IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO.

4, OCTOBER 2019 3633

Adaptive Dynamic Control for Magnetically

Actuated Medical Robots
Lavinia Barducci , Giovanni Pittiglio , Joseph C. Norton , Keith L. Obstein , and Pietro Valdastri

Abstract—In the present letter, we discuss a novel dynamic con- is the potential miniaturization; this approach permits to over-
trol approach for magnetically actuated robots, by proposing an come complex and bulky actuation system, achieving minimally
adaptive control technique, robust toward parametric uncertainties invasiveness. This is generally equated to a reduction of patient
and unknown bounded disturbances. The former generally arise
due to partial knowledge of the robots’ dynamic parameters, such discomfort and post-operative recovery time. Miniaturization
as inertial factors, the latter are the outcome of unpredictable is also feasible because functional forces can be maintained
interaction with unstructured environments. In order to show the by balancing an arbitrarily small Internal Permanent Magnets
application of the proposed approach, we consider controlling (IPMs) with a sufficiently large External Permanent Magnet
the magnetic flexible endoscope (MFE), which is composed of a (EPM).
soft-tethered internal permanent magnet (IPM), actuated with a
single external permanent magnet. We provide with experimental Due to these advantages, this class of robot has been inves-
analysis to show the possibility of levitating the MFE—one of the tigated for application to several fields of medicine, from en-
most difficult tasks with this platform—in case of partial knowledge doscopic procedures, such as colonoscopy [1]–[3], gastroscopy
of the IPM’s dynamics and no knowledge of the tether’s behavior. [4] and cardiac applications [5]–[9] to microrobotics [10].
Experiments in an acrylic tube show a reduction of contact of the Magnetic external actuation can vary from coil-based systems
32% compared to non-levitating techniques and 1.75 times faster
task completion with respect to previously proposed levitating [5], [11]–[16], rotating permanent magnet-based devices [17],
techniques. More realistic experiments, performed in a colon phan- [18] and permanent magnet-based systems [1]–[4], [19]. All
tom, show that levitating the capsule achieves faster and smoother these actuation mechanisms share similar control properties, in
exploration and that the minimum time for completing the task is fact, actuation is based on employing the previously mentioned
attained by the proposed approach. actuators for generating forces and torques focused on magnetic
Index Terms—Magnetic robots control, dynamic control, capsule agents. Since the control inputs for these robots are forces and
colonoscopy. torques, it is particularly effecacious to consider a quasi-static
[11], [12] or a dynamic control approach [20]. The latter has
I. INTRODUCTION
the advantage of considering the overall physical properties of
AGNETICALLY actuated robots have been investigated
M during the last decades, particularly in the field of medi-
cal robotics. The main advantage of magnetically actuated robots
the robots and permits faster and more accurate control.
We propose an adaptive dynamic control approach [21], able
to cope with parametric uncertainties, such as inertial factors,
and robust towards the presence of unknown bounded exter-
Manuscript received February 22, 2019; accepted July 1, 2019. Date of nal disturbances. The former are, generally, related to partial
publication July 15, 2019; date of current version July 24, 2019. This letter knowledge of the robots’ mechanical properties, the latter may
was recommended for publication by Associate Editor S. Weiss and Editor be related to unstructured forces arising from the interaction with
P. Rocco upon evaluation of the reviewers’ comments. This work was sup-
ported in part by the Royal Society, U.K., in part by the Engineering and an unknown environment. This control technique employs the
Physical Sciences Research Council, U.K., under Grants EP/P027938/1 and knowledge of the IPM pose, achieved by using an appropriate
EP/K034537/1, and in part by the National Institute of Biomedical Imaging localization technique such as [22] or [23].
and Bioengineering, USA, of the National Institutes of Health under Award
R01EB018992. (Lavinia Barducci and Giovanni Pittiglio contributed equally In order to discuss the application of the proposed tech-
to this letter.) (Corresponding author: Lavinia Barducci.) nique, we focus on the control of the Magnetic Flexible En-
L. Barducci, G. Pittiglio, J. C. Norton, and P. Valdastri are with the STORM doscope (MFE) [1], a innovative minimally invasive platform
Lab UK, Institute of Robotics, Autonomous Systems and Sensing, School
of Electronic and Electrical Engineering, University of Leeds, Leeds LS2 for colonoscopy. We consider the case of actuating a single
9JT, U.K. (e-mail: [email protected]; [email protected]; j.c.norton@leeds. soft-tethered IPM by employing a robotically manipulated EPM.
ac.uk; [email protected]). Moreover, we consider partial knowledge of the mass and dimen-
K. L. Obstein is with the Division of Gastroenterology, Hepatology, and
Nutrition, Vanderbilt University Medical Center, Nashville, TN 37232 USA, and sions of the IPM and no information about the tether. While we
also with the STORM Lab, Department of Mechanical Engineering, Vanderbilt focus on one platform, this proposed method could be applied
University, Nashville, TN 37235 USA (e-mail: [email protected]). to other actuation systems and untethered capsules [17].
This letter has supplementary downloadable material available at http://
ieeexplore.ieee.org, provided by the authors. This video shows the results of Herein, we show successful magnetic levitation [20] which
the experiments, context of the experiments, and the robustness of the control helps overcoming the major issue of previously proposed control
technique in the presence of external disturbances. This video is known to be techniques [24]: continuous attraction between the IPM and
compatible with Windows Media Player, VLC, etc. Contact: Lavinia Barduccie
([email protected]) for further questions about this letter. EPM. Successful levitation can encourage obstacle avoidance
Digital Object Identifier 10.1109/LRA.2019.2928761 and a smoother navigation. It can also result in a reduction of

2377-3766 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
3634 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 4, OCTOBER 2019

Fig. 2. Control scheme.

