Multi-Agent Systems Using Diamagnetic

Micro Manipulation - From Floating Swarms to

Mobile Sensors
Ron Pelrine, Allen Hsu, Cregg Cowan, Annjoe Wong-Foy
Advanced Technology and Systems Division
SRI International
Menlo Park, CA United States
[email protected]

Abstract—Multi-agent robotic systems on small scales Figure 1 shows a typical one-robot system, also referred to as a
typically use, or envision using, small mobile robots. This paper DiaMagnetic MicroManipulation (DM3) system. Robotic
takes a broad-brush look at a relatively recent development in agents in DM3 systems may be diamagnetically enhanced to
milli and micro robots, Diamagnetic Micro Manipulation (DM3) reduce sliding friction, or they may be fully levitated, or,
systems, in a multi-agent context. We report various multi-agent because the robots are mobile, they may be both within the
operations, such as multi-agent part manipulation, robots-acting- same system. Sliding DM3 systems [6] share many of the traits
on-other-robots for enhanced functionality, and robot-supplied of their non-diamagnetic, PCB-driven counterparts [7, 8, 9],
local feed, to assess the state of the art of this new type of system. and their fundamental mechanical behavior in space is
In addition, we explore and analyze the multi-agent implications
comparable to traditional machines. In this report, we describe
of known DM3 properties and earlier single-agent results to
several new multi-agent operations with sliding DM3 systems
motivate future research in this area. As reported here, some of
these multi-agent implications include mobile high-precision
to illustrate their versatility in this domain, and how a multi-
sensor transducers, near-field communications channels using agent perspective can help implement robotic tasks in this
robots as both transmitters and receivers, and novel swarm architecture. Sliding DM3 robots can be made fairly small,
architectures. down to roughly 1 mm, and large numbers of agents should be
feasible in compact systems, as earlier reported and
Keywords—automation, diamagnetic, multi-agent, levitation, demonstrated with 1024 agent g systems
y [[10].
]
micro-assembly, sensing

