[ieee 2011 ieee conferecne on technologies for practical robot applications (tepra) - woburn, ma,...
TRANSCRIPT
![Page 1: [IEEE 2011 IEEE Conferecne on Technologies for Practical Robot Applications (TePRA) - Woburn, MA, USA (2011.04.11-2011.04.12)] 2011 IEEE Conference on Technologies for Practical Robot](https://reader031.vdocuments.us/reader031/viewer/2022020203/5750a4c21a28abcf0caccf20/html5/thumbnails/1.jpg)
Practical Aerial Grasping of Unstructured Objects
Paul E. I. Pounds, Member, IEEE, Daniel R. Bersak and Aaron M. Dollar, Member, IEEE
Abstract— We aim to extend the functionality of UnmannedAerial Vehicles (UAVs) beyond passive observation to activeinteraction with objects. Of particular interest is graspingobjects with hovering robots. This task is difficult due to theunstable dynamics of flying vehicles and limited positionalaccuracy demonstrated by existing hovering vehicles. Con-ventional robot grippers require centimetre-level positioningaccuracy to successfully grasp objects. Our approach employspassive mechanical compliance and adaptive underactuation ina gripper to allow for large positional displacements between theaircraft and target object. In this paper, we present preliminaryanalysis and experiments for reliable grasping of unstructuredobjects with a robot helicopter. Key problems associated withthis task are discussed, including hover precision, flight stabilityin the presence of compliant object contact, and aerodynamicdisturbances. We evaluate performance of the initial proof-of-concept prototype and show that this approach to object captureand retrieval is viable.
I. INTRODUCTION
Unmanned Aerial Vehicles (UAVs) are rapidly evolving
into capable mobility platforms that manoeuver, navigate and
survey with much autonomy. However, UAVs ‘look, but don’t
touch’, avoiding contact with their surroundings. A natural
progression is to increase their scope beyond observation
to include interaction with objects. Rotorcraft, in particu-
lar, could hover over objects in order to manipulate them
(Fig. 1). Robust object retrieval and deposition capabilities
are compelling for aerial vehicles because of their unique
abilities — fast traversal of impassible terrain, movement
around 3D environments and unlimited vertical workspace.
Potential applications include object retrieval, intelligence
gathering, explosives disposal and courier services.
The idea of aerial payload retrieval is not new: early
aviators used long hooks to snare mail sacks on the ground as
the aircraft passed overhead [30], allowing towns lacking an
airstrip to send airmail. In the 1950s the Fulton skyhook was
developed by the CIA for retrieving packages and personnel
in the field [6]. The system consisted of a balloon tethered to
the payload. As the retrieving aircraft flew into the tether, a
special prong mounted to the nose snagged the line, drawing
the payload up into the fuselage slipstream for recovery.
Limited examples of autonomous aircraft interacting with
objects have been demonstrated, such as in-flight refueling
[27], [18] and the transport of slung loads, both individually
and cooperatively [3], [2], [20]. In these examples the
interacting object is either not contacted automatically (such
This work was supported in part by the Office of Naval Research grantN000141010737.
P. E. I. Pounds, D. R. Bersak and A. M. Dollar are with Dept. ofMechanical Engineering, Yale University, New Haven CT, 06511, USA.{firstname.lastname}@yale.edu
Fig. 1. Yale Aerial Manipulator Interacting with an Object.
as a load attached by a human operator on the ground) or
else highly structured (eg. refueling drogues with optical
markers). Efforts to develop autonomous helicopter UAV
payload acquisition have similarly relied on structuring of
the target object to simplify the task. One approach consists
of a hanging magnet at the end of a probe to collect ferrous
objects [1], [4]. Another method employs a hook at the end of
a probe to snag a hoop on an object [13]. The need for heavy
structuring of the object limits the potential utility of these
methods. We aim to demonstrate generalized object retrieval
and transport of unstructured objects from aerial platforms.
II. OBJECTIVES
We are developing the Yale Aerial Manipulator, a robot
helicopter integrated with a gripper, to hold and transport
objects (Fig. 1). Capturing an object with a helicopter UAV
poses challenges due to the precise hover positioning needed
to grasp a target with a conventional gripper, and the inherent
instability of helicopters.
