文档库

最新最全的文档下载
当前位置:文档库 > Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

A Pivoting Gripper for Feeding Industrial Parts

Brian Carlisle Adept Technology Ken Goldberg

USC

Anil Rao

Univ.Utrecht

Jeff Wiegley

USC

To be cost effective and highly precise,many industrial as-sembly robots have only four degrees of freedom(D.O.F.) plus a binary pneumatic gripper.Such robots commonly permit parts to be rotated only about a vertical axis.How-ever it is often necessary to reorient parts about other axes prior to assembly.In this paper we describe a way to ori-ent parts about an arbitrary axis by introducing a rotating bearing between the jaws of a simple gripper.

Based on this mechanism,we are developing a rapidly con?gurable vision-based system for feeding parts.In this system,a camera determines initialpart pose;the robot then reorients the part to achieve a desired?nal pose.We have implemented a prototype version in our laboratory using a commercially-available robot system.

1Introduction

To automate the assembly of mechanical components,parts must precisely oriented prior to packing or insertion.A parts feeder is a machine that orients parts.Currently,the design of parts feeders is a black art that is responsible for up to30%of the cost and50%of workcell failures [21,3,5,30,31].“The real problem is not part transfer but part orientation.”,Frank Riley,Bodine Corporation[27, p.316,his italics].Thus there is a demand for a parts feeder that can be reprogrammed rather than physically modi?ed when part geometry changes.

Our feeder design combines machine vision with a high-speed robot arm.The system is programmed based on the type of part to be fed.During operation,a collection of like parts are randomly scattered on a?at worktable where they are subject to the force of gravity.An overhead vision system determines the pose(position and orientation)of Adept Technology,Inc.,150Rose Orchard Way,San Jose,CA 95134.(408)432-0888.

Institute for Robotics and Intelligent Systems,Computer Sci-ence Department,University of Southern California,Los Angeles, CA90089-0273,(213)740-9080,goldberg@http://www.wendangku.net/doc/bee3908b84868762caaed59b.html.Gold-berg and Wiegley are supported by the National Science Founda-tion under Award No.IRI-9123747,and by an equipment grant from Adept Technology Inc.

Department of Computer Science,Utrecht University,P.O. Box80.089,Padualaan14,3508TB Utrecht,The Netherlands. Rao is supported by the ESPRIT Basic Research Action No.6546 (project PRoMotion).

Institute for Robotics and Intelligent Systems,Computer Sci-ence Department,University of Southern

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

California.Figure1:A parts feeder using a commercially-available system that integrates machine vision,high-speed robot arm, and pivoting gripper.This illustration shows the system feeding rectangular parts into a square pallet.

each part.The robot arm then picks up each part and moves it into a desired?nal pose as illustrated in Figure1.

To be cost effective,fast,and highly precise,commercial assembly robots usually have only four degrees of freedom (4D.O.F.).However parts must be reoriented and reposi-tioned through six degrees of freedom(6D.O.F.).To close this gap,we have designed a gripper with a rotational pivot between its jaws to provide an extra degree of freedom. Since cost and weight are critical,we note that the pivoting axis need not be actuated:it is possible to pick up each part along an axis offset from its center of mass and use on the force of gravity to rotate the part as illustrated in Figure2. In this paper we describe related work on parts feeders and robot gripper mechanisms.In Sections3and4we describe the pivoting mechanism in detail and our experiments with a vision-based parts feeding http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlst,we discuss the advantages and disadvantages of the proposed system and describes avenues for future research.

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

Brian Carlisle Adept Technology A Pivoting Gripper for Feeding Industrial Parts

Figure 2:The ?gure shows 4snapshots progressing from left to right.The gripper grasps the rectangular part along an axis offset from its center of mass,lifts it off the table,and uses the force of gravity to rotate the part into a standing orientation.