Fig. 1. Schematic representation of the platform.

Compared to the previously proposed solution [20], we
present a novel approach which takes into consideration the dy-
namics of IPM and deals with possible parametric uncertainties.
pressure applied to the environment which will reduce discom-
The proposed technique is an adaptive control, which considers
fort for the patient and risk of adverse events.
uncertainties such as the mass, dimensions of the IPM and the
This letter is organized as follows: in Section II we provide a
dynamics of the tether. In particular, the IPM mass is strongly
general overview of the method, which is utilised and explored
affected by the tether. In fact, during levitation, a consistent
further in Section III. Sections IV and V present the experimental
section of the tether is lifted. The proposed control strategy
results, comparing the proposed approach with the ones dis-
autonomously modifies its parameters in order to adapt them
cussed in [20], [24]; the former discusses free space levitation in
to the actual system dynamics. Therefore, a further control loop
a L-shaped acrylic tube, the latter reports the results obtained in
has been inserted in the dynamic control in order to achieve
a more realistic environment (a colon phantom). In Appendix A,
convergence of the estimated dynamic parameters.
we discuss the basic concepts of the magnetic actuation, while
This control technique uses capsule localization (100 Hz,
Appendix B reports proofs of lemmas and theorems employed
4 mm accuracy) [22].
in the letter.

II. CONTROL OVERVIEW III. DYNAMIC CONTROL

The dynamic control approach discussed in the following is The control strategy proposed herein is based on the backstep-
achieved using a single EPM controlled by a serial manipula- ping technique which, compared to [4], [24], adds a control loop
tor. Another magnet, referred to as IPM1 is housed within the on the force. This improves controller stability, as discussed in
capsule. This is shown in Fig. 1. [20]. Moreover, as an advancement of [20], we added a further
In this scenario, the most unstable equilibrium is the one along internal loop which estimates the dynamics of the IPM and
the gravity direction. In fact, we need to guarantee that the force takes into account the effect of the tether. The scheme in Fig. 2
on the IPM counteracts gravity in an equilibrium state that is shows the proposed control strategy (solid lines). The external
highly unstable. A dynamic control approach takes into account loop (dashed lines) represents the user interface which is not
all forces that act on the system. In particular, the coupling investigated in the present work.
between magnets is directly expressed in terms of interaction Therefore, the control has three key components: pose control
(generalized) forces; levitation is the outcome of the equilibrium (Section III-A), parameters estimation (Section III-B) and force
of these forces with gravity. control (Section III-C). Pose control, considered as the external
In our previous work [20], we proposed a dynamic control loop, aims to steer the IPM to the desired pose. The parameter
approach that allows the IPM to levitate in a realistic environ- estimation improves the controller properties, since this allows
ment, such as a colon phantom. The main drawback of this the assumptions to be weakened by the knowledge of the system
technique lies in two main assumptions: the desired trajectory dynamics. Force control, referred to also as “internal loop”, aims
was considered as a piece-wise constant function of the time and to converge the actual force to the desired one. The stability of
IPM-tether interactions were assumed negligible. The former this backstepping approach, as shown in Section III-D, guaran-
restricts velocity of the IPM movements, while the latter does tees the overall convergence.
not guarantee convergence in a general scenario. In this letter, Fundamentally, the dynamics of the IPM is subjected to the
we aim to weaken both assumptions by employing an adaptive magnetic interaction between EPM and IPM. The magnetic
dynamic control and by proving ultimately uniform bounded fields are approximated by the dipole model and are consid-
stability of the proposed approach. ered accurate for our purpose: given the geometry and relative
distance between the two magnets, we can infer that the error
1 In the following the soft-tethered capsule is also referred to as Internal does not significantly affect the control of the IPM. Possible
Permanent Magnet. errors related to dipole modelling are discussed in Section IV.