Small UAVs are sensitive to gusts and disturbances during
hover which makes maintaining a single position difficult.
Typical UAV station-keeping accuracy is tens of centime-
tres; the Rotomotion SR-100 UAV has a reported 200 mm
position-keeping accuracy [13]. Common positioning meth-
ods used by UAVs, such as GPS, are not yet accurate enough
to place the helicopter within the very small bounds required
for precise manipulation with rigid grippers. The amount of
positional error a typical end-effector may tolerate in order
to capture an object is significantly smaller than such an
aircraft can accommodate; misalignment of the gripper by
as little as 3 mm can lead to failed grasps and large contact
forces. As helicopters are inherently unstable, such forces
can potentially destabilize the aircraft, leading to a crash.
There are several obstacles to be overcome in developing
grasping functionality, specific to the aerial manipulation
task. A naive approach to aerial grasping would make use
of a conventional stiff robot gripper to take hold of objects.
978-1-61284-481-7/11/$26.00 ©2011 IEEE 99
![Page 2: [IEEE 2011 IEEE Conferecne on Technologies for Practical Robot Applications (TePRA) - Woburn, MA, USA (2011.04.11-2011.04.12)] 2011 IEEE Conference on Technologies for Practical Robot](https://reader031.vdocuments.us/reader031/viewer/2022020203/5750a4c21a28abcf0caccf20/html5/thumbnails/2.jpg)
However, this accentuates the factors that make the aerial
grasping task challenging: the need to precisely position the
aircraft over the target object for grasping in the presence
of aerodynamic disturbances, and the inherent instability of
helicopters.
Our approach is to instead use a compliant underactuated
manipulator based on the SDM gripper [8]. The character-
istics of the design — open-loop adaptive grasping, wide
finger span, insensitivity to positional error up to 15 cm —
closely match the challenges of the task. The compliance
in the grasping mechanism makes this problem tractable by
decoupling the motion of the aircraft from the object and
preventing transmitted forces from destabilizing the vehicle
[22]. Underactuation in the gripper allows for adaptivity in
finger joint configuration, with grasps conforming to the
surfaces of irregular objects, producing more robust grips.
Unlike rigid grippers, hooks, magnets or sticky pads, it
requires no preparation or structuring of the target object
(no catches, ferromagnetic material, flat surfaces) in order
to successfully capture payloads in the presence of large
positioning errors.
In this paper we identify key problems associated with
aerodynamics and flight control in aerial grasping. We de-
scribe how underactuated, compliant grippers can be used
to simplify the capture task. A proof-of-concept system for
use in grasping experiments, consisting of a gripper module
mated to a small-scale helicopter platform, is presented. The
system is used to grasp numerous objects and to capture op-
erational performance data. Results from these experiments
are given, along with an analysis of observed phenomena. A
brief conclusion summaries our findings.
III. AERIAL MANIPULATION CHALLENGES
There are several obstacles in developing grasping capabil-
ities hovering platforms. Limited hover positioning accuracy,
flight instability and shift in aircraft trim must all be over-
come to allow successful recovery of an object on the ground
by a helicopter.
A. Hover Station-Keeping Performance
The success of any grasp is highly dependent on placing
the gripper squarely over the target object. Small VTOL
UAVs are sensitive to gusts and disturbances during hover
which makes maintaining a single position difficult. Typical
station-keeping accuracy is tens of centimeters; the Rotomo-
tion SR-100 UAV has a reported 200 mm position-keeping
accuracy [13].
The amount of positional error a typical end-effector
may tolerate in order to capture an object is significantly
smaller than such an aircraft can accommodate. Existing
sensor modalities such as GPS, vision or inertial systems
are not accurate enough to localize the vehicle with sub-
centimeter precision. While it should be possible to im-
prove the position-keeping performance of VTOLs through
advancements in flight control, sensing and aircraft respon-
siveness, this requires substantial work outside the scope of
our current efforts.
