Contact-less manipulation of millimeter-scale objects via ultrasonic levitation
CContact-less Manipulation of Millimeter-scale Objects via UltrasonicLevitation
Jared Nakahara , Boling Yang , and Joshua R. Smith , Dept. of Electrical and Computer Engineering and Paul G. Allen School of Computer Science and Engineering
Abstract — Although general purpose robotic manipulatorsare becoming more capable at manipulating various objects,their ability to manipulate millimeter-scale objects are usuallyvery limited. On the other hand, ultrasonic levitation deviceshave been shown to levitate a large range of small objects, frompolystyrene balls to living organisms. By controlling the acousticforce fields, ultrasonic levitation devices can compensate forrobot manipulator positioning uncertainty and control thegrasping force exerted on the target object. The materialagnostic nature of acoustic levitation devices and their abilityto dexterously manipulate millimeter-scale objects make themappealing as a grasping mode for general purpose robots. Inthis work, we present an ultrasonic, contact-less manipulationdevice that can be attached to or picked up by any generalpurpose robotic arm, enabling millimeter-scale manipulationwith little to no modification to the robot itself. This device iscapable of performing the very first phase-controlled pickingaction on acoustically reflective surfaces. With the manipulatorplaced around the target object, the manipulator can graspobjects smaller in size than the robot’s positioning uncertainty,trap the object to resist air currents during robot movement,and dexterously hold a small and fragile object, like a flowerbud. Due to the contact-less nature of the ultrasound-basedgripper, a camera positioned to look into the cylinder caninspect the object without occlusion, facilitating accurate visualfeature extraction.
I. I
NTRODUCTION
Over the past decade, robots have become increasinglycapable of dexterous manipulation, accurate object poseestimation, sophisticated motion planning, and improvedmotor skills [1], typically achieved by arm and manipulatordesigns with high degrees of freedom. However, in mostgeneral purpose robots, dexterity is limited to manipulatinglarger objects on the scale of centimeters or above, lack-ing the ability to perform the same level of manipulationon smaller objects. With limited gripping force resolutionand positioning accuracy, these robots can miss or damagethe target object in cases when the object of interest issmall or fragile. However, millimeter-scale manipulation isa skill required for daily life, scientific research, and inthe manufacturing industry. For example, in biology, neuralscience, and other related research areas, experiments thatinvolve handling small, deformable and fragile objects likeinsects [2]–[4], biological tissues [5], [6], and drops offluid [7], [8], are very common. In microassembly and PCBmanufacturing industries, objects like bare silicon dyes andelectronic components are assembled with machines thatrequire an extra level of precision because of their size andfragility. The increasing need to automate and improve theseexperiments and manufacturing processes [9]–[14] highly
Fig. 1. The PR2 robot is using the contact-less ultrasonic manipulator tohold a flower. motivates further research into enabling general purposerobots to manipulate millimeter-scale objects.Acoustic levitators and traps have demonstrated the ca-pability to manipulate objects on the scale of millimetersthrough the processes of localized pressure modulation viahigh frequency acoustic wave interference, at frequenciesinaudible to the human ear. By taking advantage of the non-linearity of sound in air, sound can be used to generate lift onan object within the acoustic field, allowing it to overcomegravitational forces. In this work, we present an acousticlevitation device that allows general purpose robots to pickand manipulate millimeter-scale objects. This paper makesthe following contributions: • Development of the first acoustic levitation device andcontrol scheme that can pick up an object from a tabletop • Implementation of an acoustic manipulator that can beattached to or be picked up by general purpose robotswith little to no robot modificationWhile other publications have shown that small objects canbe acoustically levitated and manipulated, the contact-lessultrasonic robotic manipulator is the first device to enable a r X i v : . [ c s . R O ] F e b general purpose robot to pick up small millimeter-scaleobjects from an acoustically reflective surface, such as a tabletop.This acoustic manipulator’s size and weight allow it to beeasily attached to most general purpose robots. The levitationdevice can manipulate objects that are placed within an areacalled the basin of attraction . This large manipulation areaadds extra robustness to compensate for robot positioninguncertainty.In addition, we use an acoustic field modeling method,called the method of images, to improve the simulatedmodel of force dynamics inside the levitator. This modelingmethod improves the dexterity of our acoustic manipulatorby enabling novel acoustic manipulation skills includingpicking object from a flat surface without the need for objectsupport structures or specialized dispensing of the object.The manipulator was tested for real world usability viaa PR2 robot, where the PR2 attempted to pickup an ob-ject, extract visual features from the object, and performobject sorting based on the extracted features. We foundthat the acoustic levitator allows general purpose robots tomanipulate millimeter scale objects, by providing “in-hand”dexterity and enhanced tolerance to the robot’s kinematicuncertainty. Further more, the contact-less nature of thismanipulator provides an unblocked view of the whole object.To the best of our knowledge, this is the first research thathas successfully enabled a PR2 robot (or robots of similarform factors) to pick up and inspect small and fragile objectslike mosquitoes, integrated circuit chips, and flowers withoutdamaging them. II. R ELATED W ORK
