Collective Manipulation and Construction -...

Chapter 1

CollectiveManipulation andConstruction

1.1 Summary

Many practical applications could make use of robot col-lectives that can manipulate objects and construct struc-tures. Examples include applications in warehousing,truck loading and unloading, transporting large objectsin industrial environments, and assembly of large-scalestructures. Creating such systems, however, can be chal-lenging. When collective robots work together to ma-nipulate physical objects in the environment, their inter-actions necessarily become more tightly coupled. Thisneed for tight coupling can lead to important controlchallenges, since actions by some robots can directlyinterfere with those of other robots. This chapter ex-plores techniques that have been developed to enablerobot swarms to effectively manipulate and construct ob-jects in the environment. The focus in this chapter is ondecentralized manipulation and construction techniquesthat would likely scale to large robot swarms (at least10 robots), rather than approaches aimed primarily atsmaller teams that attempt the same objectives. Thischapter first discusses the swarm task of object trans-portation; in this domain, the objective is for robots tocollectively move objects through the environment to agoal destination. The chapter then discusses object clus-

1

2CHAPTER 1. COLLECTIVEMANIPULATION AND CONSTRUCTION

tering, which requires objects in the environment to beaggregated at one or more locations in the environment.This is followed by the related task of object sorting,which aims to have objects collected based on type. Thefinal task discussed is that of collective construction andwall building, in which robots work together to build apre-specifed structure. While these different tasks varyin their specific objectives for collective manipulation,they also have several commonalities. This chapter ex-plores the state of the art in this area.

1.2 Object Transportation

Some of the earliest work in swarm robotics was aimedat the object transportation task [1, 2, 3, 4, 5, 6], whichrequires a swarm of robots to move an object from itscurrent position in the environment to some goal desti-nation. The primary benefit of using collective robotsfor this task is that the individual robots can combineforces to move objects that are too heavy for individualrobots working alone or in small teams. However, thetask is not without its challenges; it is non-trivial to de-sign decentralized robot control algorithms that can ef-fectively coordinate robot team members during objecttransportation. A further complication is that the in-teraction dynamics of the robots with the object can besensitive to certain object geometries [7, 8] and object ro-tations during transportation [8], thus exacerbating thecontrol problem.

There are many ways to compare and contrast alter-native distributed techniques to collective object trans-port. The most common distinctions are:

• Local knowledge only, vs. some global knowledgerequired (e.g., of team size, state, position)

• Homogeneous swarms vs. heterogeneous swarms (e.g.,teams with leaders and followers)

• Manual controller design vs. autonomously learnedcontrol

• Two (2) dimensional vs. three (3) dimensional en-vironments

• Obstacle-free environments vs. cluttered environ-ments

1.2. OBJECT TRANSPORTATION 3

• Static environments vs. dynamic environments

• Dependent on fully-functioning robots vs. systemsrobust to error

Alternatively, we can compare transportation tech-niques by focusing on the specific manipulation techniqueemployed. The manipulation techniques used for collec-tive object transportation can be grouped into three pri-mary methods [9]: pushing, grasping, and caging. Thepushing approach requires contact between each robotand the object, in order to impart force in the goal di-rection; however, the robots are not physically connectedwith the object. In the grasping approach, each robot inthe swarm physically attaches to the object being trans-ported. Finally, the caging approach involves robots en-circling the object so that the object moves in the de-sired direction, even without the constant contact of allthe robots with the object.

This section outlines some of the key techniques de-veloped to address this object transportation task, orga-nized according to these three main techniques.

1.2.1 Transport by Pushing

A canonical task often used as a testbed in distributedrobotics is the box pushing task. The number, size, orweight of the boxes can be varied to require differenttypes of multi-robot cooperation. This task typicallyinvolves robots first locating the box, positioning them-selves at the box, and then moving the box cooperativelytoward a goal position. Typically, this task is explored intwo dimensions. The domain of box pushing is also pop-ular because it has relevance to several real-world appli-cations [10], including warehouse stocking, truck loadingand unloading, transporting large objects in industrialenvironments, and assembly of large-scale structures.

