See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/301347901

Two-Agent Formation Control of Magnetic Microrobots

Conference Paper · July 2016

DOI: 10.1109/MARSS.2016.7561726

CITATIONS READS
10 217

2 authors:

Mohammad Salehizadeh Eric D. Diller

Harvard University University of Toronto
16 PUBLICATIONS   168 CITATIONS    76 PUBLICATIONS   2,114 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

untethered undulatory swimming using a smart soft magnetic composite sheet View project

3D untethered magnetic microgripper View project

All content following this page was uploaded by Mohammad Salehizadeh on 23 November 2017.

The user has requested enhancement of the downloaded file.

This is a post peer-review version of the manuscript. For the final version, please refer to the IEEE Xplore database.

Two-Agent Formation Control of Magnetic

Microrobots
Mohammad Salehizadeh Eric Diller
Department of Mechanical and Department of Mechanical and
Industrial Engineering Industrial Engineering
University of Toronto University of Toronto
Ontario, CANADA Ontario, CANADA
Email: msalehi@mie.utoronto.ca Email: ediller@mie.utoronto.ca

Abstract—This paper presents a new method to control multi- have been explored: Martel et al. [12] achieved swarm control
ple micro-scale magnetic agents operating in close proximity to of bacterial actuators in the human microvasculature track-
each other for applications in microrobotics. Controlling multiple able by a clinical MRI system. Diller et al. [8] employed
magnetic microrobots close to each other is difficult due to
magnetic interactions between the agents, and here we seek to geometrically or magnetically distinct microrobots to realize
control those interactions for the creation of desired multi-agent independent control of small teams of magnetic agents. Becker
formations. We use the fact that all magnetic agents orient to et al. [13] exploited differences in cell population to steer
the global input magnetic field to modulate the local attraction- cells to goal positions using ensemble control. Nevertheless,
repulsion forces between nearby agents. Here we study these all these studies have been done only for a small number of
controlled interaction magnetic forces for agents at a water-air
interface and devise two controllers to regulate the inter-agent agents with severe limitations such as a lack of path-following
spacing and heading of the set, for motion in two dimensions. capability.
Simulation and experimental demonstrations show the feasibility These studies on magnetic micro-agents are also limited
of the idea and its potential for the completion of complex tasks in how close the agents can operate to one another. When
using teams of microrobots. Average tracking error of less than they do operate close together, the agents exert large magnetic
73 µm and 14◦ is accomplished for the regulation of the inter-
agent space and the pair heading angle, respectively, for identical interaction forces on one another, which results in control
disk-shape agents with nominal radius of 500 µm and thickness instability. Such interactions have been used to attach nearby
of 80 µm operating within several body-lengths of each other. magnet micro-agents together [14], but have not allowed
for independent agent operation in close proximity without
I. I NTRODUCTION coming into contact. Most work in the field of magnetic
Microrobots can effectively access small remote spaces with micro-agents assume that inter-agent magnetic fields are small
a wide range of potential applications in drug delivery [1], in comparison with the driving actuation field strength, with
cell lysis/sorting [2], and micro-assembly/disassembly [3]. The the associated requirement that the agents be kept far apart
ability to exert independent control over a team of microrobots from each other (typically several microrobot body-lengths).
working together on a task has potential to increase task This constraint limits the ability of teams of agents from
speed and capability [4]. Among many proposed strategies [5], working close together. In Miyashita et al. [15], local magnetic
remote actuation using a magnetic field is a common choice interaction forces are used to create a few stable formations in
because it can penetrate most materials, remotely generate both two dimensions of a set. These formations are modulated by
forces and torques on magnetic materials, and is easy and safe dynamically remagnetizing some of the agents, with promise
to generate and manipulate [4], [6]. However, swarm control for controlling sets of up to three agents and limited control
of magnetic micro-agents remains an open-ended problem as over a four-agent set. However, that method is limited to a
in most actuation systems, all magnetic micro-agents share small set of stable configurations, has no control over the
a global driving magnetic signal. In this way, all agents formation orientation, cannot be generalized to microrobots
receive identical control inputs and thus it is difficult to steer moving in three dimensions, and is only applicable to sets
independently for complex task completion [7], [8]. Outside of agents which are each magnetically unique (and so cannot
the microrobotics field, several approaches to particle assem- scale up to larger sets of agents).
bly have been explored. For instance, fluidic interactions are Here we pursue a more general approach for the dynamic
considered to achieve dynamic self-assembly of magnetized regulation of the inter-agent magnetic forces between nearby
objects such as magnetic disks [9], gears [10], and colloidal microrobots. We model the radial and transverse magnetic
asters [11]. These particles rotate at a liquid-air interface with forces between these agents, and devise a set of controllers to
complex behaviors and motions that are not possible with maintain any desired inter-agent separation and pair heading
conventional systems, but are limited to two-dimensional (2D) in the presence of these forces. We show for the first time, the
applications. In the microrobotics field, a variety of approaches stable motion control of multiple identical magnetic micro-
 y
50
radial Fr
global frame Ba counter-clockwise transverse Ft
z