In this section, we will review the previous work onenabling high accuracy and gentle manipulation from the per-spective of general purpose, industrial, and medical robotics,and discuss previous work in acoustic levitation devices.To manipulate small objects, a robot should have sufficientmanipulation accuracy and dexterity. Although dexterousmanipulation has been a popular research topic, explorationof millimeter scale object manipulation on general purposerobotic platforms has been extremely limited. Many groupshave developed solutions to reduce the positioning uncer-tainty and increase dexterity of general purpose robots, [15]–[17], yet none of the objects tested have volumes closeto millimeter scale. Industrial robotic arms are known forhaving great positioning accuracy, equipped with specializedmanipulators and sensors, and are capable of the auto-matic assembly of small screws [18]–[20]. However, objectfragility has not been considered by any of the researchabove.In contrast, millimeter scale manipulation is very commonin the field of surgical robotics [21]. Surgical manipulatorshave great manipulation accuracy and dexterity [22]–[24],yet their actuation system, kinematics and end-effectors aredesigned for surgical scenarios. This level of specializationcould make transferring design concepts to general purpose robots difficult due to the broadness of tasks these robots arerequired to perform.On the other hand, robots can handle fragile objectsthrough constant object state estimation and adjusting grasp-ing strategy accordingly. One effective way to monitor objectstate through a manipulation process is to mount sensors onthe end-effectors of a robot. Lancaster et. al. [25] createda fingertip-mounted sensor that provides proximity, contactdetection, and force sensing. Koyama et. al. [26] designed aproximity sensor that allows a robot hand to catch a fallingpaper balloon with insignificant deformation. Lee et. al. [27]proposed a soft linear actuator allowing a robot hand to holda potato chip without crushing it. Likewise, development ofsoft robotic grippers is another popular research directiontoward gentle manipulation [28]–[32]. However, the abovemethods have only been tested on larger objects and willfail to manipulate millimeter scale objects.Acoustic levitation devices use a propagating pressurewave to exert a lifting or trapping force on an object.Xie [33] showed that a single-axis acoustic levitator wascapable of lifting bio-materials, small living organisms, andtissue, in air, without damage. Some traps are capable ofperforming dexterous manipulation of objects in air [34]and water [35] using Bessel functions. Others are capableof pseudo-picking actions [36], translation, and rotation [37]of objects using coordinated phase delays. Kozuka et. al.[38] showed that small objects could be manipulated in airby controlling the phase of two interfering sound beams. Inthese devices, objects were placed inside the levitator andobjects were translated using a combination of phase delaysand acoustic source axes deflections. It was also found thatslight variations in acoustic source frequencies could causeobject translation.Other applications of acoustic levitation based manipu-lators include display technology, [39], and a microassem-bly device, [40]. These devices use large arrays of phasecontrolled transducers to translate small particles, therebydemonstrating that acoustic levitation could be used forcomplex three-dimensional manipulation of levitated smallparticles, such as polystyrene balls. While these systemsperform object translation and rotational control, both thetransducer arrays and device are large and would not befeasible as a robotic manipulator attachment.Using a similar multi-transducer array, Marzo [36]presents a hand mountable device (GauntLev), that uses aplanar transducer array and phase manipulation to translateand rotate objects. Additionally, GauntLev is capable ofperforming a picking action by first trapping a particle withina standing wave and then lifting the particle by hand to ahigher vertical position. The device and particle can then bemoved away from the acoustic reflector. Due to the one-sidedarray geometry, the levitating force is low when compared toa double-sided array geometry, limiting the types of objectsthat can be manipulated. The device also requires the objectto be move by hand to achieve object picking which is notdesirable for a robot mounted manipulator.II. P
RINCIPLES OF O PERATION
A. Acoustic Levitation and Trapping Force