The pushing technique was first demonstrated in theearly work of Kube and Zhang [1], inspired by the co-operative transport behavior in ants [7]. They proposeda behavior-based approach that combined behaviors forseeking out the object (illuminated by a light), avoidingcollisions, following other robots, and motion control. Anadditional behavior to detect stagnation was used to en-sure that the collective did not work consistently againsteach other. In this approach, all robots acted similarly;there was no concept of a leader and followers. While


Figure 1.1: Demonstration of five robots collectivelypushing a lighted box. (Reproduced from [7].)

some of the robots in the swarm might not contributeto the pushing task due to poor alignment or position-ing along the non-dominant pushing direction, Kube andZhang showed that careful design of these behaviors en-abled the robot swarm to distribute along the boundaryof the object and push it. Figure 1.1 shows five (5) robotscooperatively pushing a lighted box.

Other researchers have explored different aspects ofbox pushing in multi-robot systems. While much ofthis early work involved demonstrations of smaller robotteams, many of these techniques could theoretically scaleto larger numbers of robots. Task allocation and ac-tion selection are often demonstrated using collective boxpushing experiments; examples of this work include thatof Parker [11, 12], who illustrates aspects of adaptivetask allocation and learning; Gerkey and Mataric [13],who present a publish/subscribe dynamic task allocationmethod, demonstrated in cooperative pushing tasks; andYamada and Saito [14], who develop a behavior-based ac-tion selection technique that does not require any com-munication.

Other work using box pushing as an implementationdomain for multi-robot studies includes Donald, et al.[15], who illustrates concepts of information invariance


and the interchangeability of sensing, communication,and control; Simmons, et al. [16], who demonstrate thefeasibility of cooperative control for building planetaryhabitats; Brown and Jennings [17], and Bohringer, etal. [18], who explore notions of strong cooperation with-out communication in pusher/steerer models; Rus, et al.[19], who study different cooperative manipulation pro-tocols in robot teams that make use of different combi-nations of state, sensing, and communication; and Jonesand Mataric [20], who develop general methods for au-tomatically synthesizing controllers for multi-robot sys-tems, demonstrated in a box pushing domain.

Most of existing work in box pushing has focused noton box pushing as the end objective, but rather has usedbox pushing for demonstrating a variety of techniquesfor multi-robot control. However, for studies whose pri-mary objective is to generate robust cooperative trans-port techniques, work has more commonly focused onmanipulation techniques involving grasping and caging,rather than pushing, since grasping and caging providemore controllability by the robot team. The followingsections focus on techniques for grasping and caging.

1.2.2 Transport by Grasping

Grasping approaches for object transportation in swarmrobotics typically make use of form closure and force clo-sure properties [21]. In form closure, the object motion isconstrained via frictionless contact constraints; in forceclosure, frictional contact forces exerted by the robotsprevent unwanted motions of the manipulated object.The earliest work representing the grasping technique isthat of Wang, et al. [4]. This approach uses form clo-sure, along with a behavior-based control approach thatis similar to the early swarm robot pushing techniqueof Kube and Zhang [1]. The technique of Wang, et al.,called BeRoSH (for Behavior-based Multiple Robot Sys-tem with Host for Object Manipulation), incorporatesbehaviors for pushing, maintaining contact, moving, andavoiding objects. In this approach, the goal pose of theobject is provided directly to each robot from an exter-nal source (i.e., the Host); otherwise, the robots workindependently according to their designed behaviors. Asa collective, the swarm exhibits form closure. Wang, etal., show that this form closure technique can success-fully transport an object to its desired goal pose from a


Figure 1.2: Cooperative transport of a pallet using tank-like robots. (Reproduced from [2].)

variety of different starting locations.

Another early work using the force closure graspingtechnique is that of Stilwell and Bay [2], and of Johnsonand Bay [3]. They developed distributed leader-followertechniques that enable swarms of tank-like robots totransport pallets collectively while maintaining a levelheight of the pallet during transportation (see Fig-ure 1.2). In their approaches, a pallet sits atop severaltank-like robots; the weight of the pallet creates a cou-pling with the robots that could be viewed similar to agrasp. To transport the pallet, one vehicle is designatedas the leader. This leader then perturbs the dynamics ofthe system to move the swarm in the desired direction,and with the desired pallet height. The remaining robotsin the swarm react to the perturbations to stabilize theforces in the system. The system is fully distributed,and requires robots to only use local force informationto achieve the collective motion. The individual robotsdo not require knowledge of the pallet mass or inertia,the size of the collective, or the robot positions relativeto the pallet’s center of gravity. They showed the con-trol stability of their approach for this application, evenin the presence of inaccurate sensor data.