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
BARDUCCI et al.: ADAPTIVE DYNAMIC CONTROL FOR MAGNETICALLY ACTUATED MEDICAL ROBOTS 3635

The magnetic force and torque, exerted by the EPM on the IPM, A. Pose Control
can be written as a vector τm (x, q) ∈ Rn .
As a first step, we define a pose controller that attempts to
Consider the nominal dynamics of the IPM steer the IPM to a desired trajectory xd . We aim to find a set
B(x)ẍ + C(x, ẋ)ẋ + G(x) = τm (x, q), (1) of desired forces and torques, referred to as τd , that steer the
IPM to the desired pose. Compared to [20], we consider partial
where x ∈ Rn is the IPM pose (i.e. position and orientation) and knowledge of the dynamics of the system, using the Adaptive
q ∈ Rm embeds the robot joint variables; B(x), C(x, ẋ), G(x) Backstepping Control [21], [26].
are referred to as inertia, Coriolis matrix and gravity [25] of the The control law can be determined directly through a standard
IPM, respectively. Our aim is to find q such that x approaches a Lyapunov approach, by defining
desired value xd . τd = B̂(x)ẍr + Ĉ(x, ẋ)ẋr + Ĝ(x) − Kd s − x̃
This is achieved in two steps: first the value of the desired (5)
torque (τd ) is found for x → xd , considering the dynamics of = Y (x, ẋ, ẋr , ẍr )π̂ − Kd s − x̃
the unknown parameters, then we define q̇ for which τm → τd ,
where B̂, Ĉ and Ĉ are the estimated dynamic matrices, whose
according to the dynamics of the force and torque
parameters are embedded in π̂. The position error of the IPM
∂τ (x, q) ∂τ (x, q) is defined as x̃ = xd − x and s = x̃˙ + Λx̃ = ẋ − (ẋd − Λx̃) =
τ̇m = ẋ + q̇ = Jx ẋ + Jq q̇. (2) ẋ − ẋr , with Λ symmetric, positive definite design matrix; ẋr is
∂x ∂q
referred to as the reference velocity, being the velocity the IPM
The analytical computation of the matrices Jx and Jq is thor- is controlled to.
oughly explained in Appendix I. The variables q̇ can be inte- The present control loop guarantees x → xd , as τ → τd . This
grated to control the robot through its Direct Kinematics (DK) statement holds under the following assumption.
[25]. The novelty of our control system, compared to [24], is that Assumption 1: The steering of the IPM is achieved under
we apply a closed-loop control on τm , as in [20]. Compared to these conditions:
[20], we introduced a further control loop in which we guarantee r the force control, described in Section III-C, is faster than
the convergence of the unknown parameters of the dynamic the system dynamics in (1);
system. r the unknown parameters vector π in (3) is constant.
The proposed approach takes into consideration how the tether The former leads to consider almost instantaneous conver-
can affect the dynamics of the IPM. It is herein considered gence of force and torque; in fact, we assume there exists an
an unmodelled disturbance on the IPM dynamics, in order to instant T , 0 < T  1, such that τ (t) = τd (t), for any t ≥ T .
underline the robustness of the proposed approach. However, we This assumption is needed to prove Lemma 1 on which the final
show the stability of the proposed technique (Theorem 1) also in proof of this work (Theorem 1) is based. Furthermore, the need
absence of the tether, as in the case of untethered capsules [4]. We for π = const is not limiting, since the inertial, Coriolis and
do not consider the case of known tether properties since, even gravity parameters do not generally vary over time.
in the case tether dynamics can be predicted, interaction with Lemma 1: Under Assumption 1, the pose controller in (5)
the environment would confound them. Therefore, we consider achieves asymptotic stability of the error x̃, for any positive
the most general case of dynamic control of a single IPM. definite design gains Kd and Λ.
In order to consider possible parametric uncertainties, embed- Appendix B includes further details on this.
ded in the parameters vector π ∈ Rp , we rewrite the dynamics
in (1) as B. Parameters Estimation

B(x)ẍ + C(x, ẋ)ẋ + G(x) = Y (x, ẋ, ẍ)π, (3) This internal loop estimates the unknown parameters of the
IPM dynamics, such as the mass and the dimensions of the IPM;
where Y (x, ẋ, ẍ) ∈ Rn×p is the dynamic regressor, [25]. The this allows us to adapt our controller to the real dynamics of the
update law of π allows the unknown parameters to converge to system.
their real values, guaranteeing the robust asymptotic stability of The control law is derived from the Lyapunov theory, defining
the overall system. Appendix B describes this in more detail. π̃˙ = π̇ − π̂˙ = −π̂˙ = uπ , under the assumption that the unknown
The overall dynamics of the system we aim to control reads parameters vector π is constant. The control law reads as
as uπ = R−1 Y T (x, ẋ, ẋr , ẍr )s (6)

Y (x, ẋ, ẍ)π = τ where R is a positive, definite designed gain. The choice for uπ
, (4)
τ̇ = Jx ẋ + Jq q̇ + ν̇ is justified by the proof of Lemma 1, reported in Appendix B.
where ν models the tether interaction with the environment
C. Force Control
(such as: drag, elastic behaviour, friction and colon motions)
and π embeds the uncertain parameters of the IPM, such as the As the third step, we design a controller that ensures the
mass, the length and the diameter. The localization method [22] magnetic force (τm ) converges on the desired force (τd ). Accord-
measures x and ẋ, while the robot joints are measured by the ing to Assumption 1, this loop is required to converge almost
embedded encoders. instantaneously. The magnetic force and torque are computed

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
3636 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 4, OCTOBER 2019

from x and q by employing the localization output and dipole TABLE I

MEAN ERROR OF THE DIPOLE MODEL
model.
According to (2), the choice for
q̇ = Jq† (τ̇d + K τ̃ − Jx ẋ). (7)
yields to
τ̃˙ = τ̇d − τ̇m = −K τ̃ , (8)
with K positive definite design gain [20]. This leads to asymp-
totic stability of the force and torque error dynamics.
Lemma 2 (from: [20]): Any choice for the positive definite
gain K achieves stability of the torque dynamics, if ν = 0.