Acoustic levitation devices manipulate objects by modulat-ing air particles using ultrasonic waves. The ultrasonic wavescreate a spatially distributed, time average acoustic energypattern. Objects inside the acoustic field will move frompositions of high acoustic potential energy, to areas of lowacoustic potential energy. Based on the transducers used andthe voltage applied to the transducers, the acoustic pressureand air velocity distribution can be calculated. Given theacoustic pressure distribution ( p ), and air velocity distribution( v ), the acoustic potential energy or Gor’kov potential ( U )can be found by applying the expression in Eq.(1), [41], U = 2 πR (cid:32) (cid:10) p (cid:11) ρ o c o − ρ o (cid:104) v · v (cid:105) (cid:33) (1)where R is the radius of the levitated particle, ρ o is thedensity of air and c o is the speed of sound in air.The acoustic force is the gradient of the Gor’kov potential.From the Gor’kov acoustic potential scalar field, the acousticforce vector field ( F a ) can be calculated using Eq. (2). F a = −∇ U (2)The force vector convergence, Fig.2, and force body dia-grams, Fig.3, show the acoustic force dynamics experiencedby the trapped object. The movement of the object fromposition 1 to 2 is enabled by the levitator geometry whichwill be discussed in a later section. The resting position of theobject is located in mid air at a local minima of the Gor’kovpotential where gravity ( F g ) equals the upward acoustic liftforce. B. Modeling Acoustic Field
Since the picking task requires an object to be lifted off anacoustically reflective surface, both the incident and reflectedacoustic waves must be modeled to accurately represent theacoustic field and predict object behavior. The individualtransducers are modeled as sources that superimpose linearly.The rigid table surface is a boundary at which the particle
Fig. 2. Quiver plot showing the simulated acoustic force field exerted on anobject 2mm in diameter during object picking. The object is pulled upwardto the stable convergence point located 8mm above the table surface. Fig. 3. (a) A Gor’kov potential distribution incapable of lifting an objectdue to the stable node located on the rigid surface. A particle initially restingon the table top experiences a downward acoustic force, pinning it to thetable. This potential is due to a single ring of transducers above a rigidreflecting surface. (b) A Gor’kov potential distribution capable of lifting anobject off a table top. Under the influence of this potential, a particle thatis initially resting on the table top (position 1) experiences a net upwardforce. As illustrated in the force-body diagram, at position 1 the upwardforce due to the acoustic field exceeds the downward force of gravity. Oncethis potential is enabled by turning on the ultrasound field, the particlerises until it reaches position 2, a stable local minimum at which gravitybalances the upward acoustic force. This potential is due to two stackedrings of transducers above a rigid reflecting surface. .velocity normal to the table and acoustic pressure of the airmust be zero, since the air cannot penetrate the table.The rigid table surface is analogous to a ground plane inelectrostatics, located at the x-y plane in Fig. 2 and 3. Inelectrostatics, point sources in the vicinity of a ground planecan be modeled very effectively via the method of images. Inthe method of images, the ground plane is removed, but itseffect is modeled exactly by a new set of so-called “image”harges. If a real charge of amplitude q is at coordinates ( x, y, z ) , its image has charge − q and is placed at ( x, y, − z ) .By symmetry the z components of the field cancel, and thusthe image charges emulate the effect of the ground plane.The method of images can also be applied to acoustics,as in for example [42]. In our acoustic case, we introduce aset of image transducers, mirrored below the plane. Becauseof symmetry considerations similar to the electrostatic case,the particle velocity contributions of each real and mirrortransducer cancel in the normal direction at the table surface,satisfying the boundary condition of a rigid surface.The method of images allows us to efficiently compute theGor’kov potential and resulting force field while consideringthe boundary condition for the transducers in close proximityto the reflective surface. Simply neglecting the reflectivesurface provides qualitatively incorrect results. Correctlymodeling and controlling the reflections from the tablesurface is essential for being able to successfully lift objectsoff of the surface. C. Lifting Objects from Rigid Surfaces
From the models of the Gor’kov potential, the acousticfield must be shaped such that an upward vertical lift forceis generated to lift the object off of the rigid surface. Oneinitial approach would be to create a single ring of inwardfacing transducers of the same height to generate a stableGor’kov potential node in the center of the ring just above thetarget object. However, due to the reflection of the acousticfield off the rigid surface, a stable Gor’kov node would formdirectly on the target object, pinning the object down on therigid surface, rather than lifting the object up, Fig. 3a. Thisis due to the reflected acoustic field originating from theimaginary sources located below the acoustically reflectiveboundary. As a result, the Gor’kov node is placed on thereflective boundary at a height midway between the real andimaginary acoustic source heights, at a height of zero.To place a Gor’kov node above the target object, likein Fig. 3b, at least two rings of inward facing transducersstacked on top of each other are needed. Like in the sin-gle ring of transducer example, Gor’kov nodes are createdbetween the respective heights of each transducer ring,including the imaginary rings. The second transducer ringcancels the Gor’kov node located on the reflective surface.This creates a region of vertical instability at the reflectiveboundary, pushing the object into a stable Gor’kov node asdepicted in Fig. 3b.IV. S