I. INTRODUCTION
Multi-agent systems of multiple or many robots have been
under active investigation for many years [1, 2]. For example,
multiple fixed robots are often used in manufacturing
workcells. These robots can cooperate on various tasks to
achieve significant benefits as compared to single robots.
Fig. 1. Single robotic agent of a DM3 system. At left, photo of levitated
If the robots have mobility, (i.e., they can move many body robot. At right, typical agent components showing levitated bot driven by
lengths), then the multi-agent system can achieve still more a 4-layer PCB. Currents through traces drive the robot. The diamagnetic
capabilities, and operations are no longer limited to one graphite layer provides the passive levitation force.
location. Much of the recent interest in multi-agent systems is
Fully levitated DM3 robots, however, behave in
focused on mobile robots, such as quadcopter swarms, terrain
fundamentally different mechanical ways than conventional
mapping, and warehouse systems [3, 4, 5].
machines, independent of the distinction between macro and
The behavior and capabilities of multi-agent systems micro. At one level, levitation enables relatively
depends both on the purely system-level control, but also on straightforward consequences such as wearless reliability and
the specific properties and capabilities of the agents. For small size robots (< 1 mm). At a deeper physics level,
robotic systems, one important aspect is robot size. Robot size however, diamagnetically levitated milli- and micro-robots can
determines much of the physics of their operating environment, be controlled and moved with near-zero noise in their force,
as well as the possibility of having large numbers of agents in a and they can be made intrinsically stable. Near-zero noise in
practical system. Generally speaking, small robot sizes enable the force can be used for highly optimized control as
larger numbers of agents, but the physics of their operating previously reported [11], but one of the more striking
environment on small scales is perhaps less well understood consequences of diamagnetic levitation physics on robots is
than their macro counterparts. This paper examines a particular that, unlike the vast majority of large or small robots, they can
class of systemsrobotic systems based on printed circuit also be used for high-precision transduction for sensing [12].
board (PCB) drivers using diamagnetic materials as Thus, a multi-agent system of diamagnetically levitated robots
bearingsand specifically examines multi-agent aspects. is also a multi-agent system of mobile sensor transducers.
 Other aspects of DM3 systems that relate to multi-agent the drive system and diamagnetic bearings (see below) makes
system characteristics include the simplicity of the manipulator the DM3 system very repeatable in both position and trajectory
(enabling self-assembly and self-repair at the system level), the as a robotic system, with open loop position repeatability
use of zone or area control rather than individual robot control, currently measured on the order of 100 nm rms and trajectory
and robot-to-robot magnetic interactions (enabling robot-to- repeatability measured on the order of 1-μm rms [12].
robot communications but also disturbances and limits on robot
Two other aspects of the drive system deserve particular
packing at the system level).
note with respect to multi-agent systems. One aspect is that
These and other features of DM3 as they relate to multi- robots and other devices in DM3 systems are controlled by the
agent systems will be analyzed and discussed in subsequent local trace pattern and currents. That is, the robot’s control is
sections of this article. We emphasize, however, that multi- area- or zone-based rather than one that moves with the robot
agent capabilities of complex DM3 systems is just beginning to (as would be the case for conventional motorized robots). =RQH
be investigated and understood. This report, therefore, takes a control is not in itself new [13, 14], and has both advantages
broad-brush picture of what is known and/or has been and disadvantages for multi-agent control. The disadvantage of
demonstrated in a multi-agent context, rather than an in-depth zone control is that two robots in close proximity may not be
focus on a specific DM3 system and application. independently controlled unless they are in different
independent zones. The advantage of zone control is that large
II. EXPERIMENTAL SETUP AND BACKGROUND PHYSICS numbers of robots may be controlled to execute the same
motion in parallel using only a few control channels. The
A. Principle of Operation control zone approach generally reduces the numbers of
control channels needed since the robots do not need to carry
Referring to Figure 1, DM3 milli- and micro-robots use an extra control channels in areas which need, for example, only