D. Overall Control
In the following, we describe the overall control strategy by
considering the previous sections. We show that the new choice
of q̇
⎧
⎨ τd = Y (x, ẋ, ẋr , ẍr )π̂ − Kd s − x̃
⎪
q̇ = Jq† (τ̇d + K τ̃ − Jx ẋ − ẋ) , (9)
⎪
⎩
uπ = R−1 Y T s
leads to
τ̃˙ = −K τ̃ + ẋ,
which achieves overall convergence. This is is discussed in the
following theorem.
Theorem 1: Under the assumption π = const, the controller
defined in (9) attains, for any positive definite design gains Kp , Fig. 3. Sensorised platform. (a) EPM, (b) IPM, (c) Environment (acrylic tube),
(d) Force/Torque sensor, (e) Top acrylic sheet (constrained in negative z) and
Kd and K, (f) One of the ball transfer units.
a) asymptotic stability of x̃ if ν  0;
b) ultimately uniformly bounded error x̃ if ν is piece-wise
constant. 60 mm and a 90 degrees bend in the middle. Each half of the
This is discussed in Appendix B, where we underline that, tube has a length of 250 mm and the first part is inclined by
even in the presence of unknown unmodelled disturbances re- approximately 20 mm over its length. In this case, the desired
lated to the tether dynamics, stability of the error is ensured. trajectory is a pre-planned path since we aim to objectively
Moreover, in the absence of disturbances (e.g. untethered IPMs), evaluate the levitating performance, without the user in the loop.
asymptotic stability is guaranteed. We compared this control approach with the techniques pro-
posed in [20] and [24]. The latter imposes a continuous force
IV. EXPERIMENTAL ANALYSIS: FREE SPACE LEVITATION along the gravity direction to maintain the magnetic coupling
The goal of this experimental work is to validate the con- between the two magnets and therefore, imposes continuous
trol strategy and show that IPM height (i.e. levitation) can be contact with the environment; the former is able to levitate the
controlled and compare its performance with the two previous IPM, but with limited velocity, due to the drawn assumptions.
control strategies mentioned in this letter [20], [24]. We performed 5 trials inside the tube with each method. A video
The IPM is a cylindrical permanent magnet with an axial of the experiments is reported in the attached media of the letter.
magnetization of 1.48 T (N52), diameter and length of 10.16 During the tests, we controlled the IPM to stay in the center of
mm and a mass of 15 g. The EPM is a permanent magnet with the lumen on the x − y plane and to maintain levitation on the z
a diameter and length of 101.6 mm and an axial magnetization axis. In the first tract of the tube, the main challenge for the
of 1.48 T (N52). The EPM is attached to the flange of a serial controller is levitating the IPM, while in the second half of
manipulator (KUKA LBR Med robot).2 the tube, the stiffness of the tether and tube causes the IPM
Table I reports the errors related to the dipole model, con- to maintain contact with the wall of the tube. However, the
sidering the mean distance between the two magnets during the experiments show that the current control technique and the
experiments, described in [27]. technique used in [20] are both able to resume IPM levitation
To show how our control performs, we chose to navigate the after the disruption of moving past the corner.
IPM through an acrylic tube (Fig. 3) with an inner diameter of To give a quantitative indication of the IPM’s contact with
the environment and, crucially, be able to compare the three
2 https://www.kuka.com/en-de/industries/healthcare/kuka-medical-robotics control strategies, a custom sensorised force platform (Fig. 3)

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
BARDUCCI et al.: ADAPTIVE DYNAMIC CONTROL FOR MAGNETICALLY ACTUATED MEDICAL ROBOTS 3637

Fig. 4. 3D tracking.

Fig. 5. Overview of the IPM-tube contact. (a) Adaptive backstepping control. (b) Gravity compensating PD control. (c) PD control.