2Related Work

To feed parts,manufacturers generally rely on passive de-vices that use mechanical ?lters to reject all parts except those in a desired orientation.Rejected parts are recycled for another pass through the ?lter.The most common type of feeder is the vibratory bowl feeder ,where parts in a bowl are vibrated with a rotary motion so that they climb a helical track.As they climb,a sequence of baf?es and cutouts in the track creates a mechanical “?lter”that causes parts in all but one orientation to fall back into the bowl for another attempt at running the gauntlet [3,27,29].To improve feedrate,it is sometimes possible to design the track so as to mechan-ically rotate parts into a desired orientation (this is known as conversion ).Related methods use centrifugal forces [5],reciprocating forks,or belts to move parts through the ?lter [26].The biggest disadvantage with such methods is that when part geometry changes,the ?lters must be mechani-cally redesigned with a manual trial-and-error process.There are currently no systematic methods for designing ?lters;it is a “black art”performed by specialists through trial and error.Filters must be tested and incrementally modi?ed when they cause parts to jam.The dif?culty is that a tweak that avoids parts jamming in one orientation may cause parts to jam in another [8,39].

Sony’s APOS parts feeder [12]uses an array of nests (silhouette traps)cut into a vibrating plate.The nests and the vibratory motion are designed so that the part will remain in the nest only in a particular orientation.By tilting the plate and letting parts ?ow across it,the nests eventually ?ll up with parts in the desired orientation.Although the vibratory motion is under software control,specialized mechanical nests must be designed for each part and jamming must be avoided as with bowl feeders [19].Singer and Seering [32]proposed several designs for programmable parts feeders where programmed vibration was used to drive parts into a stable orientation.These methods can be useful for bringing parts into one of several “low-energy”poses where its center of mass is as low as possible.

In the early 1980’s,several researchers used sensors to determine the pose of parts delivered by a vibratory track [26].Sensors such as tactile probes [9,36],photocells [10],?beroptic sensors [23],and machine vision systems [11,33]were employed.Once part pose was determined,air-jets and

trapdoors were used to group parts in similar poses.Adept Technology Inc.has several commercial installations that use a 4DOF robot to reorient parts in the plane.We know of no system that combines vision with a 4DOF robot to rapidly orient parts in 6DOF.

Developing a task-based robot motion planner has been a goal in robotics since the early days [22,34,16].Paul’s HAND EYE system for solving the “instant insanity”prob-lem was the ?rst to automatically generate intermediate part poses when a transformation could not be accomplished in a single motion due to limits on reachability or arm kinemat-ics.Since then,a variety of algorithms have been developed for robot motion planning [15].In particular Tournassoud,Lozano-Perez,and Mazer [35]considered the problem of moving polyhedral parts from one pose to another using a sequence of regrasping actions with a 6DOF robot and parallel-jaw gripper.The authors took advantage of the fact that parts generally have a ?nite number of stable poses.We note that both HAND EYE and HANDEY used a 6DOF arm to reorient parts by regrasping.Quoting from [35]:

Part motions during which the object remains in contact with the hand...can only be performed by a dextrous hand as they rely on precise control of forces exerted on the object.

The authors suggest how a dextrous hand might reorient a pen by grasping it below the center of mass and relaxing the grasp force so that it rotates into a new orientation.In this paper we show that such manipulation can be achieved with-out precise force control using a mechanical modi?cation to the parallel-jaw gripper.

The text by Kato and Sadamoto [13]includes detailed photos and drawings of over 150novel mechanisms for robot end effectors.Perhaps the closest ancestor of our pivoting design is the mechanism developed by John Birk at the University of Rhode Island,which was also motivated by the need to orient parts.Birk’s hand [1]has a powered belt that causes parts to rotate about an axis parallel to the gripper jaws rather than about an axis perpendicular to the jaws as in our design.A similar pivoting joint is standard on most C-clamps available in hardware stores.We note that such clamps only have a pivot on one jaw so that the pivot is immobilized once the part is grasped.