Fig. 2. Helicopter Dynamic Model with Elastic Tether.
For an aerial grasping system to be practical, the gripper
must successfully grasp with offsets greater than the range
of landing position error. Compliance in the gripper reduces
sensitivity of the grasp to positional error. Previous test-
bench results of the SDM Hand demonstrated acceptable
object misalignment of up to 100 mm. Crucially, compliance
also reduces the forces transmitted to the airframe through
the gripper; there is little margin to compensate for distur-
bances introduced by contact with the object during landing.
B. Coupled Flight Stability
During contact with the environment, stability of the
aircraft in flight must be guaranteed [10]. When the aircraft
contacts an object, but has not yet lifted it, ground forces are
imparted to the airframe through the manipulator (Fig. 2).
These forces couple with the motion of the helicopter to
produce distinctly different dynamics from free flight. The
gripper can be treated as an elastic tether connecting the
helicopter to a fixed object.
Previous tethered helicopter stability research has consid-
ered the helicopter flying far from the tether point, where the
line tension and direction is approximately constant and the
tether cannot support moments [12], [26], [21]. In this case,
however, the elastic tether link has both prismatic and angular
reaction forces which change sign as the aircraft moves over
the equilibrium point. If not correctly compensated-for by the
flight controller, these forces will destabilize the helicopter,
leading to a crash. Rather than develop special controllers
solely to stabilize the aircraft during the aerial manipulation
task, it is desirable to show that a standard Proportional
Integral Derivative (PID) control architecture will remain
stable during hovering object capture for a given gripper and
helicopter configuration.
Passive ground effect and inflow damping produces in-
herently stable vertical dynamics in rotorcraft; added tether
elasticity does not destabilize vertical motion. However,
the combined dynamics of the object-helicopter system in
coupled horizontal and pitch motion can be destabilized by
certain tether stiffness configurations. It was shown in earlier
analysis that a hovering helicopter grasping an object fixed
to the ground with a compliant gripper under PID control
100
![Page 3: [IEEE 2011 IEEE Conferecne on Technologies for Practical Robot Applications (TePRA) - Woburn, MA, USA (2011.04.11-2011.04.12)] 2011 IEEE Conference on Technologies for Practical Robot](https://reader031.vdocuments.us/reader031/viewer/2022020203/5750a4c21a28abcf0caccf20/html5/thumbnails/3.jpg)
will be stable for if the ratio of lateral and angular tether
stiffnesses, kx and kθ are below some value:
kxk′θ
<mh
I(h+ q1dz)(1)
where m is the mass of the helicopter, h is the height of
the rotor above the helicopter center of gravity, I is the pitch
rotational inertia, q1 is rotor flapping angle linear velocity
coupling and dz is the tether length [22]. Intuitively, a
helicopter with zero lateral stiffness and free to slide whilst
held level by high angular stiffness will not be unstable,
whereas a helicopter held with a pin joint and free to rotate
will pivot into the ground.
C. Payload and Trim
Given a successful grasp, the helicopter must then trans-
port its payload. Helicopter rotors produce a limited amount
of thrust to lift the combined weight of the vehicle and cargo
plus an additional fraction to allow for vertical acceleration,
called the ‘thrust margin’. Typical thrust margin for vertical
take-off and landing UAVs in hover is 30 percent of the
vehicle weight [23], while specialized acrobatic model he-
licopters can have thrust margins in excess of 100 percent.
These aircraft can carry up to half of their unloaded weight,
and still maintain 30 percent thrust margin.
Conventional helicopter balancing places the payload’s
center of mass coaxial with the rotor shaft so that the thrust
force is aligned with the load. Due to the limited positioning
accuracy of current hovering platforms and the uncertainty
inherent with grasping a potentially unknown object, the
payload may be significantly off-center when grasped.