YSTEM D ESIGN
The ultrasonic levitation manipulator is specifically de-signed to pick small, fragile objects up off of acoustically re-flective surfaces through transducer phase manipulation. Thisallows robots such as the PR2 to gain additional dexteritythrough contact-less object manipulation. To maximize trapstability and acoustic force, the trap leverages both carefulselection of the array geometry and phase manipulation ofthe transducer array. By using a cylindrical geometry ratherthan a planar geometry, acoustic energy is focused along
Fig. 4. (a) The ultrasonic transducers are arranged in a cylinder consistingof four ring levels, numbered 1 to 4 from bottom to top. (b) The four levelring geometry generates a stacked node structure with stable points locatedalong the axis of the cylinder. These stable points trap objects and canbe easily translated along the central axis of the cylinder. The acousticallyreflective boundary is shown as the black horizontal line. the cylindrical axis and concentrated into a smaller trappingvolume rather than being dispersed.
A. Acoustic Trap Geometry
To help maximize trapping force and enable dexterousmanipulation, an inward facing cylindrical array of 56 trans-ducers arranged into four rings, Fig. 4a, aims to exert a higherdegree of force and lift objects further away from the surfaceof the table than a planar transducer design. To perform thepicking action, more than one ring is need to correctly shiftthe Gor’kov potential. While only the bottom two rings ofthe manipulator are necessary for lifting an object off of areflective surface, the top two rings lift the object further offof its initial resting surface and secure the object at a higherposition. This reduces the effect of the imaginary acoustic
Fig. 5. A comparison between a planar and cylindrical trap geometrywith respect to the maximum acoustic pressure delivered to a focusedtarget location. The cylindrical geometry is capable of delivering 27% morepressure than the planar geometry.ig. 6. The ultrasonic levitation device is a material agnostic manipulator capable of suspending (a) plant pieces, (b) small insects [mosquito], (c)integrated circuits, and (d) food [potato chip crumb]. Unlike magnet based levitators, the levitated object is not restricted to specific materials. Objects staysuspended in the middle of the trap, with the force of gravity directed downward in the frame of the pictures. sources on the object, allowing the whole device to be liftedwithout affecting the object.A cylindrical geometry allows all transducers to be posi-tioned equidistantly from a central axis with pairs of trans-ducers facing each other. The transducers form stacked ringsof a diameter five times the acoustic wavelength, 8.5mm.When activated, the trap generates multiple 40kHz standingwaves which intersect to form a stacked node structure, Fig.4b, along the central axis of the cylinder. This concentratesthe acoustic energy along the central axis of the device andaids in the device’s specialized task of vertical object pickingand lifting allowing an object to be translate upward fromnode to node. In order to get a better estimate of relativegrasping pressure, Fig. 5 compares the picking node pressurebetween a 56 transducer planar array and a cylindrical array.
B. Hardware Design
To generate the acoustic field, a field programable gatearray, FPGA, generates 56 channels of individually phasecontrolled square wave outputs. Each output channel has aphase resolution of 2500 steps or approximately 0.15 ◦ . TheFPGA allows all output channels to be updated on the sameclock edge for precise transducer clock timings. The FPGAoutputs are level translated from 3.3V logic to 12V logicand a second signal with inverted phase is generated. Theoriginal and phase inverted signals drive the 56 MA40S4Stransducers differentially such that the transducers see a V pp signal. C. Trap Control System
To control the ultrasonic manipulation device, the acousticfield’s low potential nodes must be spatially translated. Thefield is changed by manipulating the phase of each transduceroutput channel through either a increment, decrement, orphase angle set command sent from a computer to the FPGA.The commands are decoded by the FPGA and the phasevalues for the next state of each channel are buffered. Oncethe buffer is full with the next set of phases, a synchronous update occurs to update the output channels at the same time.This system allows the ultrasonic manipulation device tobe manually controlled by a human, through increment anddecrement commands, or by a computer program that cancompute and send all 56 phase angles for a given position.