Magnetic force Fm (nN)

repulsion repulsion
x ψG 0
x

zero radial-force angle

Ba clockwise
et
-50
local
frame ψG −φ attraction attraction
ψL =

−54.72◦
er zero radial-force
Ft r 12 -100 angle +54.72◦
Fr m2
φ
agent 2
m1 -90 -60 -30 0 30 60 90
agent 1 Control angle ψL (deg)
Fig. 2: Inter-agent magnetic force vs. control angle based on eq. (1), for the
Fig. 1: Pairwise orientation parameters defined in global and local coordinates case of r = 5R (R denotes the radius of agents), |M| = 105 A/m. The critical
for agents in close proximity with magnetic moments m 1 and m 2 aligned with angles here include: ψL = 54.72◦ corresponding to the zero-radial force angle,
the actuation field B a . r 12 represents the pairwise distance vector connecting ψL = 0◦ and ψL = ±90◦ at which the radial force becomes maximum attractive
agent 2 to agent 1. F r and F t are radial and transverse forces exerted on the (negative) and repulsive (positive) value, respectively.
first agent by the second agent (corresponding forces acting on agent 2 are
not shown).

horizontal plane. As the applied field is uniform over space,

robot agents operating in close proximity. In our design, the no external magnetic forces are generated (∇B Ba = 0). Fig. 1
spatially-uniform external field creates only torque to orient shows two agents with the relevant kinematic parameterization
micro-agents and as a result the inter-agent force appears and interaction forces.
between agents. We modulate the agents’ magnetic moment B. Inter-agent magnetic force
angle to regulate this force. We choose to make the agents
We study the forces acting on agent 1 caused by field
interact at a water-air interface to allow for two-dimensional
gradients created by agent 2. The position of agent 1 relative to
study.
agent 2 is denoted by r 12 , as depicted in Fig. 1. The magnitude
The paper is structured as follows. Section II describes the
of this vector r12 is termed “agent separation” while the angle
kinematics of agents along with the inter-agent force relation.
φ made by this vector with the horizontal x-axis is termed
Section III introduces our control principles to regulate the
“pair heading”. We model each agent as a magnetic point
relative motion of agents; accordingly, a systematic feedback
dipole and so only consider the magnetic moment in force
control law is synthesized to handle best performance. In Sec-
calculations. Let F 12 represent the force exerted on the first
tion IV, a simulation based on a physical model is conducted to
agent by the second agent as F 12 = ∇(m m1 · B 12 ), where B 12
enable prediction of agents’ behaviors; afterwards, fabrication
is the magnetic flux density at the position of the first agent
method and experimental setup that is used to control the
with magnetic moment m 1 created by the second agent with
relative position of agents are presented, then experimental
magnetic moment m 2 . In local cylindrical coordinates (êr , êt ),
results are carried out. This paper is concluded in Section V.
the radial and transverse components of this local magnetic
II. C ONCEPTS AND D EFINITIONS force can be written as
This section introduces the kinematics describing a pair of −3µ0 m1 m2
Fr = [2cos2 (ψL ) − sin2 (ψL )] and (1a)
agents along with the inter-agent force relation, and lays the 4πr4
foundation for controlling a two-agent configuration. −3µ0 m1 m2
Ft = [sin(2ψL )]. (1b)
4πr4
A. Magnetic actuation and inter-agent kinematics Here, µ0 is the permeability of free space, ψL the local
We base the following analysis on the assumption that the control input angle defined as the angle between the actuation
magnetic moment m of all magnetic agents in the workspace field B a and the vector r 12 as sketched in Fig. 1. Similarly,
align with the applied field B a . We use the angle of the ψG = ψL + φ is the control angle in global coordinates. These
applied field as our control input to the entire system. Consider force components as a function of control angle are sketched in
two identical magnetic agents with magnetic moments m Fig. 2. From this figure, we can see how to develop a controller
aligned with the uniform external magnetic field B a applied for the modulation of the inter-agent magnetic force as a
in the workspace for actuation in a two-dimensional (2D) function of the control angle ψL . At ψL = 0◦ and ψL = ± 90◦
the radial force attains its maximum attractive (negative) and
repulsive (positive) value, respectively. It is to be noted that
the magnitude of the attractive force is roughly twice as large
as that of the repulsive force. Importantly, as shown in Fig. 2, ψL
at ψL = 54.72◦ the radial force becomes zero. Also, at any ψL
angle between 0 and 90◦ , a non-zero transverse force occurs ψL
which causes the pair of agents to rotate about one another.