Small weight imbalances can be trimmed out by adjusting
the cyclic blade pitch inputs, so that the thrust vector is
tilted to pass through the true center of mass. One degree
of cyclic blade pitch results in one degree of thrust angle
deflection [15]. A typical small helicopter has a useful blade
pitch range of 20 degrees [24], which requires that the
center of mass of the helicopter/gripper/object system fall
within a cone of approximately 10 degrees from the rotor
axis (Fig. 3a). However, utilizing the full range of cyclic
control to actively trim out load offsets comes at the expense
of limited maneuvering control. An unbalanced helicopter
has less ‘maneuvering margin,’ the available cyclic range to
affect maneuvers [15]. If the combined center of mass is
sufficiently far from the rotor axis that it lies outside the
cyclic cone, the aircraft cannot counteract the offset and
regain trim (Fig. 3b).
IV. GRIPPER FOR AERIAL GRASPING
As described above, conventional robot graspers require
a level of precision incompatible with that of a typical
helicopter platform. The SDM Hand has two crucial fea-
tures that enable reliable grasping in the presence of large
positioning errors: passive mechanical compliance in the
joints, and an adaptive and underactuated transmission. These
have substantial advantages in facilitating the capture of
target objects from hovering platforms with poor positional
accuracy.
Fig. 3. Cyclic Control Limits of a Helicopter: Center of mass inside (a)and outside (b) cyclic range.
A. Mechanical Compliance
One approach to dealing with positional uncertainty is
through compliance in the grasping mechanism, so that offset
errors do not result in large forces and the mechanism
conforms to the object.
Robot compliance has often been considered in the context
of active control, where sensors and actuators are used to
achieve a desired force-deflection relationship, with many
studies devoted to impedance analysis and control (e.g.
[25], [19], [7]). However, this requires the active use of
position/velocity and force/torque sensor signals in the robot
joints or end-effector. This approach fails in impacts, where
the speed of the force transients is too fast to use sensing
and control to avoid damage, as well as when contact occurs
at locations on the mechanism that are not sensorized.
In contrast, passive mechanical compliance, implemented
through springs in robot joints, allows for large joint deflec-
tions and can ensure low contact forces. This can minimize
disturbance to objects during the first phases of acquisition
and in impacts, where active stiffness control often fails. The
elimination of the sensing required for active compliance can
also lower weight and increase durability.
B. Adaptive Underactuation
It is desirable for a gripper to have many links and
joints to allow for object conformation and form closure.
However, due to limitations in space, manipulator payload
and available control channels, there is a need to simplify
the number of actuators and, hence, controlled degrees of
freedom of robot hands.
Mechanical coupling allows many joints with few actua-
tors and can take the form of both fixed-motion joint coupling
and underactuation. A ‘fixed-motion coupled’ hand has more
joints than degrees-of-freedom, with each degree-of-freedom
controlled by a dedicated actuator, affording the mechanism
no adaptability (e.g. [5], [17]). In these hands, motion of
one joint always results in a fixed proportional motion of
the joint(s) coupled to it. In the same way, if contact occurs
on one joint thus fixing its position, all coupled joints are
thereby fixed. An underactuated hand has fewer actuators
than degrees-of-freedom, and therefore demonstrates adap-
tive behavior. In these hands, motion of the distal links can
continue after contact on the coupled proximal links occurs,
allowing the finger to passively adapt to the object shape
101
![Page 4: [IEEE 2011 IEEE Conferecne on Technologies for Practical Robot Applications (TePRA) - Woburn, MA, USA (2011.04.11-2011.04.12)] 2011 IEEE Conference on Technologies for Practical Robot](https://reader031.vdocuments.us/reader031/viewer/2022020203/5750a4c21a28abcf0caccf20/html5/thumbnails/4.jpg)
Fig. 4. Yale Aerial Manipulator Systems.
(e.g. [9], [28], [29], [11], [14], [16]). By carefully selecting
joint coupling schemes, much of the functionality of a hand
can be retained while reducing the number of actuators and
the overall complexity of the grasping mechanism.
A key aspect of the hand’s tendon transmission design
is that it allows the hand to be very compliant during
approach and initial contact, and stiffer during grasp due
to the compliance in the fingers being placed in parallel
with the actuator. Before the hand is actuated, the tendon
cables remain slack and the hand is in its most compliant
state, keeping contact forces low and maximizing passive
adaptability. After actuation, the tendon is tensioned, taking
much of the compliance out of the fingers, resulting in a
stiffer grasp with greater stability. This method permits use
of actuators that are not backdrivable and prevents inertial
load of the actuator from increasing the passive stiffness.