D. Simulating the Design
To validate the picking action of the acoustic field, a MAT-LAB simulation renders the acoustic field corresponding tothe activation of the first and second transducer ring levels.Fig. 2 shows the vertical acoustic force as an object is pickedfrom the table surface to a stable levitation node about 8mmabove the table surface. If the object is subjected to a lateralperturbation, left or right, the object will return to the stablelevitation node due to the lateral trapping force depicted bythe arrow convergence. V. R
ESULTS
The acoustic picking device is capable of picking upobjects with a maximum dimension of 2mm or less. Wetested our picking device using polystyrene balls rangingfrom 2mm to 1mm in diameter. The polystyrene balls arean example of an object which is small enough to beoccluded by the parallel jaw gripper of the PR2 robot and softenough that the PR2 will crush the balls in the gripper whentrying to pick them up. The contact-less ultrasonic levitationdevice can also levitate objects such as plants, small insects,integrated circuit chips, and potato chip crumbs, see Fig. 6.The mass of each object is given in Table. I.
A. Contact-less Ultrasonic General Purpose Picking Device
The acoustic picking device serves as a modular manipula-tor for the PR2 robot. By grasping the acoustic manipulationdevice, the PR2 is able to manipulate objects much smallerthan it could with its gripper. Additionally, the parallel jawgripper of the PR2 robot securely fits into a groove builtinto the handle of the levitation device, allowing the PR2gripper to mate with the acoustic manipulation device tooptimize ease of use. This easy to grasp design allows the
TABLE IM
ASS OF OBJECTS PICKED UP BY DEVICE
Object Flower Bud Plant Piece Mosquito IC Potato Chip Red Polystyrene BallWeight (mg) 3 3 3 12 5 1ig. 7. The picking action is performed in 6 steps. In the figure, green represents activated transducers involved in the picking process. (a) The manipulatoris placed around the target object and brought to the center of the basin of attraction. (b) The first two levels of the transducer rings are activated and anincremented phase shift is applied to the second level of transducers, bringing the object to about 10mm off the acoustically reflective surface of the table.(c) An incremental phase shift is applied to the bottom transducer array, moving the object from 10mm to 15mm. (d) The third level ring of transducersare activated and the phase of the second and third level transducers are incremented. This moves the object from 15mm to 30mm in height. (e) The firstlevel transducers are deactivated, and the fourth level transducers are activated. The phase of the third and fourth, transducer ring levels are incremented,moving the object from 30mm in height to 45mm in height. (f) The second level transducer ring is deactivated. The fourth level transducer ring phase isdecremented. At this height, 50mm, the levitation device can be lifted from the surface of the table.
PR2 to pick up the device to manipulate small objects andset down the manipulator to regain the gripper’s originalfunctionality. This system functions as a practical extensionto the manipulation capabilities of general purpose roboticgrippers.The picking procedure for the double ring levitation devicetakes advantage of both phase manipulation and selectiveactivation and deactivation. Fig. 7 shows the sequence ofring activity and phase shifts which allow the object to betranslated from the surface of the table, to the center of thetop pair of rings. To test the picking action of the device,the gripper was first fitted with the ultrasonic manipulationdevice. The picking device was then placed on the tablearound a polystyrene ball. The device is turned on and the
Fig. 8. This chart represents the acoustic manipulator’s basin of attraction ,the area that objects can be levitated or translated from to be levitated. Theacoustic manipulator has been tested to successfully perform the pickingaction on objects at test point positions represented by the green dots andpull objects from the yellow dots to the center to be levitated. The red dotsrepresent positions that were tested where objects could not be pulling intoa position capable of picking. Each grid unit is 1mm and each dot is spaced5mm apart. ring state and phase sequence in Fig. 7 is performed manuallyby a human operator, allowing the ball to be picked up. Oncetrapped in the upper ring, the acoustic levitation device canbe picked up and moved.In an ideal scenario, the robot would be capable ofpositioning the acoustic manipulation device with the objectdirectly in the center of the cylindrical transducer array,however, robots designed for larger manipulation tasks, suchas the PR2 used for testing, have a positioning error largerthan the size of the target object. The acoustic manipulationdevice compensates for positioning error by increasing the“grasping” range of the robot. For contact-less manipulationvia acoustic levitation, we call this the basin of attraction ,shown in Fig. 8. This basin refers to the area in which objectsinside the cylinder can be either picked up or pulled bythe acoustic field toward the inner area where the pickingaction via acoustic levitation is possible. The extent of thebasin of attraction is shown in green and yellow. The greenarea denotes positions that the acoustic levitation devicecan directly lift without lateral manipulation and the yellowdenotes the area where the manipulator can laterally moveobjects into the green area from. To compensate for the10mm positioning uncertainty of the PR2, the basin ofattraction is about 30mm in diameter, an area of 706.86mm ,allowing for objects to be picked up despite positioningmisalignment. B. Acoustic Manipulation Device for Object Sorting
When mounted to the PR2 robot, we tested the generalusability of the acoustic manipulator with an object sortingtask. In the test, a polystyrene ball of either red or bluewas placed on the table in front of the robot. The PR2first picked up the ball using the acoustic manipulator asdescribed above. The PR2 then brought the picker up toa camera and extracted visual features within the picker.The PR2 successfully localized the object and identifiedts’ color, then moved the picker above the correspondingcolor circle and dropped the polystyrene ball into the correctarea. This experiment suggested that (1) the weight andsize of this device does not impact the robot arm’s normalfunctionality, (2) the contact-less grasp was strong enoughto resist air turbulence generated by robot movements, (3)the contact-less grasp provides unoccluded object view forvisual inspection.