array of NdFeB (1.4 x 1.4 mm) magnets arranged with one degree-of-freedom for transport.
alternating magnetization (checkerboard pattern). This
configuration, in conjunction with a diamagnetic layer of The other aspect to note for DM3 is that PCB, flex, and
pyrolytic graphite (5 ȝP, provides stable levitation of the other lithographically patterned micro-circuits enable large and
micro-robot. However, in order to confine the robots to a complex drive systems to be made relatively easily using
specific location in (x, y, z), an underlying PCB, it is necessary conventional batch fabrication. Current DM3 systems are
to generate a magnetic potential well to localize the robot. The routinely made as large as 30 cm x 30 cm, and larger systems
magnetic force is generated by four PCB current traces (IX1, are undoubtedly possible. We also note that DM3 robots can
IX2, IY1, IY2). The pairs of traces in X and Y are typically driven transition between totally separate PCBs if they are in
in quadrature, behaving very similarly to a linear stepper proximity of one another. These capabilities are important for
motor. While driving the currents in quadrature controls the future versions of multi-agent DM3 systems in which large
relative phase between the pairs of currents and therefore the numbers and/or large sizes are desired.
robot in-plane position, modulating the absolute magnitude of
the traces increases or decreases the out-of-plane force between III. MULTI-AGENT OPERATIONS
the board and the robot providing ~40-ȝPRI=PRWLRQ$V
shown in Figure 2, end effectors can also be attached to the top DM3 systems can employ many different milli-robots and
of the robot for accomplishing specific tasks. end effectors to execute various tasks. It is relatively
straightforward to imagine that using thousands of robots in
Graphite thickness on the order of 0.5 mm is used for parallel could achieve higher throughput with a given process
levitated systems, while thinner graphite sheets, on the order of than a single or a few robots; however, systems of thousands of
25- to 100-Pm thick, are used for diamagnetically enhanced robots with thousands of different tools and processes that can
sliding systems. We refer the reader to the following papers for be executed are harder to understand and characterize.
more details about the DM3 milli-robots [6, 11, 12].
A. Cooperative Robots
B. DM3 Drive System Robots may cooperate by doing different steps in a building
The DM3 drive system described in connection to Figure 2 process for a single object, such as having one robot deposit
has its own physics and fabrication characteristics. From a glue while another robot brings a part to place for gluing [6,
magnetic drive standpoint, it is similar to voice coil actuation, 10]. In this simple form of cooperation, DM3 systems are
using only permanent magnets actuated by currents in similar to other multi-robot systems.
conductors (Lorentz force). The absence of soft ferromagnetic
materials, such as iron-cores, enables ultra-low hysteresis, More interesting examples of cooperation involve robots
which is why high-fidelity acoustic speakers are made using working together more directly. The simplest multi-agent
voice coil designs. system uses two robots. The bots may operate on each other to
achieve greater functionality, or they may operate jointly on a
With respect to DM3 systems, the consequence of a Lorentz single part for manipulation. Figure 2 shows two examples of
force design is that the force and torque exerted on the rare Earth the former, while Figure 3 shows two examples of the latter.
magnet array at a given location and orientation is extremely
linear and repeatable in the trace currents. The repeatability of In the top photos of Figure 2, two independent robots used
to make a controllable syringe illustrate how one robot can
The authors gratefully acknowledge funding support for this work under actuate a device carried by another robot to achieve greater
DARPA Contract # HR0111-12-C-0040.
functionality. In the case of the syringe, one robot controls the those in Figure 4 can implement local feed, by using the bots
plunger for a tube robot that carries the syringe tube. Related themselves to bring large quantities of building materials to the
devices can have one robot mechanically actuate a gripper or build site. This is in contrast to central feed for the same carbon
other device. The common theme is that it may be problematic rods described in previous reports [6]. Local feed may have
to carry power and implement control on one milli-robot for significant advantages in throughput speed depending on the
some actuated functions, but the problem may be much easier specific operations and parts.
to solve within the DM3 system if one just uses another robot.
Transferred
However, note that two or more robots in proximity must be in Carbon Rod
Carbon Rod
independent DM3-control zones to be controlled independently; 13 mm