was used. The force sensor (6 axis Force/Torque sensor, Nano17,

ATI) was secured between two acrylic sheets, with the top sheet
constrained in the negative z direction but allowed to move freely
in x, y and positive z due to the ball transfer units used to support
it. The sensor was used to precisely record (via a National
Instruments cDAQ 9171, 1 kHz sampling frequency) all forces
acting on the acrylic sheet through the environment (i.e. acrylic
tube). This setup was chosen to allow flexibility in the tube layout
and the use of an extremely precise sensor (resolving down to
approximately 0.003 N) that could be damaged if connected to
an unconstrained platform.
In Fig. 4, we report the 3D trajectories of the IPM for each
approach, while in Fig. 5 we report the force on the z direction
(gravity direction), the norm of the lateral force (along x and y Fig. 6. Setup for colon phantom experiments.
axis) and the overall contact between the IPM and the tube wall,
expressed as norm of the force vector.
The results show that the amount of contact with the current into account force in the x and y direction) is significantly larger
method is comparable to the results obtained with the gravity than the force in the z direction with the Adaptive Backstepping
compensating PD approach, but is reduced with respect to control and the Gravity compensating PD control approaches.
the amount of contact achieved with the PD method. In fact, the This also underlines the robustness of the proposed approach
force along the gravity direction is significantly lower with the towards significant interaction between the tether and the en-
first two approaches. Fig. 5(a) shows a negative mean value for vironment. On-the-other-hand, this issue is likely to be less
the force with the Adaptive backstepping control and the Gravity significant in a more flexible environment, such as the colon.
compensating PD control. This is due to the force transmitted Therefore, the method is expected to attain a more satisfactory
by the tether on the negative z axis. In general, we can infer that performance.
the IPM does not have contact with the tube wall; the limited Moreover, in Fig. 4 we can notice that, with respect to the
contact with the tube is indicated from the standard deviation. Gravity compensating PD control, the Adaptive backstepping
Moreover, due to the stiffness of the acrylic tube and the control is able to maintain a better levitation also after the corner
interaction of the tether, the norm of the overall force (i.e. taking (where the rigidity of the tube and the stiffness of the tether affect

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
3638 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 4, OCTOBER 2019

Fig. 7. Trials on the colon simulator.

the behaviour of the IPM). The gravity compensating PD control, assistance with overcoming tether drag). However, since the cap-
instead, has more difficulty to exert necessary force to levitate sule is levitating, the friction related to environmental interaction
the IPM while trying to reduce the IPM-tube contact. is reduced and feeding the tether is more effective. A video of the
Concerning the time to travel the tube and to resume the experiment performed with the Adaptive backstepping control
IPM levitation after the corner, we can infer that the Adaptive is reported in the attached media of the letter.
Backstepping control is faster than the gravity compensating PD During these experiments, the sensorised platform was not
method: with the Adaptive Backstepping control the IPM was used to measure the force that the IPM exerts on the environ-
able to traverse the tube in a mean time of 72 s with a standard ment. This was because the transmitted forces (from the IPM
deviation of 9.1 s, while the PD controller reports a mean time interactions to the sensor) were sufficiently low to be comparable
of 126.2 s with standard deviation of 23 s. with background noise. The attributing factor being the highly
deformable environment that absorbs (dissipates) the low (<1 N)
V. EXPERIMENTAL ANALYSIS: COLON PHANTOM contact forces. In the future, to investigate the real performance
To show the practical feasibility of this approach we provide of the proposed approach, a deeper analysis will be performed
experiments in a more realistic environment. For this purpose, with expert users.
we used the M40 Colonoscope Training Simulator3 in standard A supplementary video shows the robustness of the control
configuration. As the previous set of the experiments in Section technique in the presence of external disturbances.
IV, we compared the current approach with the method used in
[20] and the PD control computed in [24].
VI. CONCLUSION
We performed 5 trials with each approach to compare all
techniques. The user (with no prior endoscopic experience, but The present letter discusses a dynamic approach for the con-
knowledge of the platform) was able to guide the IPM to traverse trol of the magnetically actuated medical robots. In particular, we
the colon from the end of the sigmoid to the caecum. The IPM show the application of the proposed technique to the MFE [1].
is equipped with a camera that provides a visual feedback to the We prove that proposed method is enough accurate to achieve
user, which manipulates the IPM’s desired pose (xd ) with a 3D levitation of the IPM, which is one of the most complex tasks
mouse, as shown in Fig. 6. with this type of platforms. Moreover, levitating the IPM leads
In Fig. 7, we report the fastest trial with each approach. These to reduced contact with the environment, avoiding obstacles
experiments show that the current approach is able to reduce the and folds. This can aid locomotion and reduce tissue stress (i.e.
mean completion time for the overall task (a path of approx- patient discomfort and risk of trauma).
imately 0.85 m). The Adaptive Backstepping control achieves The control strategy is based on the Adaptive Backstepping
a mean completion time of 248.5 s with standard deviation of Control, which facilitates IPM levitation. The novelty of the
31.8 s; the mean time achieved by the gravity compensating PD current approach is the fact that we take into account the dy-
control [20] was 306.3 s with a standard deviation of 69.6 s; the namics of the IPM, considering all the uncertainties the system
PD control [24], instead, had a mean completion time of 551.9 s is subjected to. This overcomes some of the assumptions drawn
with standard deviation of 138.4 s. Moreover, since we set a in [20]. In particular, the assumption that the desired trajectory
maximum time for each trial of 600 s, the PD control failed two is a piece-wise constant function of the time is weakened. This
times over all five trials - the ceacum was not reached on time. allows an increase in the velocity of the IPM, even if this
In order to levitate the capsule, the force exerted on the IPM is always subjected to the limitation of the IPM localization
along the z axis is reduced, compared to [24]; this leads to a lower algorithm. In fact, the current localization frequency (100 Hz) is
functional steering force and the need for feeding the tether (i.e. not fast enough to guarantee that the IPM dynamics is handled
completely and so increasing this would have a direct impact on
3 https://www.kyotokagaku.com/products/detail01/m40.html system performance.