V. HARDWARE PLATFORM
A proof-of-concept prototype was constructed to test the
feasibility of aerial grasping with a gripper. The demonstrator
consists of a compliant underactuated gripper fitted to a
commercial radio-controlled helicopter.
The helicopter platform is a T-Rex 600 ESP (Align RC,
Taiwan), a large high-end fully-electric aircraft with a 1.35 m
diameter rotor driven by a 1.6 kW brushless motor, boasting
a maximum carry weight of 1.8 kg (Fig. 4). The gripper
is mounted ventrally between the skids. The skids retract,
allowing the gripper unobstructed access to objects on the
ground. The aircraft is stabilised by Helicommand RIGID
commercial off-the-shelf control system (Captron, Germany),
and directed by a human pilot. The Helicommand systems
regulate aircraft flight attitude with inertial sensors and
use a downward-pointing optical system to compensate for
drift. The system was later upgraded to a Helicommand
Profi, which adds additional optics and a pressure sensor
to allow hover position-keeping and height stabilisation. The
T-Rex has a cyclic control range of 20◦; when fitted with
the gripper, this gives the platform a maximum allowable
payload offset of 53 mm for a 1 kg load.
Compact routing of the adaptive gripper transmission is
required to reduce weight and size for mounting in the
available space under the helicopter. The prototype gripper
is actuated by a pair of HSR-5990TG high-performance
hobby servos (Hitec, South Korea) operated in parallel and
controlled by an onboard microprocessor. A pair of sense
Fig. 5. Flexible Compliant Gripper Module.
resistors measures current to each servo and the micropro-
cessor modulates the servo position commands to regulate
constant effort applied by the gripper. The servo controller
also limits the stall current of the servos, preventing damage
from overheating. The complete gripper assembly, including
servos, batteries and transmission, weighs less than 750 g.
As the helicopter can only approach objects from above,
all grasps are finger-tip or power grasps with opposing
finger forces. Buckling modes in finger-tip grasps can lead
to azimuthal instability, due to the limited torsion stiffness
of the proximal joints. Early experiments showed that this
twisting would regularly eject objects from a fingertip grasp
as, unlike in a power-grasp, the flats of the finger pads had
insufficient contact to stabilize the twist motion. To improve
grasp stability, ‘syndactylic bars’ joining the proximal links
of adjacent digits together were added, constraining them
to move as a pair while still allowing the distal links to
tension independently. This adaptation prevents the proximal
joints from twisting due to a force couple induced by
the intermeshing finger configuration. The addition of the
syndactylic bars has largely eliminated this effect, but at the
expense of lower overall adaptability of the gripper.
To acquire an object, the helicopter approaches the target,
descends vertically to land over the target and then closes its
gripper. Once a solid grasp is achieved, the aircraft ascends
with the object. The UAV manipulation system has thus far
demonstrated grasp and transportation of objects up to 1.5 kg.
In bench-tests, the gripper was able to stably grasp a range
of sizes and shapes of objects, with sufficient force to lift
masses well above the helicopter’ lifting capacity (>2 kg).
VI. EXPERIMENTAL PERFORMANCE EVALUATION
The ability of the system to grasp a variety of objects
(Fig. 6), and the reliability of the system grasping a single
object repeatedly, were tested experimentally. For these tests,
an expert pilot directed the helicopter with a radio handset,
taking care to avoid ground vortex phenomena. While this
introduces non-repeatable human influences, the expert pilot
is considered to be a near optimal controller with perfor-
mance representational of that of an equivalent autonomous
system.