C. Limitations of the Acoustic Manipulator
Although the contact-less ultrasonic manipulator can per-form the picking action on polystyrene balls, the manipula-tion device has limitations and can be further improved toincrease robustness and predictability, allowing the processto be automated and fully controlled by the robot.Additional elements, such as a feedback loop capable oflocalizing the object in both height and lateral position withinthe trap, would improve fine tuned and coordinated objectmovement. Movement planning and ultrasonic transducerphase calculation becomes easier to automate and moreaccurate with a localization method, enabling fine tunedrotation and translation of objects within the trap.Another improvement to the system also includes objectgeometry recognition and mapping to help optimize acousticfield strength based on target object size and shape. Whilepicking objects are limited to small objects approximately2mm in size, acoustic lift force could be further maximizedto accommodate larger objects.Designed for a controlled environment, variations in envi-ronmental variables can effect the performance of the device.One example of this is heat generated by the transducers.Variations in air temperature can change the air densitywithin the trap, causing errors in device control.VI. C
ONCLUSION
This work presents an ultrasonic, contact-less manipulatordesigned to add millimeter scale object control to generalpurpose robots, like the PR2. The picking function is enabledby the acoustic field modeling technique, improving the sim-ulated force dynamics inside the trap. With these simulationswe were able to improve the accuracy of the system andinform the geometry of the device design. Likewise, thebasin of attraction helped to improve the picking functionby expanding the grasping range. Acoustic levitation andphase based object manipulation lends itself to robotics bycompensating for the robot’s positioning uncertainty andgrasping force control. To make these actions more robustand fully automatic, additional improvements to the softwareand hardware could help with environmental factor immunityand increase object control. Implementing an object localiza-tion and feedback loop would help to improve this device’sdesign and long term usage in performing more complicatedtasks. A
CKNOWLEDGMENTS
This work was funded in part by the Milton and DeliaZeutschel Professorship, and in part by National ScienceFoundation awards CNS-1823148 and EFMA-1832795. R
EFERENCES[1] R. R. Ma and A. M. Dollar, “On dexterity and dexterous manipula-tion,” in . IEEE, 2011, pp. 1–7.[2] B. G. Bolling, N. Vasilakis, H. Guzman, S. G. Widen, T. G. Wood,V. L. Popov, S. Thangamani, and R. B. Tesh, “Insect-specific virusesdetected in laboratory mosquito colonies and their potential implica-tions for experiments evaluating arbovirus vector competence,”
TheAmerican journal of tropical medicine and hygiene , vol. 92, no. 2, pp.422–428, 2015.[3] A. V. Batchelor and R. I. Wilson, “Sound localization behavior indrosophila melanogaster depends on inter-antenna vibration amplitudecomparisons,”
Journal of Experimental Biology , vol. 222, no. 3, p.jeb191213, 2019.[4] H. Gu and D. K. O’Dowd, “Whole cell recordings from brain of adultdrosophila,”
JoVE (Journal of Visualized Experiments) , no. 6, p. e248,2007.[5] J. P. Levy, R. R. Muldoon, S. Zolotukhin, and C. J. Link, “Retroviraltransfer and expression of a humanized, red-shifted green fluorescentprotein gene into human tumor cells,”