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
BARDUCCI et al.: ADAPTIVE DYNAMIC CONTROL FOR MAGNETICALLY ACTUATED MEDICAL ROBOTS 3639

To prove the strength of the current approach we performed APPENDIX B

two sets of experiments. The first set of experiments was devel- PROOFS OF LEMMAS AND THEOREMS
oped in free space, using a sensorised platform to measure the In the following we provide the proofs of Lemma 1 and
forces the IPM exerted on the environment. The measured force
Theorem 1.
can be read as a measure of the amount of contact between the
Proof of Lemma: Consider the positive definite Lyapunov
IPM and the tube wall. The results demonstrate that our method function
is able to reduce the contact with the surrounding and increase
the velocity of the IPM respect to the previous solutions [20], 1 T 1 1
W (x̃, s, π̃, t) = x̃ x̃ + sT B(x)s + π̃ T Rπ̃.
[24]. 2 2 2
To prove the feasibility of our approach in a more realistic with R is a positive definite gain. Since the unknown parameters
environment, we performed the second set of experiments in are assumed constant, their error dynamics is π̃˙ = π̇ − π̂˙ =
a colon phantom. These proved that with the current strategy −π̂˙ = −uπ , for some choice for the update law uπ . The time
def
the IPM was able to traverse the colon from the sigmoid to the derivative of the chosen Lyapunov function reads as
caecum with a reasonable completion time. Prior work using a
1
latex phantom and Olympus colonoscope demonstrates a median Ẇ = x̃T x̃˙ + sT B(x)ṡ + sT Ḃ(x)s − π̃ T Ruπ
cecal intubation time (i.e. travelling from anus to cecum) of 2
150.19 s (115.68, 197.48) for 4 experted physicians [28]. Even 1
= x̃T x̃˙ + sT (B(x)ẍ − B(x)ẍr ) + sT Ḃ(x)s − π̃ T Ruπ
if the proposed method is slower compared to the standard 2
technique, it is worth mentioning that the lower invasiveness of
the MFE has the potential for eliminating the need for anesthesia Summing and subtracting sT Cs we obtain
and so achieve an overall shorter procedure and recovery time. Ẇ = x̃T x̃˙ + sT (−B(x)ẍr − C(x, ẋ)ẋr − G(x) + τ )
In the future, we will investigate the performance of the
current approach in a more realistic experimental setting (i.e 1
+ sT (Ḃ(x) − 2C(x, ẋ))s − π̃ T Ruπ .
animal and cadaver models). Moreover, we aim to investigate 2
the application of the proposed control approach to multiple Due to the skew-symmetry of Ḃ(x) − 2C(x, ẋ), ṡT (Ḃ(x) −
IPMs. 2C(x, ẋ))ṡ = 0 [25]. We define the generalized forces as

APPENDIX A τ = B̂(x)ẍr + Ĉ(x, ẋ)ẋr + Ĝ(x) − Kd s − x̃.

MAGNETIC ACTUATION thus
Consider the vector between End Effector (EE) position (pE )
Ẇ = sT Y π̃ − sT Kd s − x̃T Λx̃ − π̃ T Ruπ .
and IPM position (pI ), referred to as p = pE − pI . According
to the dipole model, the (generalized) force between EPM and The update law uπ = R−1 Y T s, for the parameters dynamics,
IPM is yields to
 3M 
||p||4 (m̂E m̂I + m̂I m̂E + (m̂I Z m̂E )I)p̂ Ẇ = −sT Kd s − x̃T Λx̃
T T T
τm =
||p||3 m̂I × D m̂E
M
and leads to conclude for, at least, uniform stability of the origin.
According to La Salle-Yoshizawa’s theorem [29], since
where M = μ0 ||mI4π || ||mE ||
with mI = ||mI ||m̂I and mE = limt→∞ Ẇ (x̃, s, π̃) = 0, limt→∞ x̃ = 0 and asymptotic stability
||mE ||m̂E magnetic moments of IPM and EPM, respectively; of x̃ is guaranteed. 
μ0 = 4π10−7 AN2 permeability of vacuum, p̂ = ||p|| p
, Z=I− Proof of Theorem: Consider the positive definite Lyapunov
3×3
5p̂p̂ and D = 3p̂p̂ − I. Herein I ∈ R
T T
is referred to as the function
identity matrix and || · || is the Euclidean norm. 1
We consider the time derivative of τm [20] V (x̃, s, π̃, t, τ̃ ) = W (x̃, s, π̃, t) + τ T τ ,
2
ṗE ṗI where W (x̃, s, π̃, t) is the Lyapunov function defined in the
τ̇m = ∂τm ∂τm
˙E − ∂τm ∂τm
˙I proof of Lemma 1 and τ = τd − τ . The time derivative of the
∂p ∂ m̂E m̂ ∂p ∂ m̂I m̂
  chosen Lyapunov function is
I 03,3
= J q̇ V̇ (x̃, s, π̃, t, τ̃ ) = −sT Kd s − x̃T Λx̃ + (τ̃ + ν)T τ̃˙
∂τm ∂τm
∂p ∂ m̂E
03,3 (m̂E )T×
  = −sT Kd s − x̃T Λx̃ − τ̃ T K τ̃ − ν T K τ̃ ,
I 03,3
− ∂τm
∂p
∂τm
∂ m̂I ẋ under the assumption ν̇  0.
03,3 (m̂I )T×
First, we prove statement (a). In absence of disturbances, i.e.
= Jq q̇ + Jx ẋ, ν = 0, by applying the La Salle-Yoshizawa’s theorem [29], we
can conclude for asymptotic stability of x̃, as in the proof of
with J manipulator’s Jacobian matrix. Lemma 1.