therefore, it may not be possible without zone planning to actuate 13 mm

multiple robots in this fashion in the desired workspace. This is Milli-bot

more clearly seen in the bottom of Figure 3, where a robot carrying carbon
Milli-bot originally
carrying carbon rod
carries a ramp and another smaller robot drives onto it to rod
change tilt angle. This example uses an uncoated PCB so the
zone boundaries are easily identified. 7 mm 7 mm

Tube-carrying bot Plunger bot

Wetting Variable Angle

Pads Carbon Rod Carbon Rod

10 mm
Figure 2. Frame grabs from video showing DM3 robots acting jointly on an
object. Top: A carbon rod is transferred from one bot on an angled flex to
another bot. Bottom: A carbon rod is jointly held by two bots using surface
tension. Differentially driving the bots changes the rod angle for building.

Cartridge bot

Zone Boundary
10 mm
Pick bot

Figure 1. Frame grabs of video showing DM3 robots acting on each other.
Top: A mobile, programmable syringe is made using a bot to carry the syringe
tube, and a second bot plunger carries and actuates the plunger. Bottom: A
larger bot carries a graphite ramp, and a second smaller bot drives onto it to
change the tilt angle. The PCB is uncoated in this test to show the independent
zones controlling each bot.
Multiple robots may also actuate or manipulate an object
together. Figure 3 shows two examples not previously
reported. Figure 3 (bottom) shows cooperative robots changing 10 mm
the orientation of a carbon rod while holding it jointly, while
Figure 3 (top) shows one robot on a flex circuit handing off a
Figure 3. A pick robot using wetted sponges approaches a bot carrying a
carbon rod to another robot (part transfer) so as to change the cartridge with 10-20 carbon rods for local feed. The pick bot can extract a
part’s angle. In both cases, the part is held using water wetting single carbon rod for building.
[6, 10]. Part transfer processes where one bot pushes or pulls a
part off another bot can be surprisingly effective, and have B. Mobile Sensor Transducers
demonstrated 100+ transfers without error in testing. The agent shown in Figure 2 uses trace currents in the PCB
to control three or more degrees of freedom (DOF) with a tool
Cooperative robots can also implement local feed. Figure 4
or end effector. It is therefore, by most common mechanical
shows a previously unreported example of cooperative robots
definitions, a robot. The classification as a robot does not,
in which one robot locally supplies materials to another robot
however, preclude it from being other things. For example, it is
for building. In the example in Figure 4, one robot brings in a
well-known that many macro direct drive robots (no gearing)
cartridge of 10-20 carbon rods, which are then extracted from
can crudely weigh the objects they lift up by measuring the
the cartridge singly by a second bot. Cooperative robots like
motor currents needed to hold the object stationary against
gravity. In this mode, the robot is acting as a sensor transducer, highest sensitivity readouts usually need the highest precision
converting force from gravity into another physical quantity, in positioning the transducer. Here the high robotic precision
namely motor currents. of DM3 is a major advantage, and one can even imagine DM3
systems in which both the transducer and the readout are
Measurements of weight using macro robots are typically
mobile precision agents (< 1 Pm positioning).
very crude (~ 1 part in 10-20) because the robot’s bearings
have friction and other unpredictable forces that make the
macro robot a noisy transducer. These unpredictable forces can
be referred to as bearing noise. Mechanical-sensor bearings,
such as flexures and other spring elements, are designed to be
highly repeatable and predictable so they have low bearing
noise in their force transduction, in contrast to actuator and
motor bearings, which typically emphasize load-bearing and
motion range rather than force repeatability properties.
Sensor-grade flexure bearings are highly repeatable
because they are nearly perfectly elastic, with low or zero
hysteresis and zero friction. Diamagnetic bearings have these
same properties, and indeed, much of the early work in
diamagnetic levitation focused on mechanical sensor
applications [15, 16, 17]. It should be noted that the effect is
not small – measurements of diamagnetically levitated robots Fig. 5. Position readout sensors, such as photointerrupters, can be fixed in the
showed roughly a 100-1000 fold decrease in bearing noise workspace. More advanced designs might make the readout sensor a mobile
agent that is carried on a robot with local electrical contacts.
compared to friction bearings using the same system with
sliding contact [12], as indicated in Table 1. While there is limited data on mobile DM3 robotic systems
Roughly speaking, for a robot, low bearing noise means that have the multi-functional capability of precision mobile
that with a suitable drive the force can theoretically be transducers, a few examples exist. One reported example [12]