Bmin (mT)
ψL
The transverse force peaks at 45◦ . Additionally, it can be seen
from the figure that the radial and transverse forces are even
and odd functions of the control angle about 0◦ , respectively.
In other words, by reflecting the control angle about ψL = 0◦ ,
the transverse force can be reversed without affecting the radial
force. This reflection will be used to control the pair heading.
C. Input field strength requirement
The total field at the location of an agent is the vector
sum of the inter-agent field B 12 (field on agent 1 created by
Agent separation r (agent radii)
agent 2) and the actuation field B a . It is convenient to assume
that the global field B a dominates the local field such that all Fig. 3: Minimum required input field strength as a function of agent separation
r for multiple control angles to limit the total angle error to θe = 5◦ (|M| =
agents always align with the actuation field. Here we check 105 A/m).
our assumption that the local field created by a nearby agent
does not rotate the total field at an agent’s location. To avoid
this phenomenon, the actuation field strength can be chosen holds. The agents stay suspended at the interface by capillary
to keep the total field angle error less than a threshold given forces. Depending on the materials used for the agents and
by θe = ψa − ψL , where ψa represents the angle of total field. the liquid container, horizontal capillary forces between agents
For a given angle error threshold θe , the minimum required and between the container and the agents may exist [9]. Our
field strength Bmin is estimation shows these forces have a magnitude 100 times less

