Systems: Celebrating 50 Years of Robotics (IROS ’11), pp. 1093– IEEE International Conference on Robotics and Automation, pp.
1098, San Francisco, Calif, USA, September 2011. 1270–1275, Barcelona, Spain, April 2005.
[80] D. B. Camarillo, C. F. Milne, C. R. Carlson, M. R. Zinn, and J. K. [93] A. Kapadia, E. Tatlicioglu, D. Dawson, and I. D. Walker, “A
Salisbury, “Mechanics modeling of tendon-driven continuum new approach to extensible continuum robot control using the
manipulators,” IEEE Transactions on Robotics, vol. 24, no. 6, pp. sliding-mode,” Computer Technology and Application, vol. 2, no.
1262–1273, 2008. 4, pp. 293–300, 2011.
[81] D. B. Camarillo, C. R. Carlson, and J. K. Salisbury, “Task-space [94] D. Braganza, D. M. Dawson, I. D. Walker, and N. Nath,
control of continuum manipulators with coupled tendon drive,” “Neural network grasping controller for continuum robots,” in
in Experimental Robotics: The 11th International Symposium, O. Proceedings of the 45th IEEE Conference on Decision and Control
Khatib, V. Kumar, and G. Pappas, Eds., pp. 271–280, Springer, (CDC ’06), pp. 6445–6449, San Diego, Calif, USA, December
2009. 2006.
[82] S. Neppalli, M. A. Csencsits, B. A. Jones, and I. D. Walker, [95] N. Giri and I. D. Walker, “Three module lumped element model
“Closed-form inverse kinematics for continuum manipulators,” of a continuum arm section,” in Proceedings of the IEEE/RSJ
Advanced Robotics, vol. 23, no. 15, pp. 2077–2091, 2009. International Conference on Intelligent Robots and Systems:
[83] F. Renda and C. Laschi, “A general mechanical model for Celebrating 50 Years of Robotics (IROS ’11), pp. 4060–4065, San
tendon-driven continuum manipulators,” in Proceedings of the Francisco, Calif, USA, September 2011.
IEEE International Conference on Robotics and Automation, pp. [96] R. S. Penning, J. Jung, J. A. Borgstadt, N. J. Ferrier, and M.
3813–3818, St. Paul, Minn, USA, 2012. R. Zinn, “Towards closed loop control of a continuum robotic
[84] I. A. Gravagne and I. D. Walker, “Manipulability, force, and manipulator for medical applications,” in Proceedings IEEE
compliance analysis for planar continuum manipulators,” IEEE International Conference on Robotics and Automation, pp. 4822–
Transactions on Robotics and Automation, vol. 18, no. 3, pp. 263– 4827, 2011.
273, 2002. [97] T. Zheng, D. T. Branson III, R. Kang et al., “Dynamic continuum
[85] L. G. Torres and R. Alterovitz, “Motion planning for concentric arm model for use with underwater robotic manipulators
tube robots using mechanics-based models,” in Proceedings of inspired by Octopus vulgaris,” in Proceedings of the IEEE
the IEEE/RSJ International Conference on Intelligent Robots and International Conference on Robotics and Automation, pp. 5289–
Systems: Celebrating 50 Years of Robotics (IROS ’11), pp. 5153– 5294, St. Paul, Minn, USA, 2012.
5159, San Francisco, Calif, USA, September 2011. [98] M. Ivanescu, N. Bizdoaca, and D. Pana, “Dynamic control for a
[86] J. Jung, R. S. Penning, N. J. Ferrier, and M. R. Zinn, “A tentacle manipulator with SMA actuators,” in Proceedings of the
modeling approach for continuum robotic manipulators: effects IEEE International Conference on Robotics and Automation, pp.
of nonlinear internal device friction,” in Proceedings of the 2079–2084, Taipei, Taiwan, September 2003.
 ISRN Robotics 19