controlled ~1000 times better than with a comparable friction weighs the mass of a grain of salt on the fly, i.e., while the
bearingenabling superfast optimized motion as previously robot/transducer is in relatively high-speed motion (1.5 cm/s).
reported [11] and exquisite force and stiffness control. For a The mass of the salt grain, about 140 Pg, was measured using
sensor, this means the device can be 1000 times more sensitive an optical readout (analog photointerrupter) to detect changes
in converting force to displacement. The two modes, actuator in the trajectory position as the robot passed through the optical
and sensor, can be combined, such as for a tunable sensor beam. Mass resolution of this method was about 10 Pg, or
whose resonance frequency for determining sensitivity is about 1 part in 1300 of the robot’s 13 mg mass.
controlled by the robot driver. We also note the robot driver More recently, we have used milli-robots with optical
can implement a commonly used sensor design, sometimes readouts for detecting the position of a surface. Using a system
referred to as a forcer or torquer design, which uses active similar to that shown in Figure 6, a robot can be oscillated by
feedback to hold the transducer in a fixed position, thereby the robot’s drive, with the center of the oscillation controlled to
increasing sensor range and reducing non-linearities. gradually move toward the surface open loop (incrementing
Unlike flexures and springs, diamagnetic levitation can be TABLE 1. QUALITATIVE COMPARISON OF THEORETICAL BEARING NOISE
designed as mobile bearings, suitable for arbitrarily large VERSUS MOBILITY FOR SOME MAJOR BEARING CLASSES.
displacements and rotations. In this respect, it is almost unique Intrinsic
in bearing technology because it can be both a mobile bearing Bearing Type Mobility? Comments
Bearing Noise
and a precision sensor-grade bearing. Classic sensor bearing;
Flexures & only certain flexure
The impact of precision mobile sensor transducers for =HUR No
springs materials have zero
converting force to displacement in DM3 systems is little hysteresis
explored and poorly understood. This is due to in part to the Comparable to flexures
in force noise; typically
small size domain and multi-agent context. Large number, Diamagnetic
=HUR Yes able to reach lower
fixed mechanical sensor arrays on small scales are well known bearings
frequencies compared to
using microelectromechanical systems (MEMS) technology, flexures of the same size
but mobility is typically lacking. High; friction,
adhesion, High and unpredictable
Contact bearings
Small mobile robots can, of course, carry separate sensors, (bushings, ball
particulates,
Yes
force noise; Almost
but mechanical sensors also usually need a readout, such as a local plastic never used for sensor
bearings, sliding)
deformation and bearings
position optical readout for force-to-position transduction. The hysteresis
readout typically needs power and electronics, meaning the Powered bearings Specifics of force noise
integration problem on very small robots becomes that much (pumped air, Medium; noise depend on particular
more difficult. liquids, active from power
Yes
system; some powered
feedback source, bearings may have noise
The readout can be external to the robot with both DM3 magnetic particulates low enough for useful
(see Figure 5) and other types of robots, but note that the levitation) sensor functionality
high resolution micro-steps using trace currents). When the bot robots to date (note: dimensions refer to magnetic array; a
hits the surface, the regularly oscillating AC signal from the magnetically inert end effector could be much larger).
optical readout changes drastically to a larger-amplitude,
At high precision, the spacing for undetectable changes in
chaotic signal from impact. Surface position was set as the first
robot equilibrium position may need to be much larger than
change in motion amplitude from impact was greater than 2X
when less precision is required. However, we note that at large
the normal (free space) oscillation amplitude. Note that the
separations the system is just as repeatable, suggesting that first-
optical sensor itself doesn’t measure position, and is in fact
order or higher compensation algorithms might be effective.
only AC coupled in this experiment. Although data should be
regarded as preliminary given the limited testing, the system While static forces can be dealt with through calibration and/or
determined the open loop position of the surface to about a feedback, dynamic forces can lead to interesting effects
100-nm rms error as shown in Figure 6, or one could convert to between various robots. Figure 7 shows measurements between
the absolute position of the surface with a calibration of trace two adjacent 3 x 3 milli-robots, each in independent control
currents to position [18, 19]. zones and fully levitated over graphite. Because of the low-loss
94 nm rms deviation bearing between the robot and the graphite surface, the robot
12
11
in detected position behaves as almost an ideal harmonic oscillator, in which the
magnetic potential from the board provides a restoring force
Microns From Home Position