102
![Page 5: [IEEE 2011 IEEE Conferecne on Technologies for Practical Robot Applications (TePRA) - Woburn, MA, USA (2011.04.11-2011.04.12)] 2011 IEEE Conference on Technologies for Practical Robot](https://reader031.vdocuments.us/reader031/viewer/2022020203/5750a4c21a28abcf0caccf20/html5/thumbnails/5.jpg)
Fig. 6. Target Objects. Fig. 7. Flight Trajectory for Capture. Fig. 8. Onboard Camera View.
For each grasping attempt, the pilot flew the helicopter
from a starting position greater than 1 m away from the
target, attempted to come to a hovering stand-still above the
target, and descend vertically to the ground (viz. Fig. 7). Less
than 30 seconds were spent on each attempt. The pilot stood
4.25 m up-range from the target, behind the helicopter.
The system was used to attempt to grip a series of
objects of different sizes, shapes and masses (Fig. 9b): a
softball (160 g, 89 mm), a PVC tube (390 g, 280 mm),
a water bottle (600 g, 220 mm), a wood block (700 g,
265 mm) and a cylinder (900 g, 390 mm). Each object was
attempted several times, recorded by an onboard camera and
an offboard camera. In each trial, the lateral positioning error
was extracted from the onboard camera, and the longitudinal
error from the off-board camera, utilizing known reference
points on the helicopter and the object edges.
The grasps were completed over two sets of trials. A
grasp was considered ‘successful’ if the gripper held the
object and the helicopter could lift, transport and return the
object without dropping it. Two objects, the block and hollow
tube, had no failed grasps. The water bottle, cylinder, and
softball had 67 per cent, 70 per cent and 80 per cent success
rates respectively. The gripper demonstrated object capture
even with significant positional error, successfully grasping
objects with as much as 103.5 mm longitudinal offset and
46.7 mm lateral offset.
The reliability of the gripper was tested by landing on an
object known to demonstrate good grasping performance, the
700 g wood block (Fig. 9c). The block center of mass was
determined using the onboard camera.
Twenty attempts were made of which all grasps succeeded,
even with position errors in excess of 100 mm longitudinally
and 40 mm laterally. Although the increasing experience of
the pilot in performing the grasping task may play a role in
the perfect result of this experiment, all grasp attempts of
the block in the previous trials were also successful.
As the gripper is constrained to always grasp objects
from above, all grasp attempts are functionally similar.
Consequently, with one exception, the failure mode observed
during gripping consisted of the object being grasped and
lifted, but slipping out of one pair of fingers, and rotating
downwards before sliding out of the grasp entirely. In the
exception, the softball had an displacement error 90 percent
of its diameter, which instead caused the gripper to eject it
during the grasp attempt.
Fig. 9. Assorted Object Grasp Scatter Plot (a), 700 g Block Grasp ScatterPlot (b).
In no trials did the gripper entirely fail to grip and lift an
object which was within its grasp, even if it did not retain
it. Only one object within 40 mm of center failed grasping.
Objects were generally more sensitive to lateral error than
to longitudinal error, primarily due to the aspect ratio of the
tested objects, with most successful attempts being within
0–40 mm of the aircraft centerline.
The geometries that gripped best were rough textured
objects with flat sides that allowed the gripper to resist slip
and rotation. The block performed especially well, as the top
edges of the block contact the finger pads of the proximal
links as the block rotated downwards, producing a stable grip
akin to a power grasp. The water bottle performed poorly
because its smooth sides and uneven contours promoted
slipping, and as the bottle rotated the grasping pair of fingers
would squeeze the object out of the grip. The heavy cylinder
slipped easily out of grasp on one end, but was too big to be
ejected by the grip; instead, its great weight pulled it free of
the two gripping fingers due to insufficient friction to keep
it in place.
Across the objects attempted, the helicopter grasping sys-
tem demonstrated 85 per cent success out of more than
50 grasp attempts, rising to 100 per cent success for the
best performing objects. However, the objects chosen did
not span the entire range of potential target objects. The
current gripper design was not optimized for the described
application, and significant future work will be devoted to
enable grasping a wider range of object size and shape,
particularly for small objects.