Figure3:An exploded view of the mechanism for one pivoting pad.Frictional resistance to torsional and axial forces are reduced with needle and thrust bearings as illustrated.The contacting pad can be of?at hard rubber(in our case a pencil eraser).The pad turns with the part,thereby insuring frictional resistance to slip as the part is lifted and transported.

Goldberg and Furst developed a parallel-jaw gripper with a translational bearing rather than a rotational bearing.The translating bearing serves to reduce effective friction parallel to the support plane and thus aids in achieving a stable grasp. One similarity to the pivoting gripper is that the additional degree of freedom does not require an active servo system. Also,both grippers were designed for application to parts feeding[6].

3The Pivoting Gripper Mechanism

Our idea for rotating parts requires two“hard?nger”con-tacts,which are de?ned as contacts that can apply forces pointing into the friction cone but cannot resist torques about the contact point[28].Clearly,a grasp with two hard?nger contacts cannot achieve form closure as the part is free to ro-tate about the contact axis.Indeed,[17]showed that4hard ?nger contacts are both necessary and suf?cient to achieve form closure on a polyhedron.However for feeding parts we do not require form closure;we want to insure that the part will not translate when lifted,and in contrast to most work on grasping,we want to insure that the part will rotate about the grasp axis.

One way to achieve hard?nger contacts is to use sharp-ened points.Such point contacts are sensitive to small vari-ations in part orientation and do not have the“self-aligning”bene?ts of?at contacts.Furthermore,point contacts may damage the part.To implement true hard?nger contacts, the biggest problem is eliminating frictional resistance to torques about the pivot axis.

We propose to use a bearing mechanism as illustrated in ?gure3.

An important consideration is that the bearing mechanism have a small footprint and also permit the pivoting axis to reach as close as possible to the worksurface to grasp small parts.Machining such a mechanism poses practical dif?culties.We machined our own bearing races and used custom needle bearings to build a prototype with footprint 1322mm permitting the pivot axis to be lowered to within 6mm of the worktable.We used?at-topped pencil erasers to provide high frictional resistance to translational forces. We mounted a pair of these bearings on a commercial pneumatic parallel-jaw gripper.When the jaws are closed under air pressure,compressive forces tax our thrust bear-ings and introduce frictional resistance to rotation.Fortu-nately,we have found that it is a simple matter to achieve compliant rotation by brushing parts against a?xed“lip”in the environment.The brushing movement can be achieved with a combination of vertical and horizontal movements of the arm,thus requiring no actuation at the pivot.

4The Parts Feeding System

Our feeder design is based on a commercially-available sys-tem that integrates machine vision with a high-speed robot arm.The system is programmed based on the type of part to be fed.During operation,a collection of like parts are

dropped and randomly scattered on a?at worktable where they are subject to the force of gravity.A set of conveyor belts is used to singulate parts and transport them into the ?eld of view of an overhead camera.Note that in general, parts will only have a?nite number of stable poses under the in?uence of gravity.Each of these is treated as a unique pattern by the vision system,which identi?es each pattern and its pose(position and orientation).Each pattern indexes into a pre-programmed manipulation routine for the pivot-ing gripper which picks up the part,reorients it if necessary, and moves it into a desired?nal pose.The process is then repeated for all parts that can be identi?ed by the vision sys-tem.Remaining parts are recycled for another pass through the system.

We implemented a prototype system in our laboratory using a commercially-available robot system:Adept604-S SCARA type arm,standard CCD camera,and Adept’s AIM vision software.Figure4shows a photo of the system in our laboratory.In one case we placed registration marks on the parts to cope with ambiguous and symmetric poses(one method for locating such marks is described in[24]).We also used diffused overhead lighting to reduce shadows and specularities.

In our initial experiments,we hand-singulated the parts by dropping them one-by-one into the camera’s?eld of view. We also hand-coded the grasp points for each pattern.After the robot reorients the part and places it into a pallet,it returns to a home position and waits for the next part to be dropped into its?eld of view.We tested the system with two sets of parts:rectangular parts made of steel and

orange insulating caps made of plastic.The latter parts were suf?ciently lightweight that gravity did not provide suf?cient torque to overcome friction;in this case we brushed parts past a?xed lip to achieve the same effect.