10
9 when the robot is perturbed. Using robots designed with a net
8
7
magnetic dipole, we can couple/transmit information from one
6 robot to another robot’s position. To excite the system, we
5 perform a step function motion of one of the robots (transmitter
4
3
robot) while we monitor the stationary (receiver robot) using
2 an optical sensor as a function of distance between them (d).
1 As shown in Figure 7 (b), as the two robots become closer and
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
closer, the receiver robot’s position begins to pick up the
Sample # motion/presence of the transmitter robot. However, this effect is
only seen when the transmitter robot performs a step function
Figure 6. Example data of 18 measurements of position of a hard glass surface
using a levitated robot with a photointerrupter showing 94 nm rms error.
or motion at resonance; relatively slowly varying motions of
Precision surface detection is one example using a levitated robot as a sensor. nearby robots, or motion without resonant frequency
components, will only perturb the remote robot’s equilibrium
Force-to-position transducers can be used to measure many position. This implies information can be selectively
physical quantities by converting them into a force. Mass, tilt, transmitted by controlling the dynamics of nearby robots.
acceleration, magnetic field, electrostatic field, and many other
quantities can be measured this way. Chemical measurements,
such as those using selective absorption, are also feasible using
mass measurements. Resonant frequencies using micro-robot
designs are typically in the 3-30 Hz range, suggesting high
sensor sensitivity (inversely proportional to resonance
frequency squared in many mechanical sensor designs).
The connection to mobile sensing is further extended by
using mobile readouts, and/or simply carrying other types of
transducers on the DM3 robot. In this latter case, the transducer
is not based on DM3, but rather relies on precise positioning in
the readout using DM3. A simple example would be a robot
carrying a rod with a known temperature coefficient of
expansion. Measuring the end of the rod, such as in the optical
readout of Figure 5, could determine temperature for the Figure 7. Near-field magnetic interactions between robots at resonant
system, either locally or globally. Thus, the readout can be frequencies suggest a natural robot-to-robot communications channel might
be implemented in multi-agent systems with little or no extra hardware
shared to measure many different physical quantities with
needed, a significant potential advantage for some swarm systems.
mobility, but this method hinges on the precision requirements
for transducer placement not overwhelming the system, as
would be the case for many non-DM3 methods. D. DM3 Swarms and Advanced Biomimetic Systems
Large numbers of agents are possible with DM3 systems,
C. Robot-Robot Magnetic Interactions such as the previously mentioned system of 1024 manipulators
[10]. These systems have been centrally controlled, so a natural
Magnetic interaction between robots is a constraint in DM3 research question to ask is whether DM3 systems can function
systems, but may also offer opportunities. As a constraint, if as swarms, defined here as multi-agent systems controlled in
the robots get too close they can adversely affect each other’s whole or in part by low-level, insect-like intelligence.
motion because of the magnetic forces. The magnitude of the
interaction depends on specific robot design and robot size. One path to DM3 swarms would use low-level intelligence
Robot center-to-center spacing as low as 2.6 mm has been integrated onto the mobile magnet arrays. This is a viable
demonstrated, but 10-15 mm is more typical of 4- to 8-mm technical path, and electrical connections to power on-board,
low-level intelligence electronics can be made by touching REFERENCES
contact pads. However, while on-board low-level intelligence [1] V. Scheinman, Robot world: “A multiple robot vision guided assembly
is a possible path to swarms, it faces the non-trivial challenge system,” Robotics Research, Fourth International Symposium, R. Bolles
of integrating electronics and other necessary components and B. Roth, Eds. Santa Cruz, California: MIT Press, 1987, pp. 23–27.
(sensors, power, etc.) onto small platforms with payload limits. [2] D. Wong, J. Wang, E. Steager, V. Kumar, “Control of multiple magnetic
micro robots,” Proc. ASME 2015 International Design Engineering
An alternate view is that the zones on the driving PCB Technical Conferences & Computers and Information in Engineering
really control the system, and the low-level intelligence might Conference IDETC/CIE 2015, August 2-5, 2015, Boston, MA, USA.
be integrated with the zone control rather than on the [3] Harvard Kilobot Project.
manipulators to achieve swarm-like behavior. Such systems http://www.eecs.harvard.edu/ssr/projects/progSA/kilobot.html
might have many of the characteristics of swarms, but be more [4] E. Diller, J. Giltinan, M. Sitti, “Independent control of multiple
magnetic microrobots in three dimensions,” International Journal of
akin to coral colonies or similar natural systems with Robotics Research 32:614, 2013.
synergistic collections of essentially immobile organisms. In
[5] A. Kortschack, O. Hanbler, C. Rass, S. Fatikow, “Driving principles of
this approach, the zones are the individual “organisms” while
mobile microrobots for micro- and nanohandling,” Intelligent Robots
the manipulators are similar to raw materials that are acted on and Systems, 2003. (IROS 2003) Proceedings, October 27-31, 2003.