A related approach is that of Kosuge and Oosumi [5],who also use a decentralized leader-follower approach formultiple holonomic robots grasping and moving an ob-


ject, in a manner similar to that of [2]. Their approachdefines a compliant motion control algorithm for eachvelocity-controlled robot. The main difference of thiswork compared to [2] is that the control algorithm speci-fies the desired internal force as part of the coordinationalgorithm. This approach was validated in simulation forrobots carrying an aluminum steel pipe.

Another related approach is that of Miyata, et al.[6], who addressed the need for non-holonomic vehiclesto regrasp the object during transport. Their approachis a hybrid system that makes use of both centralizedand decentralized planners. The centralized planner de-velops an approximate motion plan for the object, alongwith a regrasping plan at low resolution; the decentral-ized planner precisely estimates object motion and robotcontrol at a much higher resolution. They demonstratedthe effectiveness of this approach in simulation.

Sugar and Kumar [22] developed distributed controlalgorithms enabling robots with manipulators to graspand cooperatively transport a box. In this work, a novelmanipulator design enables the locomotion control tobe decoupled from the manipulation control. Only asmall number of the team members need to be equippedwith actively controlled end effectors. This approach wasshown to be robust to positioning errors related to themisalignment between the two platforms and errors inthe measurement of the box size.

Cooperative stick pulling [23, 24] was explored byIjspeert, et al.; this task requires robots to pull sticks outof the ground (see Figure 1.3). The robot controllers arebehavior-based, and include actions such as looking forsticks, detecting sticks, gripping sticks, obstacle avoid-ance, and stick release. Experiments show that the dy-namics are dependent on the ratio between the numberof robots and the number of sticks; that collaborationcan increase super-linearly with certain team sizes; thatheterogeneity in the robots can increase the collabora-tion rate in certain circumstances; and that a simplesignalling scheme can increase the effectiveness of thecollaboration for certain team sizes. A main objective ofthis research was to explore the effectiveness of variousmodeling techniques for group behavior. These model-ing techniques are discussed in more detail in a separatechapter.

The SWARM-BOTS project is a more recent exam-ple of the use of grasping for collective transport, but


Figure 1.3: Stick pulling experiment using robot collec-tives. (Reproduced from [23].)

Figure 1.4: An s-bot, developed as part of the SWARM-BOTS project. (Reproduced from [25].)

which also makes use of self-assembly as a novel approachfor achieving distributed transport. In this work [25], s-bot robots are developed that have grippers enabling therobots to create physical links with other s-bots or ob-jects, thus creating assemblies of robots. These assem-blies can then work together for navigation across roughterrain, or to collectively transport objects. The s-botsare cylindrical, with a flexible arm and toothed gripperthat can connect one s-bot to another (see Figure 1.4).

The decentralized control of the SWARM-BOT robotsis learned using evolutionary techniques in simulation,then ported to the physical robots. The learned s-botcontrol [26] consists of an assembly module, which is aneural network that controls the robot prior to connec-tion, and a transport module, which is a neural networkthat enables the s-bot to move the object towards thegoal after a grasp connection is made. The self-assembly


process involves the use of a red-colored seed object, towhich other s-bots are attracted. S-bots initially lightthemselves with a blue ring, and then are attracted tothe color red, while being repulsed by the color blue.Once robots make a connection, they color themselvesred.

The interaction of these attractive and repulsive forcesacross the s-bots enables the robots to self-assemble intovarious connection patterns. Once the s-bots have self-assembled, they use the transport module to align to-wards a light source, which indicates the target position.The s-bots then apply pushing and pulling motions totransport the object to the destination. Similar to theapproach of Kube and Zhang [1], the s-bots also checkfor stagnation and execute a recovery move when needed.The authors demonstrate [8] how the evolutionary learn-ing approach allows the collective to successfully dealwith different object geometries, adapt to changes in tar-get location, and scale to larger team sizes.