The resulting system worked quite reliably.We achieved feed rates of5seconds per part,including visual recognition of pose.In an industrial setting,the vision and manipulation can be pipelined to speed operation.The system was able to successfully feed parts about90%of the time.Failures were typically due to errors in visual recognition,grasp location, or incomplete pivoting along the lip.A videotape of these experiments is available.

5Comparison with Existing Feeders

Our goal is to develop a programmable system that ef?-ciently feeds a broad class of industrial parts.Parts to be fed will be singulated with the aid of vibration and a series of conveyor belts traveling at different speeds;parts that are not recognized or are unable to be reoriented will be rejected and recirculated.

This design for a programmable feeder offers the follow-ing advantages over previous methods:

Fast set-up and changeover:The camera,robot arm, and gripper can be used for a wide variety of part shapes.To set-up for a new part,only the vision and manipulation routines must be changed.

Figure4:Experimental system in our laboratory.

Less damage to sensitive parts:The proposed feeder is highly suited for delicate parts such as electronic or optical components.

Compatibility with large and small parts.Vibratory bowls are generally limited to parts of diameter?ve inches or less due to restrictions on bowl diameter.

Less Likely to jam:No mechanical?lters are required.

Lower Noise.OSHA Noise limits are90dBa,maxi-mum employee exposure over eight hours.Shrouds to contain noise can add up to25%to the cost of vibratory bowls[39].

This feeder design is not intended as a universal panacea. The initial investment for the?rst feeder will be higher than that for conventional(hard-tooled)feeders.And certainly conventional feeders will be more cost effective for high-volume parts such as fasteners and washers.But our hunch is that the?exibility of the vision-based system will be cost effective in the long run for a variety of industrial applica-tions.

6Planning Algorithms

To con?gure such a system for a new part,two changes in software are required:vision and manipulation routines.In our initial experiments the former was programmed using proprietary statistical decision software from Adept,and the latter was hand-coded.We are now working to develop plan-ning algorithms for automatically programming the system based on part geometry.

Given a CAD model of a part and its center of mass, the?rst step is to determine all of the part’s stable poses under the force of gravity.For polyhedral parts,we can compute the convex hull of the part and check if the center of mass projects into each face.Clearly,a polygonal part with faces can have no more than stable poses.For more general classes of curved parts,we are working with Dave Kriegman to build on a recent algorithm described in[14]. Each stable pose presents a projective image that can be identi?ed by the vision system.Well-known methods exist for automatically generating shape recognition rou-tines(some of these have been incorporated into Adept’s commercially-available software).We are also studying al-gorithms for automatically locating registration marks to reduce ambiguities.

A primary challenge is to develop algorithms that,given a CAD model of a part,will automatically plan appropriate grasp strategies for each stable pose.Given polyhedral part shape and coef?cient of friction as input,the problem is to ?nd a(possibly empty)set of grasps that will achieve the following objectives[25]:

The line connecting these contacts(the contact axis)is parallel to the worktable(due to the limited kinematics of the robot arm),

The part will not slip as it is lifted off of the plane,

The part will rotate about the contact axis as it is lifted due to the force of gravity and come to rest such that:

When the part is replaced on the table it will assume the desired pose.

Automatic planning software would permit of?ine analy-sis of the“feedability”of a proposed part,thereby providing designers with rapid feedback[38].It may also be possible to to predict feedrates of?ine using a probabilistic analysis of pose stability,ie,system feedrate is related on the proba-bility that a part will land on each of its stable faces[20,7]. We have implemented a method similar to that presented in [2],which starts with the convex hull of the part.Each face of this hull is projected onto a unit sphere centered on the part’s center of mass.This projected face de?nes a“cap-ture range”,a set of contact orientations that will converge on enclosed face as the part falls under quasi-static condi-tions.An2implementation is described in[37].We acknowledge that more accurate prediction of pose proba-bilities would require us to model the complex dynamics of frictional collisions which is notoriously dif?cult[18].