Nature biotechnology , vol. 14,no. 5, p. 610, 1996.[6] C. G. Stefanoff, R. Hassan, A. C. Gonzalez, L. A. B. Andrade, D. G.Tabak, S. Romano, and I. R. Zalcberg, “Laboratory strategies forefficient handling of paraffin-embedded tissues for molecular detec-tion of clonality in non-hodgkin lymphomas,”
Diagnostic MolecularPathology , vol. 12, no. 2, pp. 79–87, 2003.[7] F. Kong, L. Yuan, Y. F. Zheng, and W. Chen, “Automatic liquidhandling for life science: a critical review of the current state of theart,”
Journal of laboratory automation , vol. 17, no. 3, pp. 169–185,2012.[8] H. Song, D. L. Chen, and R. F. Ismagilov, “Reactions in dropletsin microfluidic channels,”
Angewandte chemie international edition ,vol. 45, no. 44, pp. 7336–7356, 2006.[9] C. Cork, J. O’Neill, J. Taylor, and T. Earnest, “Advanced beamlineautomation for biological crystallography experiments,”
Acta Crystal-lographica Section D: Biological Crystallography , vol. 62, no. 8, pp.852–858, 2006.[10] B. Miles and P. L. Lee, “Achieving reproducibility and closed-loopautomation in biological experimentation with an iot-enabled lab of thefuture,”
SLAS TECHNOLOGY: Translating Life Sciences Innovation ,vol. 23, no. 5, pp. 432–439, 2018.[11] J. Melin and S. R. Quake, “Microfluidic large-scale integration: theevolution of design rules for biological automation,”
Annu. Rev.Biophys. Biomol. Struct. , vol. 36, pp. 213–231, 2007.[12] S. K. Chollet, L. Benmayor, J.-M. Uehlinger, and J. Jacot, “Costeffective micro-system assembly automation,” in , vol. 1. IEEE,1999, pp. 359–366.[13] K.-F. Bohringer, K. Goldberg, M. Cohn, R. Howe, and A. Pisano, “Par-allel microassembly with electrostatic force fields,” in
Proceedings.1998 IEEE International Conference on Robotics and Automation(Cat. No. 98CH36146) , vol. 2. IEEE, 1998, pp. 1204–1211.[14] D. H´eriban and M. Gauthier, “Robotic micro-assembly of micropartsusing a piezogripper,” in . IEEE, 2008, pp. 4042–4047.[15] B. Yang, P. Lancaster, and J. R. Smith, “Pre-touch sensing forsequential manipulation,” in . IEEE, 2017, pp. 5088–5095.[16] L. Johannsmeier, M. Gerchow, and S. Haddadin, “A framework forrobot manipulation: Skill formalism, meta learning and adaptive con-trol,” in . IEEE, 2019, pp. 5844–5850.[17] L. U. Odhner, L. P. Jentoft, M. R. Claffee, N. Corson, Y. Tenzer,R. R. Ma, M. Buehler, R. Kohout, R. D. Howe, and A. M. Dollar,“A compliant, underactuated hand for robust manipulation,”
TheInternational Journal of Robotics Research , vol. 33, no. 5, pp. 736–752, 2014.[18] R. M. Aronson, A. Bhatia, Z. Jia, M. Guillame-Bert, D. Bourne,A. Dubrawski, and M. T. Mason, “Data-driven classification ofscrewdriving operations,” in
International Symposium on ExperimentalRobotics . Springer, 2016, pp. 244–253.[19] Z. Jia, A. Bhatia, R. M. Aronson, D. Bourne, and M. T. Mason,“A survey of automated threaded fastening,”
IEEE Transactions onutomation Science and Engineering , vol. 16, no. 1, pp. 298–310,2018.[20] X. Cheng, Z. Jia, A. Bhatia, R. M. Aronson, and M. T. Mason,“Sensor selection and stage & result classifications for automatedminiature screwdriving,” in . IEEE, 2018, pp. 6078–6085.[21] H. M. Le, T. N. Do, and S. J. Phee, “A survey on actuators-drivensurgical robots,”