−µ0 m 2 tan(θe + ψL ) + tan(ψL )
 than the magnetic force; in other words, the observation meets
Bmin = . (2) our assumption that the capillary force impact is negligible
4πr3 tan(ψL ) − tan(θe + ψL )
against the magnetic force at the operating separation range
Fig. 3 shows the minimum required field strength as a function for a water-air interface.
of agent separation r for multiple control angle inputs, using
a maximum angle error of θe = 5◦ . It is noted that ψL = 0 and III. C ONTROL OF T WO -AGENT C ONFIGURATION
ψL = 90◦ see no angle error as the local and global fields are This section presents our approach to regulate the agent
parallel in these cases. To ensure that the agents align to the separation and pair heading angle. The task in designing a
applied field, the minimum field strength is applied. For the controller is to choose the input magnetic field angle ψL to
experimental section of this paper, agent spacing is roughly push the relative spacing and angle of the pair towards the
5R, so we maintain a field strength of 2 mT to be higher than goal state. The basis for producing these radial and transverse
the minimum required and assume that the agents always align forces is shown in Fig. 4(a). We start with the simplest
with B a . radial controller with two input states, then generalize to a
D. Other forces proportional radial controller. Finally we introduce a transverse
angle controller.
1) Fluid drag force: The microrobots used in this study are
500 µm in radius and 80 µm in thickness, and are experiencing A. Radial (separation) controller
low Reynolds number fluid flow. Therefore, a first-order model 1) Bang-Bang control principle: The simplest radial con-
is considered to describe the agents’ motion based on the troller has two states as illustrated in Fig. 4(a) (i-ii). These two
Stokes fluid drag model as states lead to repulsion or attraction of the two agents, when
Fm their moments are perpendicular or parallel to their separation
ẋx = − , (3)
b vector, respectively.
where ẋx is the induced agent velocity and F m is the inter- (i) If the space between two agents is too small (r < rdes ),
agent magnetic force. The term b = 6π µR, fluid drag constant, the controller points the field orientation perpendicular
depends on liquid viscosity, µ, and the radius of the assumed to their separation vector r so that the agents repel each
spherical agents, R. other with full radial force. This is done by applying
2) Capillary force: In this work, we restrict the agents to ψL = 90◦ . The transverse force at ψL = 90◦ is zero.
operate at a water-air interface to constrain their motion to two (ii) If the space between two agents is too large (r > rdes ),
dimensions, and assume that the Stokes fluid drag model still the controller points the field orientation parallel to their
(i). Repulsion (ii). Attraction
ψL = 90◦ ψL = 0◦
Ba (a) (b)
Ba rdes agent 2 agent 2

rdes

r < rdes r > rdes

agent 1 agent 1

Radial P-Control Radial and Transverse P-Control

Ba

e = r − rdes (R)
e = r − rdes (R)

ψL (deg)
ψL (deg)
(iii). Zero radial-force r ≈ rdes
ψL = 54.72◦

(a)
Time (s) Time (s)
ψL (deg)
Fig. 5: Two-agent configuration control simulation. (a) Radial P-control. (b)
Radial and transverse P-control (R denotes the radius of agents equal to
90 500 µm). The top plots show the motion trajectory simulation associated
to each controller in solid line. The desired pairwise separation rdes is