103
![Page 6: [IEEE 2011 IEEE Conferecne on Technologies for Practical Robot Applications (TePRA) - Woburn, MA, USA (2011.04.11-2011.04.12)] 2011 IEEE Conference on Technologies for Practical Robot](https://reader031.vdocuments.us/reader031/viewer/2022020203/5750a4c21a28abcf0caccf20/html5/thumbnails/6.jpg)
A. Observed Phenomena
During the experiments the helicopter flight controller
maintained control of the vehicle at all stages. Although the
grasp offset expected from the target landing trials would
suggest that there is a high likelihood of grasping a heavy
object outside the range where the cyclic control margin
would saturate, this did not occur during the tests. Once
grasped, the weight of off-center objects caused the gripper
to droop, bringing the mass of the payload closer to the rotor
axis, and inside the limits of the control margin.
The greatest applied torque during the test was from the
900 g cylinder with an offset of 91 mm (applying a 0.8 Nm
bias load) which was carried without ill effect. This same
object, the largest object, could be released by the operator
mid-flight without producing large pitch or roll excursions.
VII. CONCLUSION
We have demonstrated a helicopter UAV-gripper system
capable of reliably grasping and retrieving objects under
human control. This system exploits the unique performance
capabilities of an underactuated, compliant gripper to directly
address the particular challenges of helicopter imprecision.
The system was able to grasp a variety of objects including
blocks, balls, bottles and cylinders, ranging from 160 g to
900 g. To our knowledge, this is the first time a helicopter
UAV has directly captured and retrieved objects with a
gripper. As preliminary approach to the problem, the gripper
design is not fully optimized for this task.
The most difficult object could be grasped 67 percent of
the time, and the easiest object 100 percent of the time. Load
bias disturbances of 0.8 Nm applied to the airframe by the
payload were rejected by the flight controller around hover.
Future work will consider tuning the gripper for improved
grasping and considering control aspects of grasping during
flight and from unstable platforms.
VIII. ACKNOWLEDGEMENT
The authors would like to thank Joe Acosta of Build
Right Fly Right Hobbies, Wallingford CT, USA, and Gre-
gory Brown for their support of this project. This project
was funded in part by US Office of Naval Research grant
#N000141010737.
REFERENCES
[1] Amidi O., Kanade T. and Miller J., “Vision-Based Autonomous Heli-copter Research at Carnegie Mellon Robotics Institute 1991–1997,” InProc. American Helicopter Society International Conference, 1998.
[2] Bernard M. and Kondak K., “Generic Slung Load TransportationSystem Using small Size Helicopters,” In Proc. IEEE InternationalConference on Robotics and Automation, 2009.
[3] Bisgaard M., Bendtsen J. and la Cour-Harbo A., “Modelling of GenericSlung Load System,” Journal of Guidance, Control and Dynamics, Vol.32, No. 2, 2009.
[4] Borenstein J., “The Hoverbot — An Electrically Powered FlyingRobot,” Unpublished paper, University of Michigan, 1992, http://www-personal.umich.edu/ johannb/hoverbot.htm (2009).
[5] Butterfass J., Grebenstein M., Liu H. and Hirzinger G., “DLR-HandII: Next Generation of a Dextrous Robot Hand,” In Proc. InternationalConference on Robotics and Automation, 2001.
[6] Central Intelligence Agency, “Robert Fulton’s Skyhook andOperation Coldfeet,” https://www.cia.gov/library/center-for-the-study-of-intelligence/csi-publications/csi-studies/studies/95unclass/Leary.html(Sept. 2010).
[7] Cutkosky M.R. and Kao I., “Computing and Controlling the Com-pliance of a Robotic Hand,” IEEE Transactions on Robotics andAutomation, Vol. 5, No. 2, pp 151–165, 1989.
[8] Dollar A.M. and Howe R.D., “A Robust Compliant Grasper via ShapeDeposition Manufacturing,” IEEE/ASME Transactions on Mechatron-ics, Vol. 11, No. 2, pp 154–161, 2006.
[9] Dollar A.M. and Howe R.D., “Simple, Robust Autonomous Graspingin Unstructured Environments,” In Proc. IEEE/RAS International Con-ference on Robotics and Automation, 2007.