Sensors and Actuators A: Physical , vol. 247, pp. 323–354, 2016.[22] M. D. Kutzer, S. M. Segreti, C. Y. Brown, M. Armand, R. H. Taylor,and S. C. Mears, “Design of a new cable-driven manipulator with alarge open lumen: Preliminary applications in the minimally-invasiveremoval of osteolysis,” in . IEEE, 2011, pp. 2913–2920.[23] R. J. Murphy, Y. Otake, K. C. Wolfe, R. H. Taylor, and M. Armand,“Effects of tools inserted through snake-like surgical manipulators,” in . IEEE, 2014, pp. 6854–6858.[24] H. Aoyama, N. Oko, T. Usuda, and N. Le Fort-Piat, “Precise mul-tiple wires driven manipulation with visual image measurement,” in . IEEE, 2014, pp. 70–73.[25] P. E. Lancaster, J. R. Smith, and S. S. Srinivasa, “Improved proximity,contact, and force sensing via optimization of elastomer-air interfacegeometry,” in . IEEE, 2019, pp. 3797–3803.[26] K. Koyama, M. Shimojo, T. Senoo, and M. Ishikawa, “High-speedhigh-precision proximity sensor for detection of tilt, distance, andcontact,”
IEEE Robotics and Automation Letters , vol. 3, no. 4, pp.3224–3231, 2018.[27] C. W. Lee, D.-S. Go, M. H. Heo, D.-W. Lee, and H. U. Yoon, “A softlinear actuator with gentle manipulation mechanism,” in . IEEE, 2019,pp. 775–778.[28] A. Gupta, C. Eppner, S. Levine, and P. Abbeel, “Learning dexterousmanipulation for a soft robotic hand from human demonstrations,” in . IEEE, 2016, pp. 3786–3793.[29] D. Rus and M. T. Tolley, “Design, fabrication and control of softrobots,”
Nature , vol. 521, no. 7553, p. 467, 2015.[30] K. C. Galloway, K. P. Becker, B. Phillips, J. Kirby, S. Licht, D. Tcher-nov, R. J. Wood, and D. F. Gruber, “Soft robotic grippers for biologicalsampling on deep reefs,”
Soft robotics , vol. 3, no. 1, pp. 23–33, 2016.[31] F. Ilievski, A. D. Mazzeo, R. F. Shepherd, X. Chen, and G. M. White-sides, “Soft robotics for chemists,”
Angewandte Chemie InternationalEdition , vol. 50, no. 8, pp. 1890–1895, 2011.[32] J. Hughes, U. Culha, F. Giardina, F. Guenther, A. Rosendo, and F. Iida,“Soft manipulators and grippers: a review,”
Frontiers in Robotics andAI , vol. 3, p. 69, 2016.[33] W. Xie, C. Cao, Y. L¨u, Z. Hong, and B. Wei, “Acoustic method forlevitation of small living animals,”
Applied Physics Letters , vol. 89,no. 21, p. 214102, 2006.[34] S. A. Seah, B. W. Drinkwater, T. Carter, R. Malkin, and S. Subrama-nian, “Correspondence: Dexterous ultrasonic levitation of millimeter-sized objects in air,”
IEEE transactions on ultrasonics, ferroelectrics,and frequency control , vol. 61, no. 7, pp. 1233–1236, 2014.[35] C. R. Courtney, B. W. Drinkwater, C. E. Demore, S. Cochran,A. Grinenko, and P. D. Wilcox, “Dexterous manipulation of micropar-ticles using bessel-function acoustic pressure fields,”
Applied PhysicsLetters , vol. 102, no. 12, p. 123508, 2013.[36] A. Marzo, “Gauntlev: A wearable to manipulate free-floating objects,”in
Proceedings of the 2016 CHI Conference on Human Factors inComputing Systems . ACM, 2016, pp. 3277–3281.[37] A. Marzo, S. A. Seah, B. W. Drinkwater, D. R. Sahoo, B. Long, andS. Subramanian, “Holographic acoustic elements for manipulation oflevitated objects,”
Nature communications , vol. 6, p. 8661, 2015.[38] T. Kozuka, K. Yasui, T. Tuziuti, A. Towata, and Y. Iida, “Noncontactacoustic manipulation in air,”
Japanese Journal of Applied Physics ,vol. 46, no. 7S, p. 4948, 2007.[39] Y. Ochiai, T. Hoshi, and J. Rekimoto, “Three-dimensional mid-airacoustic manipulation by ultrasonic phased arrays,”
PloS one , vol. 9,no. 5, p. e97590, 2014.[40] O. Youssefi and E. Diller, “Contactless robotic micromanipulation in air using a magneto-acoustic system,”
IEEE Robotics and AutomationLetters , vol. 4, no. 2, pp. 1580–1586, 2019.[41] H. Bruus, “Acoustofluidics 7: The acoustic radiation force on smallparticles,”
Lab on a Chip , vol. 12, no. 6, pp. 1014–1021, 2012.[42] G. Gaunaurd and H. H, “Acoustic scattering by a spherical body neara plane boundary,”