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.
3640 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 4, OCTOBER 2019

We can also investigate the uniform ultimate boundedness of [11] J. Edelmann, A. J. Petruska, and B. J. Nelson, “Estimation-based control
x̃ (statement (b)), by showing that for any 0 < θ < 1, of a magnetic endoscope without device localization,” J. Med. Robot.
Res., vol. 3, no. 1, Mar. 2018, Art. no. 1850002. [Online]. Available:
V̇ (x̃, s, π̃, t, τ̃ ) = (1 − θ)(−sT Kd s − x̃T Λx̃ − τ̃ T K τ̃ ) http://www.worldscientific.com/doi/abs/10.1142/S2424905X18500022
[12] A. J. Petruska and B. J. Nelson, “Minimum bounds on the number of
electromagnets required for remote magnetic manipulation,” IEEE Trans.
+ θ(−sT Kd s − x̃T Λx̃ − τ̃ T K τ̃ ) − ν̇ T K τ̃ , Robot., vol. 31, no. 3, pp. 714–722, Jun. 2015.
[13] C. Chautems, A. Tonazzini, D. Floreano, and B. J. Nelson, “A variable
therefore, stiffness catheter controlled with an external magnetic field,” in Proc.
˜ ∀ 0 < θ < 1,
V̇ (x̃, s, π̃, t, τ̃ ) ≤ λ(Kd , Λ, K)||ξ|| IEEE/RSJ Int. Conf. Intell. Robots Syst., 2017, pp. 181–186. [Online].
Available: http://ieeexplore.ieee.org/document/8202155/
[14] T. Greigarn, R. Jackson, T. Liu, and M. C. Çavuşoğlu, “Exper-
if imental validation of the pseudo-rigid-body model of the MRI-
˜ = −λ(K)||ν̇|| def actuated catheter,” in Proc. IEEE Int. Conf. Robot. Autom., May 2017,
||ξ|| = μ. pp. 3600–3605.
θλ(Kd , Λ, K) [15] S. Jeon et al., “A magnetically controlled soft microrobot steering a
guidewire in a three-dimensional phantom vascular network,” Soft Robot.,
Here ξ˜ = (sT x̃T τ̃ T )T and λ(A1 , A2 , . . . , Al ) is referred as the vol. 6, no. 1, pp. 54–68, 2019. [Online]. Available: https://doi.org/10.
maximum eigenvalue of the matrices A1 , A2 , . . . , Al . 1089/soro.2018.0019
[16] J. Sikorski, I. Dawson, A. Denasi, E. E. Hekman, and S. Misra, “Intro-
Therefore, the ultimate bound is μ.  ducing BigMag—A novel system for 3D magnetic actuation of flexible
surgical manipulators,” in Proc. IEEE Int. Conf. Robot. Autom., 2017,
ACKNOWLEDGMENT pp. 3594–3599.
[17] S. Yim and M. Sitti, “Design and rolling locomotion of a magnetically
Any opinions, findings, conclusions, or recommendations actuated soft capsule endoscope,” IEEE Trans. Robot., vol. 28, no. 1,
pp. 183–194, Feb. 2012.
expressed in this material are those of the authors and do not [18] P. Ryan and E. Diller, “Magnetic actuation for full dexterity microrobotic
necessarily reflect the views of the Royal Society, the Engineer- control using rotating permanent magnets,” IEEE Trans. Robot., vol. 33,
ing and Physical Sciences Research Council, or the National no. 6, pp. 1398–1409, Dec. 2017.
[19] C. Tremblay, B. Conan, D. Loghin, A. Bigot, and S. Martel, “Fringe field
Institutes of Health. navigation for catheterization,” in Proc. 6th Eur. Conf. Int. Federation Med.
Biol. Eng., 2015, pp. 379–382.
REFERENCES [20] G. Pittiglio et al., “Magnetic levitation for soft-tethered capsule
colonoscopy actuated with a single permanent magnet: A dynamic control
[1] A. Arezzo et al., “Experimental assessment of a novel robotically- approach,” IEEE Robot. Autom. Lett., vol. 4, no. 2, pp. 1224–1231, Apr.
driven endoscopic capsule compared to traditional colonoscopy,” Diges- 2019.
tive Liver Disease, vol. 45, no. 8, pp. 657–662, 2013. [Online]. Available: [21] J. Zhou and C. Wen, Adaptive Backstepping Control of Uncertain Sys-
http://dx.doi.org/10.1016/j.dld.2013.01.025 tems: Nonsmooth Nonlinearities, Interactions or Time-Variations. Berlin,
[2] P. Valdastri et al., “Wireless therapeutic endoscopic capsule: In vivo Germany: Springer, 2008.
experiment,” Endoscopy, vol. 40, no. 12, pp. 979–982, Dec. 2008. [22] A. Z. Taddese, P. R. Slawinski, M. Pirotta, E. De Momi, K. L. Obstein,