Many variations of the above system are possible.We might also include an active clutching mechanism for lock-ing the pivot when part rotations are undesirable.This would,for example,allow the same gripper to be used for feeding and inserting parts.

The system described in this paper is an example of what might be called a“RISC”approach to industrial robotics: R educed I ntricacy in S ensing and C ontrol[4].The idea is to reduce complex manipulations to a sequence of primitive operations that can be performed with simple mechanisms such as the pivoting gripper.A challenge is to develop robust planning algorithms that will automatically perform this reduction.We believe that such algorithms can have near-term application to manufacturing. Acknowledgements

We thank Tony Lewis,Mark Dane,and Dave Kriegman for their suggestions.We also thank John Elder for the photographs and Maria Wiegley for Figure2.

References

[1]John Birk.A computer-controlled rotating-belt hand

for object orientation.IEEE Trans.on Systems,Man, and Cybernetics,4(2),March1974.

[2]G.Boothroyd,A.H.Redford,C.Poli,and L.E.Murch.

Statistical distributions of natural resting aspects of parts for automatic handling.SME Manufacturing En-gineering Transactions,1:72,1972.

[3]Geoffrey Boothroyd,Corrado Poli,and Laurence E.

Murch.Automatic Assembly.Marcel Dekker,Inc., 1982.

[4]John Canny and Ken Goldberg.“RISC”for in-

dustrial robotics:Recent results and open prob-lems.In International Conference on Robotics and Automation.IEEE,May1994.Also available as USC Techreport IRIS-93-315and via anonymous ftp from128.125.51.19under pub/Tech Rept/Iris/1993/IRIS93315.ps.Z.

[5]Gregory T.Farnum and Bill Davis.Delivering the part.

Manufacturing Engineering,March1986.

[6]Ken Goldberg.Orienting polygonal parts without sen-

sors.Algorithmica,10(2):201–225,August1993.Spe-cial Issue on Computational Robotics.

[7]Ken Goldberg and Matthew T.Mason.Bayesian grasp-

ing.In International Conference on Robotics and Au-tomation.IEEE,May1990.

[8]Jerry Goodrich and Gary Maul.Programmable parts

feeders.Industrial Engineering Magazine,May1983.

Reprinted in Automated Assembly,Jack http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlne ed., published in1986by SME.

[9]D.D.Grossman and M.W.Blasgen.Orienting me-

chanical parts by computer-controlled manipulator.

IEEE Trans.on Systems,Man,and Cybernetics,5, 1975.

[10]J.Hartley.Picking parts from a bowlfeeder with image

sensing.Sensor Review,1(1),January1981.

[11]J.W.Hill.Programmable bowl feeder design based on

computer vision.Assembly Automation,1(1),Novem-ber1980.

[12]Hajime Hitakawa.Advanced parts orientation sys-

tem has wide application.Assembly Automation,8(3), 1988.

[13]Ichiro Kato and Kuni Sadamoto.Mechanical Hands

Illustrated(Revised Edition).Hemisphere Publishing Corp,NYC,1987(originally published in1982). [14]David http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlputing stable poses of piece-

wise smooth objects.CVGIP:Image Understanding, 55(2),March1992.

[15]Jean-Claude Latombe.Robot Motion Planning.

Kluwer Academic Press,1991.

[16]Tomas Lozano-Perez,Joe Jones,Emmanuel Mazer,

and Pat O’Donnell.Handey:A Robot Task Planner.

MIT Press,1992.

[17]Xanthippi Markenscoff,Luqun Ni,and Christos H.

Papadimitriou.The geometry of grasping.IJRR,9(1), February1990.

[18]Matthew T.Mason,Ken Goldberg,and Y u Wang.