[10] Gentili L. and Naldi R. and Marconi L., “Modeling and Controlof VTOL UAVs Interacting With the Environment,” in Proc. IEEEConference on Decision and Control, 2008.
[11] Hirose S. and Umetani Y., “The Development of Soft Gripper forthe Versatile Robot Hand,” Mechanism and Machine Theory, Vol. 13,pp 351–359,1978.
[12] Kaufman L., and Schultz E., “The stability and control of tetheredhelicopters,” Journal of the American Helicopter Society, 7(4), 1962.
[13] Kuntz N., and Oh P., “Development of Autonomous Cargo Transportfor an Unmanned Aerial Vehicle Using Visual Servoing,” In Proc.Digital Systems and Control Conference, 2008.
[14] Laliberte T., Birglen L. and Gosselin C., “Underactuation in RoboticGrasping Hands,” Machine Intelligence & Robotic Control, Vol. 4, No.3, pp 1–11, 2002.
[15] Leishman J.G., “Principles of Helicopter Aerodynamics, 2nd Ed.”,Cambridge University Press, New York, 2006.
[16] Lotti F., Tiezzi P., Vassura G., Biagiotti L., Palli G. and Melchiorri C.,“Development of UB Hand 3: Early Results,” In Proc. InternationalConference on Robotics and Automation, 2005.
[17] Lovchik C.S. and Diftler M.A., “The Robonaut Hand: A DexterousRobot Hand for Space,” In Proc. IEEE/RAS International Conferenceon Robotics and Automation, 1999.
[18] Mammarella M., Campa G., Napolitano M., Fravolini M. and Per-hinschi R., “Machine Vision/GPS Integration Using EKF for the UAVAerial Refueling Problem,” IEEE Transactions on Systems, Man andCybernetics, Vol. 38, No. 6, 2008.
[19] Mason M.T., “Compliance and Force Control for Computer controlledManipulators,” IEEE Transactions on Systems, Man, and Cybernetics,Vol. 11, No. 6, pp 418–432, 1981.
[20] Michael N., Fink J. and Kumar V., “Cooperative Manipulation andTransportation with Aerial Robots,” In Proc. Robotic Science andSystems, 2009.
[21] Oh S.R., Agrawal S., Pathak K., Pota H.R., and Garrett, M., “Ap-proaches for a tether-guided landing of an autonomous helicopter,”IEEE Transactions on Robotics, 22, pp. 536–544, 2006.
[22] Pounds P.E. and Dollar A.M., “Hovering Stability of HelicoptersWith Elastic Tethers,” In Proc. ASME Dynamic Systems and ControlConference, 2010.
[23] Pounds P. and Mahony R., “Small-Scale Aeroelastic Rotor Simulation,Design and Fabrication,” In Proc. Australasian Conference on Roboticsand Automation, 2005.
[24] Prouty R., Helicopter Performance Stability and Control, Krieg Pub-lishing Company, 2002.
[25] Salisbury K.J., “Active Stiffness Control of a Manipulator in CartesianCoordinates,” In Proc. IEEE Conference on Decision and Control,pp 95–100, 1980.
[26] Schmidt G. and Swik R., “Automatic hover control of an unmannedtethered rotorplatform,” Automatica, 10(4), pp. 393 – 394, IN1–IN2,395–403, 1974.
[27] Scott D., Toal M. and Dale J., “Vision Based Sensing for AutonomousIn-flight Refueling,” In Proc. of SPIE, Vol. 6561, 2007.
[28] Townsend W.T., “The BarrettHand Grasper - Progammably FlexiblePart Handling and Assembly,”Industrial Robot – An International Jour-nal, Vol. 10, No. 3, pp 181–188, 2000.
[29] Ulrich N. and Kumar V., “Grasping Using Fingers With CoupledJoints,” In Proc. ASME Mechanisms Conference, 1988.
[30] U.S. Centennial of Flight Commission, “Air-mail and the Growth of the Airlines,”http://www.centennialofflight.gov/essay/Government Role/1930-airmail/POL6.htm (Sept. 2010).
104