Repulsion reached when the sketched surrounding dashed-line circles around agents with
Error radius equal to rdes /2 come into contact with one another. Initial positions
54.72 e (mm)
are denoted by circle and current position with diamond. Here the initial
Setpoint Gain K separation is r = 10R assuming the desired separation to be rdes = 5R, and
|M| = 105 A/m. The bottom graphs illustrate the separation error and control
Attraction angle input as a function of time for each controller.
0

(b) B. Angle (pair heading) control

For control angle inputs between ψL = 0◦ or 90◦ , a trans-
Fig. 4: Two-agent configuration control principle. (a) States of agent attraction
or repulsion are determined by the direction of applied field B a : i) repulsion
verse force Ft will be induced on the agents according to
at ψL = 90◦ , ii) attraction at ψL = 0◦ , iii) zero radial-force at ψL = 54.72◦ . eq. (1b). If we wish to control the pair heading, we can
(b) Radial proportional controller design. choose the sign of this transverse force by choosing whether
we operate with positive or negative values of ψL as shown
in Fig. 2. Because Fr is symmetric about ψL = 0◦ , reversing
separation vector r so that the agents attract each other the sign of ψL does not affect the radial controller already
with full radial force. This is done by applying ψL = 0◦ . introduced. Although we cannot arbitrarily choose desired
The transverse force at ψL = 0◦ is zero. values of both Fr and Ft , we can regulate the radial force
(iii) (optional) If the space between two agents is close to the proportionally while introducing a bang-bang type binary
goal spacing (r ≈ rdes ), the control angle is set to ψL = controller on the transverse force. We choose to keep the
54.72◦ such that the radial force is zero. In this state, proportional controller on the radial rather than the transverse
a transverse force is created which rotates the pair (we force because this coordinate tends to be unstable (due to
ignore this rotation for now). the r−4 relationship on Fr in eq. (1a)). Using this scheme,
the control of the two parameters r and φ can be regulated
2) Proportional control: The bang-bang controller suffers using a single magnetic global input. In the next section, the
from a low level of precision due to its binary input states. effectiveness of this control method is characterized.
A more sophisticated controller would choose intermediate
angles between those shown in Fig. 4(b), centered around the IV. R ESULTS
zero-radial force angle. To create a proportional controller (P- This section presents numerical simulation results, the fab-
controller), we define the radial separation error e = r − rdes , rication process and experimental setups, and the experimental
which allows us to give the proportional control law as results of the proposed controller for two-agent configuration.
ψL = ψs − Kkek. Thus, the controller applies a larger radial
force correction for larger separation errors e (up to saturation)
as shown in Fig. 4(b). The gain K is tuned manually to result A. Simulation
in a stable controller with good performance. It can be noted Fig. 5 shows numerical simulation of the motion trajectory
that the Bang-Bang control law is a special case of the radial of magnetic agents in two-agent configuration using the P-
P-controller with infinite gain K. controller in two modes: (a) radial P-control, (b) radial and
 2.5
camera