[Online]. Available: http://www.thieme-connect.de/DOI/DOI?10.1055/s- and P. Valdastri, “Enhanced real-time pose estimation for closed-loop
0028-1103424 robotic manipulation of magnetically actuated capsule endoscopes,” Int.
[3] P. Valdastri et al., “Magnetic air capsule robotic system: Proof of con- J. Robot. Res., vol. 37, no. 8, pp. 890–911, 2018. [Online]. Available:
cept of a novel approach for painless colonoscopy,” Surgical Endoscopy, https://doi.org/10.1177/0278364918779132
vol. 26, no. 5, pp. 1238–1246, 2012. [Online]. Available: https://doi. [23] M. Turan, J. Shabbir, H. Araujo, E. Konukoglu, and M. Sitti, “A deep
org/10.1007/s00464-011-2054-x learning based fusion of RGB camera information and magnetic lo-
[4] A. W. Mahoney and J. J. Abbott, “Five-degree-of-freedom manipulation calization information for endoscopic capsule robots,” Int. J. Intell.
of an untethered magnetic device in fluid using a single permanent magnet Robot. Appl., vol. 1, no. 4, pp. 442–450, Dec. 2017. [Online]. Available:
with application in stomach capsule endoscopy,” Int. J. Robot. Res., vol. 35, https://doi.org/10.1007/s41315-017-0039-1
no. 1/3, pp. 129–147, 2016. [24] A. Z. Taddese, P. R. Slawinski, K. L. Obstein, and P. Valdastri, “Non-
[5] C. Chautems and B. J. Nelson, “The tethered magnet: Force and 5-DOF holonomic closed-loop velocity control of a soft-tethered magnetic cap-
pose control for cardiac ablation,” in Proc. IEEE Int. Conf. Robot. Autom., sule endoscope,” in Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst.,
2017, pp. 4837–4842. Oct. 2016, pp. 1139–1144. [Online]. Available: http://ieeexplore.ieee.org/
[6] M. N. Faddis et al., “Novel, magnetically guided catheter for endocardial document/7759192/
mapping and radio frequency catheter ablation,” Circulation, vol. 106, [25] B. Siciliano, L. Sciavicco, L. Villani, and G. Oriolo, Robotics: Mod-
no. 23, pp. 2980–2985, 2002. elling, Planning and Control, 1st ed. Berlin, Germany: Springer,
[7] S. Toggweiler et al., “Management of vascular access in transcatheter 2008.
aortic valve replacement: Part 2: Vascular complications,” JACC, Cardio- [26] H. K. Khalil, Nonlinear Systems. 3rd ed. London, U.K.: Pearson Education
vascular Interventions, vol. 6, no. 8, pp. 767–776, 2013. Int., Dec. 2000.
[8] S. K. Hilai, W. Jost Michelsen, J. Driller, and E. Leonard, “Magneti- [27] A. J. Petruska and J. J. Abbott, “Optimal permanent-magnet geometries
cally guided devices for vascular exploration and treatment,” Radiology, for dipole field approximation,” IEEE Trans. Magn., vol. 49, no. 2,
vol. 113, no. 3, pp. 529–540, 1974. [Online]. Available: https://doi.org/ pp. 811–819, Feb. 2013.
10.1148/113.3.529 [28] K. L. Obstein et al., “Evaluation of colonoscopy technical skill levels
[9] S. Ernst et al., “Initial experience with remote catheter ablation us- by use of an objective kinematic-based system,” Gastrointestinal En-
ing a novel magnetic navigation system,” Circulation, vol. 109, no. 12, doscopy, vol. 73, no. 2, pp. 315–321, 2011. [Online]. Available: http://dx.
pp. 1472–1475, 2004. [Online]. Available: https://www.ahajournals. doi.org/10.1016/j.gie.2010.09.005
org/doi/abs/10.1161/01.CIR.0000125126.83579.1B [29] J. T. Spooner, M. Maggiore, R. Ordonez, and K. M. Passino, Stable
[10] E. Diller, J. Giltinan, G. Z. Lum, Z. Ye, and M. Sitti, “Six-degree-of- Adaptive Control and Estimation for Nonlinear Systems—Neural and
freedom magnetic actuation for wireless microrobotics,” Int. J. Robot. Fuzzy Approximator Techniques. Hoboken, NJ, USA: Wiley, 2002.
Res., vol. 35, no. 1/3, pp. 114–128, 2016.

Authorized licensed use limited to: INRIA. Downloaded on November 14,2023 at 16:09:49 UTC from IEEE Xplore. Restrictions apply.