and transferred between “organisms”. This approach also has [6] A. Hsu et al., “Application of micro-robots for building carbon fiber
support in view of the robot-to-robot communications via the trusses,” in 2016 Int’l Conf. on Manipulation, Automation and Robotics
magnetic field discussed above. The zones have sufficient at Small Scales (MARSS), Proceedings on IEEExplore ISBN no. 978-1-
magnetic control over the manipulators to transmit to nearby 5090-1510-8.
zones, whereas alternate approaches using on-board [7] S. Chowdhury, W. Jing, D. Cappelleri, “Towards independent control of
intelligence would need separate transmitters and receivers. multiple magnetic microrobots,” Micromachines 7:1, Article 3, 2016.
[8] D. Cappelleri, D. Efthymiou, A. Goswami, N. Vitoroulis, M. =DYODQRV
Lastly, DM3 multi-agent robotic systems combine an “Towards mobile microrobot swarms for additive micromanufacturing,”
unusual combination of manipulator simplicity and mechanical Int. J. Adv. Robotic Systems, Vol 11, Issue 9 September 19, 2014.
processing power of the system. The manipulators in DM3 can [9] C. Pawashe, S. Floyd, M. Sitti, “Multiple magnetic microrobot control
be as simple as a few magnets, or even single magnets, using electrostatic clamping,” Appl. Phys. Lett. 94, 164108, 2009.
whereas systems of even a few DM3 robots have significant [10] SRI International, “Magnetically actuated micro-robots for advanced
mechanical processing capabilities in terms of six DOF mobile manufacturing applications,” 2015.
https://www.youtube.com/watch?v=Bxb3-bT8uxk. Also, “Magnetically
motion at high-precision, multi-functionality as mobile sensors,
actuated micro-robots for advanced manipulation applications,” 2014.
and so forth. This system-agent feature has been previously https://www.youtube.com/watch?v=uL6e3co4Qqc
reported in the context of self-assembly and replication of [11] R. Pelrine, A. Hsu, A. Wong-Foy, B. McCoy, C. Cowan, “Optimal
manipulators [20]. The magnet arrays can be self-assembled by control of diamagnetically levitated milli robots using automated search
DM3 systems and as few as one robot is needed to build patterns,” in 2016 Int’l Conf. on Manipulation, Automation and
complex end effectors, using, for example, drop deposition Robotics at Small Scales (MARSS), Proceedings on IEEExplore ISBN
with UV-curable polymers. Estimates are that only ~ 5-10% of no. 978-1-5090-1510-8.
the system area for a 30-cm x 30-cm system is needed to [12] R. Pelrine et al., “Diamagnetically levitated robots: An approach to
massively parallel robotic systems with unusual motion properties,” in
implement in-system replication. To be sure, such processes 2012 IEEE Int’l. Conf. on Robotics and Automation (ICRA), pp. 739–744.
would need to be refined and integrated rather than tested as
[13] R.S. Fearing, “A planar milli-robot system on an air bearing,” 7th Int.
separate experiments, but the potential for in-system Symp. Robotics Research, G. Giralt and G. Hirzinger, Eds. New York:
replication, tool-on-demand, and self-repair is striking. Springer, 1996, pp. 570–581.
We also mention that advanced multi-agent biomimetic [14] B.R. Donald, C.G. Levey, C.D. McGray, I. Paprotny, D. Rus, “An
untethered, electrostatic, globally controllable MEMS micro-robot,”
systemsin which the magnet arrays are viewed as “stem Journal of Microelectromechanical Systems, 15:1, February 2006.
cells” that the system specializes with in-system fabricated end [15] V. Ponizovskii, “Diamagnetic suspension and its applications (survey),”
effectors according to system needsare a related interesting Instruments and Experimental Techniques, 24, pp. 833–841, 1982.
topic for future research in multi-agent DM3 systems as [16] Q. Li, K.-S. Kim, A. Rydberg, “Lateral force calibration of an atomic
previously suggested [20]. force microscope with a diamagnetic levitation spring system,” Review
of Scientific Instruments, 77, 065105, 2006.
[17] I. Simon, A.G. Emslie, P.F. Strong, R.K. McConnell, Jr., “Sensitive
IV. CONCLUSION/FUTURE WORK tiltmeter utilizing a diamagnetic suspension,” Review of Scientific
Multi-agent systems are at the heart of DM3 technology. Instruments, 39, pp. 1666–1671, 1968.
However, as reported here, these systems have a potential [18] A. Hsu et al., “Automated 2D micro-assembly using diamagnetically
richness and complexity of behavior that has been little levitated milli-robots,” 2017 Int’l. Conf. on Manipulation, Automation
and Robotics at Small Scales (MARSS), Montreal, July 17-21, 2017
explored or is only theoretical. Examples discussed here (unpublished).
include the multi-functionality of levitated robots to act as
[19] M. Ammi, V. Fremont, A. Ferreira, “Flexible microscope calibration
precision mobile transducers for sensing, the existence of a using virtual pattern for 3-D telemicromanipulation,” in Proceedings of
natural robot-to-robot communication channel via magnetic the 2005 IEEE Int’l Conf. on Robotics and Automation, pp. 3888–3893.
field interactions, and the potential for on-board replication [20] R. Pelrine, A. Wong-Foy, A. Hsu, B. McCoy, “Self-assembly of milli-
with all the system adaptability and self-repair that implies. We scale robotic manipulators: A path to highly adaptive, robust automation
hope that future research in these and related areas will help systems,” in Proc. 2016 Int’l. Conf. on Manipulation, Automation and
elucidate these intriguing multi-agent aspects of DM3 systems. Robotics at Small Scales (MARSS), IEEExplore ISBN no. 978-1-5090-
1510-8.