[99] M. Ivanescu and V. Stoian, “A Variable structure controller Aerospace Conference (AERO ’11), Big Sky, Mont, USA, March
for a tentacle manipulator,” in Proceedings of the 1995 IEEE 2011.
International Conference on Robotics and Automation. Part 1 (of [114] M. Sfakiotakis, A. Kazakidi, N. Pateromichelakis, J. A. Eka-
3), pp. 3155–3160, May 1995. terinaris, and D. P. Tsakiris, “Robotic underwater propulsion
[100] A. Kapadia and I. D. Walker, “Task-space control of extensi- inspired by the octopus multi-arm swimming,” in Proceedings of
ble continuum manipulators,” in Proceedings of the IEEE/RSJ the IEEE International Conference on Robotics and Automation,
International Conference on Intelligent Robots and Systems: pp. 3833–3839, St. Paul, Minn, USA, 2012.
Celebrating 50 Years of Robotics (IROS ’11), pp. 1087–1092, San [115] L. Cowan and I. D. Walker, “The importance of continuous and
Francisco, Calif, USA, September 2011. discrete elements in continuum robots,” International Journal of
[101] M. W. Hannan and I. D. Walker, “Real-time shape estimation Advanced Robot Systems, vol. 10, pp. 1–13, 2013.
for continuum robots using vision,” Robotica, vol. 23, no. 5, pp. [116] A. Kapoor, N. Simaan, and R. H. Taylor, “Suturing in confined
645–651, 2005. spaces: constrained motion control of a hybrid 8-DoF robot,”
[102] B. Weber, P. Zeller, and K. Kuhnlenz, “Multi-camera based in Proceedings of the 12th International Conference on Advanced
real-time configuration estimation of continuum robots,” in Robotics (ICAR ’05), pp. 452–459, Seattle, Wash, USA, July 2005.
Proceedings IEEE/RSJ International Conference on Intelligent [117] A. Kapoor, K. Xu, W. Wei, N. Simaan, and R. H. Taylor, “Telema-
Robot Systems (IROS ’12), pp. 3550–3555, Vilamoura, Portugal, nipulation of snake-like robots for minimally invasive surgery
2012. of the upper airway,” in Proceedings of the 9th International
[103] J. Li and J. Xiao, “Determining “grasping” configurations for a Conference on Medical Image Computing and Computer-Assisted
spatial continuum manipulator,” in Proceedings of the IEEE/RSJ Intervention (MICCAI) Medical Robotics Workshop, 2006.
International Conference on Intelligent Robots and Systems: [118] G. P. Scott, C. G. Henshaw, D. Walker, and B. Willimon,
Celebrating 50 Years of Robotics (IROS ’11), pp. 4207–4214, San “Autonomous robotic refueling of an unmanned surface vehicle
Francisco, Calif, USA, September 2011. in varying sea states,” submitted to Journal of Field Robotics.
[104] L. A. Lyons, R. J. Webster III, and R. Alterovitz, “Planning [119] J. Merino, A. L. Threatt, I. D. Walker, and K. E. Green, “Kine-
active cannula configurations through tubular anatomy,” in matic models for continuum robotic surfaces,” in Proceedings of
Proceedings of the IEEE International Conference on Robotics the IEEE/RSJ International Conference on Intelligent Robots and
and Automation (ICRA ’10), pp. 2082–2087, Anchorage, Alaska, Systems (IROS ’12), pp. 3453–3460, Vilamoura, Portugal, 2012.
USA, May 2010.
[105] J. Xiao and R. Vatcha, “Real-time adaptive motion planning for
a continuum manipulator,” in Proceedings of the 23rd IEEE/RSJ
2010 International Conference on Intelligent Robots and Systems
(IROS ’10), pp. 5919–5926, Taipei, Taiwan, October 2010.
[106] M. Saha and P. Isto, “Motion planning for robotic manipulation
of deformable linear objects,” in Proceedings of the IEEE Interna-
tional Conference on Robotics and Automation (ICRA ’06), pp.
2478–2484, Orlando, Fla, USA, May 2006.
[107] A. Kapadia, E. Tatlicioglu, and I. D. Walker, “Teleoperation
control of a redundant continuum manipulator using a non-
redundant rigid-link master,” in Proceedings of the IEEE/RSJ
International Conference on Intelligent Robots and Systems
(IROS ’12), pp. 3105–3110, Vilamoura, Portugal, 2012.
[108] M. Csencsits, B. A. Jones, W. McMahan, V. Iyengar, and
I. D. Walker, “User interfaces for continuum robot arms,”
in Proceedings of the IEEE IRS/RSJ International Conference
on Intelligent Robots and Systems, IROS 2005, pp. 3011–3018,
Edmonton, Canada, August 2005.
[109] W. McMahan and I. D. Walker, “Octopus-inspired grasp syn-
ergies for continuum manipulators,” in Proceedings of the IEEE
International Conference on Robotics and Biomimetics, pp. 945–
950, Bangkok, Thailand, 2009.
[110] Y. Yekutieli, R. Sagiv-Zohar, B. Hochner, and T. Flash, “Dynamic
model of the octopus arm. II: control of reaching movements,”
Journal of Neurophysiology, vol. 94, no. 2, pp. 1459–1468, 2005.
[111] A. Kapadia and I. D. Walker, “Self-Motion analysis of extensible
continuum manipulators,” in Proceedings of the IEEE Inter-
national Conference on Robotics and Automation, Karlsruhe,
Germany, May 2013.
[112] I. S. Godage, T. Nanayakkara, and D. G. Caldwell, “Locomotion
with continuum limbs,” in Proceedings IEEE/RSJ International
Conference on Intelligent Robot Systems (IROS ’12), pp. 293–298,
Vilamoura, Portugal, 2012.
[113] I. D. Walker, “Continuum robot appendages for traversal of
uneven terrain in in situ exploration,” in Proceedings of the IEEE
 International Journal of

Rotating
Machinery

International Journal of
The Scientific
Engineering Distributed
Journal of
Journal of

Hindawi Publishing Corporation

World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
Sensors
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

http://www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Publishing Corporation Volume 2014
Navigation and
Observation
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
in Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com
http://www.hindawi.com Volume 2014

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014