Progress in robotic manipulation.In15th Grantees Conference on Production Research and Technology.

National Science Foundation,January1989.

[19]Paul Moncevicz,Mark Jakiela,and Karl Ulrich.Orien-

tation and insertion of randomly presented parts using viratory agitation.In ASME3rd Conference on Flexi-ble Assembly Systems,September1991.

[20]L.E.Murch and G.Boothroyd.Prediction ef?ciency

of parts orienting system.Automation,2,1971. [21]James L.Nevins and Daniel http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlputer-

controlled assembly.Scienti?c American,1978. [22]Richard Paul.Modelling,Trajectory Calculation,and

Servoing of a computer Controlled Arm.PhD thesis, Stanford AIM-177,November1973.

[23]D.Pherson.Programmable feeder for non-rotational

parts.In15th CIRP International Seminar on Manu-facturing Systems,1983.

[24]Anil Rao and Ken Goldberg.Placing registration

marks.IEEE Transactions on Industrial Electronics, Special Issue on Industrial Robotics,1993.Accepted May1993.A preliminary version appeared in the1993 IEEE International Conference on Robotics and Au-tomation,Atlanta,GA,and as USC IRIS Technical Report293,April1992.

[25]Anil Rao and Ken Goldberg.Planning grasps for a

pivoting gripper.Technical report,USC Institute for Robotics and Intelligent Machines,http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlC Techreport IRIS-93-313.[26]A.H.Redford and E.Lo.Robots in Assembly.Halsted

Press,New Y ork,1986.

[27]Frank J.Riley.Assembly Automation,A Management

Handbook.Industrial Press,New Y ork,1983. [28]J.Kenneth Salisbury.Kinematic and Force Analysis

of Articulated Hands.PhD thesis,Department of Me-chanical Engineering,Stanford University,May1982.

published in it Robot Hands and the Mechanics of Manipulation,MIT Press,1985.

[29]Ben-Zion Sandler.Robotics:Designing the Mecha-

nisms for Automated Machinery.Prentice Hall,1991.

Chapter7summarizes results originally published(in Russian)in A.N.Rabinovitsh,Automatic Orientation and Parts Feeding,Technika,1968.

[30]Berhard J.Schroer.Electronic parts presentation using

vibratory bowl feeders.Robotics,3,1987.

[31]Neil C.Singer.Utilizing dynamic and static stability

to orient parts.Master’s thesis,MIT,1985.

[32]Neil C.Singer and Warren Seering.Utilizing dynamic

stability to orient parts.Journal of Applied Mechanics, 54,1987.See also Singer’s1985MIT Masters Thesis.

[33]T.Suzuki.An approach to a?exible part-feeding sys-

tem.In1st International Conference on Assembly Au-tomation,1980.

[34]Russ H.Taylor.A Synthesis of Manipulator Control

Programs from Task-Level Speci?cation.PhD thesis, Stanford AIM-282,july1976.

[35]Pierre Tournassoud,T.Lozano-Perez,and E.Mazer.

Regrasping.In International Conference on Robotics and Automation.IEEE,May1987.

[36]S.S.M.Wang and Peter Will.Sensors for computer

controlled mechanical assembly.Industrial Robot, March1978.

[37]Jeff Wiegley,Anil Rao,and Ken http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlput-

ing a statistical distribution of stable poses for a poly-hedron.In The30th Annual Allerton Conference on Communication,Control,and Computing.University of Illinois at Urbana-Champaign,October1992. [38]Randy Wilson and Jean-Claude Latombe.Reason-

ing about mechanical assembly.In The30th Annual Allerton Conference on Communication,Control,and Computing.October1992.

[39]Richard D.Zimmerman.Updated art–parts feed-

ing.In Assemblex VIII Conference.SME,March1982.

Reprinted in Automated Assembly,Jack http://www.wendangku.net/doc/bee3908b84868762caaed59b.htmlne ed., published in1986by SME.