r (mm)
z
rdes
250 µm
orientation circle x y 0
polymer shell 90
magnetic coil system

ψL (deg)
54.72

0
0 5 10 15 0 5 10 15
Time (s) Time (s)
magnetic agents (a) (b)
1 cm Fig. 7: Experimental results comparing (a) Bang-Bang, and (b) radial (no
transverse) P-controller. Experiments are done using magnetic field strength
of |B
Ba | = 2 mT and gain K = 2 (deg/mm) to reach the desired separation
5 cm of rdes = 2.5 mm. Radial P-controller with separation tracking RMS error
of 130 µm has a more stable performance compared with Bang-Bang with
separation tracking error of 197 µm. Also in P-controller the input angle
Fig. 6: Experimental setup. In the inset image of agent, the orientation circle ψL varies about the zero-radial force angle instead of aggressively jumping
atop mimics the north side of each agent’s magnetic moment. The identical between 0◦ and 90◦ in Bang-Bang (multimedia is available at [17]).
disk-shape agents shown have a radius of 500 µm and thickness of 80 µm
that float on water surface. The agents are driven in horizontal plane by an
electromagnetic coil system with two pairs of coils capable of producing fields
in the x − y plane.
field is smaller than the coercivity of the magnetic materials
in the agent, and so the agents’ magnetization will not be
altered by the actuation field. Agent position is detected using
transverse P-control. This simulation is a time integration a camera (FO134TC, FOculus) mounted atop the workspace,
of eq. (1) for initial condition of r = 10R and parameters and a computer with custom C++ code finds agent positions
|M| = 105 A/m, µ = 8.9 × 10−4 Pa s. Although the agents have using a Hough Circle Transform in the openCV library at 60
a disk shape, for simplification we model them as spheres frames/second [16]. Two identical agents are placed in a plastic
to calculate the fluid drag force eq. (3). The inter-agent and petri dish on water as illustrated in Fig. 6.
agent-air-wall capillary forces, as well as inertial forces are
C. Experiment
ignored in the simulation. The simulation shows a particular
case where two agents are initially far away (r = 10R). It can We tested two types of controllers:
be seen that both controllers approach the goal configuration 1) Radial (no transverse) control: Fig. 7 shows the experi-
and the error reduces to a small value over time. We have seen mental results for (a) Bang-Bang, and (b) radial (no transverse)
in simulation that the controllers are stable for a wide variety P-control. While both controllers demonstrate asymptotic sta-
of initial conditions. bility during this experiment, the P-controller achieves a lower
RMS error from the goal of 130 µm as compared with
B. Fabrication of agents and experimental setup 197 µm for the Bang-Bang controller. The control angle signal
Magnetic agents are fabricated as in Zhang and Diller [16]. in Fig. 7(b) demonstrates that the attraction force tends to be
We describe briefly the fabrication process involving: (a) greater in magnitude than the repulsion force in pulling the
making mold using photolithography, (b) pouring polymer and agents toward each other, as the inter-agent attractive force is
curing toward desired shape. Identical agents are composed stronger than the inter-agent repulsive force (as discussed in
of magnetic microparticles mixed with polyurethane polymer Section II-B). As aforementioned, in Bang-Bang the control
(BJB M-3184), which is poured into a negative mold to define input can take only two states of ψL = 0◦ and 90◦ . However,
the agent shape, degassed in a vacuum pump and scraped during the experiment (multimedia is available at [17]), agents’
level using a razor blade. The mold is made using SU-8 moments physically sweep through the continuous range of
photolithography and a replica molding process. Magnetic local angles between 0◦ and 90◦ , although this leads to a neg-
fields for agent actuation are created in an electromagnetic coil ligible impact on stability. In addition, a slight displacement of
system with two pairs of coils nested orthogonally to create the pair of agents is observed due to other disturbance forces
fields in the plane, powered by two analog servo drives (30A8, potentially the one resulted from small non-uniformity of the
Advanced Motion Controls). Each pair of wire loops in the coil actuation field. It is to be noted that one can not use Bang-
system is arranged in Helmholtz configuration, resulting in a Bang to control the transverse motion, because at the possible
uniform magnetic field up to 15 mT (uniform to within 5% of input angles the transverse force will be always zero.
nominal at the center over a workspace size of 5 cm) located at 2) Radial and transverse control together: Fig. 8 shows the
center of the coil system (see Fig. 6). The strength of magnetic experimental results for separation and pair heading control
Separation (mm)
(a) desired r using larger applied fields and faster controllers. Fluid and
capillary interactions were assumed to be negligible in this
study. However, it is recommended to tailor these forces for
actual r various liquid interfaces. In future work, the proposed scheme
will be extended to 3D and agent pair formation will be
examined using field gradient under liquid surface taking into
Pair heading (deg)

(b) desired φ
account gravity. Future research will investigate the problem
of manipulating multiple agents to complete useful tasks using
a team of agents in 3D fluidic environments.
actual φ
R EFERENCES
(c) [1] B. J. Nelson, I. K. Kaliakatsos, and J. J. Abbott, “Microrobots for min-
Control input

imally invasive medicine,” Annual Review of Biomedical Engineering,

+54.72◦
ψL (deg)

vol. 12, pp. 55–85, 2010.

zero radial-force angle [2] E. B. Steager, M. S. Sakar, C. Magee, M. Kennedy, A. Cowley, and
−54.72◦ V. Kumar, “Automated biomanipulation of single cells using magnetic
microrobots,” The International Journal of Robotics Research, vol. 32,
no. 3, pp. 346–359, 2013.
0 [3] G. M. Whitesides and B. Grzybowski, “Self-assembly at all scales,”
Time (s) Science, vol. 295, no. 5564, pp. 2418–2421, 2002.
[4] E. Diller and M. Sitti, “Three-dimensional programmable assembly
Fig. 8: (a) Separation r and (b) pair heading φ tracking using radial and by untethered magnetic robotic micro-grippers,” Advanced Functional
transverse P-controller together along with (c) the applied control angle input Materials, vol. 24, no. 28, pp. 4397–4404, 2014.
ψL . The control angle flips to maintain the desired pair heading angle. RMS [5] W. Hu, Q. Fan, and A. T. Ohta, “Interactive actuation of multiple
tracking error of less than 73 µm and 14◦ is accomplished for the regulation of opto-thermocapillary flow-addressed bubble microrobots,” Robotics and
the separation and the pair heading angle, respectively (multimedia is available Biomimetics, vol. 1, no. 1, pp. 1–6, 2014.
at [17]). [6] J. J. Abbott, M. C. Lagomarsino, L. Zhang, L. Dong, and B. J. Nelson,
“How should microrobots swim?” The International Journal of Robotics
Research, vol. 28, no. 11-12, pp. 1434–1447, 2009.
[7] S. Chowdhury, W. Jing, and D. J. Cappelleri, “Controlling multiple
to track a changing goal state. RMS tracking error of less microrobots: recent progress and future challenges,” Journal of Micro-
than 73 µm and 14◦ is accomplished for the regulation of the Bio Robotics, vol. 10, no. 1-4, pp. 1–11, 2015.
separation and the pair heading angle, respectively. Although [8] E. Diller, J. Giltinan, and M. Sitti, “Independent control of multiple
magnetic microrobots in three dimensions,” The International Journal
tracking is achieved, there is a small undesired coupling of Robotics Research, vol. 32, no. 5, pp. 614–631, 2013.
between the controllers. Ideally, to be able to independently [9] J. M. Ng, M. J. Fuerstman, B. A. Grzybowski, H. A. Stone, and G. M.
control r and φ , flipping the control angle for transverse Whitesides, “Self-assembly of gears at a fluid/air interface,” Journal of
the American Chemical Society, vol. 125, no. 26, pp. 7948–7958, 2003.
control needs to occur instantaneously. One can confirm by [10] B. A. Grzybowski, H. A. Stone, and G. M. Whitesides, “Dynamic self-
tracking the agent orientation in the video [17], that the assembly of magnetized, millimetre-sized objects rotating at a liquid–air
agents remain aligned with the actuation field during the entire interface,” Nature, vol. 405, no. 6790, pp. 1033–1036, 2000.
[11] A. Snezhko and I. S. Aranson, “Magnetic manipulation of self-
experiment. However, the agents take approximately 100 ms assembled colloidal asters,” Nature Materials, vol. 10, no. 9, pp. 698–
to align with the field when it switches direction suddenly. 703, 2011.
Also, as mentioned earlier, for a rotating actuation field with [12] S. Martel, M. Mohammadi, O. Felfoul, Z. Lu, and P. Pouponneau, “Flag-
ellated magnetotactic bacteria as controlled mri-trackable propulsion
a constant strength, if the separation between agents becomes and steering systems for medical nanorobots operating in the human
too small, not only angle error interference may come to play, microvasculature,” The International Journal of Robotics Research,
but also the appeared local force is hard to overcome by a vol. 28, no. 4, pp. 571–582, 2009.
[13] A. Becker, Y. Ou, P. Kim, M. J. Kim, and A. Julius, “Feedback control
general controller. of many magnetized: Tetrahymena pyriformis cells by exploiting phase
inhomogeneity,” in IEEE International Conference on Intelligent Robots
V. C ONCLUSIONS and Systems, 2013, pp. 3317–3323.
[14] E. Diller, C. Pawashe, S. Floyd, and M. Sitti, “Assembly and disassembly
In this paper, for the first time, a method was introduced of magnetic mobile micro-robots towards deterministic 2-D reconfig-
to control the motion of small-scale identical magnetic robots urable micro-systems,” The International Journal of Robotics Research,
vol. 30, no. 14, pp. 1667–1680, 2011.
in close proximity with each other. This paper established a [15] S. Miyashita, E. Diller, and M. Sitti, “Two-dimensional magnetic micro-
tool for multi-agent control of magnetic microrobots using module reconfigurations based on inter-modular interactions,” The In-
a global magnetic field. We designed a set of controllers to ternational Journal of Robotics Research, vol. 32, no. 5, pp. 591–613,
2013.
maintain any inter-agent separation and pair heading in two [16] J. Zhang and E. Diller, “Millimeter-scale magnetic swimmers using elas-
dimensions (2D) via modulation of the inter-agent magnetic tomeric undulations,” in IEEE International Conference on Intelligent
force. Simulation and experimental results showed precise Robots and Systems, 2015, pp. 1706–1711.
[17] Experiment multimedia available online: http://microrobotics.mie.
tracking with two magnetic agents along various combina- utoronto.ca/?p=474.
tions of the separation and pair heading trajectories. For the
controller to operate without instability, the relative spacing
between agents in this work needed to be between 3 and
30 agent’s radius. Future work will widen this range by

View publication stats