This technique for collective transport using self-assembly was demonstrated [25] in an interesting ap-plication of object transport, in which 20 s-bots self-assembled into four (4) chains in order to pull a childacross the floor (see Figure 1.5). In this experiment, theuser specifies the number of assembled chains, the distri-bution of the s-bots into the chains, the global localiza-tion of the child, and the global action timing. The s-botsthen autonomously form the chains using self-assemblyand execute the pull. A video of this experiment is avail-able at http://www.swarm-bots.org/ram05.html.

Several additional interesting phenomena regardingcollective transport were discovered in related studieswith the SWARM-BOTS. Nouyan, et al., [27] showedthat the different collective tasks of path formation, self-assembly, and group transport can be solved in a sin-gle system using a homogeneous robot team. They fur-ther introduce the notion of chains with cycle directionalpatterns, which facilitate swarm exploration in unknownenvironments, and assist in establishing paths betweenthe object and goal. The paths established by the robot-generated chains mimic pheromone trails in ants. In [28],Gross and Dorigo determined that, while robots that be-have as if they are solitary robots can still collectivelymove objects, robots that learn transport behaviors in agroup can achieve a better performance. In [29], Campo,et al., showed that the SWARM-BOTS robots could ef-


Figure 1.5: SWARM-BOTS experiment in which s-botsself-assemble to pull a child across the floor. (Repro-duced from [25].)

fectively transport objects even with only partial knowl-edge of the direction of the goal. They investigated four(4) alternatives control strategies, which vary in the de-gree to which the robots negotiate regarding the goal po-sition during transport. Their results showed that nego-tiating throughout object transport can improve motioncoordination. All of these works are based on inspirationfrom biological systems.

The work of Berman, et al., [30] is not only bio-inspired, but also seeks to directly model the group re-trieval behavior in ants. Their studies examined the ants’roles during transport in order to define rules that governthe ants’ actions. They further explored measurementsof individual forces used by the ants to guide food totheir nest. They found that the distributed ant trans-port behavior exhibits an initial disordered phase, whichthen transitions to a more highly coordinated phase of in-creased load speed. From these studies, a computationaldynamic model of the ant behavior was designed andimplemented in simulations, showing that the derivedmodel matches ant behavior. Ultimately, this approach


Figure 1.6: Illustration of unmanned tugboats au-tonomously transporting a barge. (Reproduced from[31].)

could be adapted for use on physical robot teams.

Once a robot collective has begun transporting an ob-ject, the question arises as to how new robots can join thegroup to help with the transport task. Esposito [31] ad-dresses this challenge by adaping a grasp quality functionfrom the multi-fingered hand literature. This approachassumes that robots know the object geometry, the totalnumber of robots in the swarm, and the actuator limita-tion. Individual robot contact configurations are definedrelative to the object center and object boundary. Theobjective is to find an optimal position for a new robotby optimizing across the grasping wrench space. A nu-merical algorithm was developed to address this prob-lem, which incorporates the force closure criteria. Thisapproach was demonstrated on unmanned tugboats col-lectively moving a barge, as illustrated in Figure 1.6. Inthis demonstration, the robots are equipped with artic-ulated magnetic attachments that allow them to graspthe barge. This approach is scalable to larger numbersof robots, with constant best case runtime, and medianruntimes polynomial in the number of robots.

1.2.3 Transport by Caging

The caging approach simplifies the object manipulationtask, compared to the grasping approach, by makinguse of the concept of object closure [32]. In object clo-sure, a bounded movable area is defined for the object by


the robots surrounding it. The benefit of this approachis that continuous contact between the object and therobots is not needed, thus making for simpler motionplanning and control techniques, compared to graspingtechniques based on form or force closure. Wang and Ku-mar [33] develop this object closure technique under theassumptions that the robots are circular and holonomic,the object is star-shaped, the robots know the number ofrobots in the collective, and can estimate the geometricproperties of the object, along with the distance and ori-entation to other robots and the object. Their approachcauses the robots to first approach the object indepen-dently, and then search for an “inescapable formation”,which is a configuration of the robots from which theobject cannot escape. Finally, the robots execute a for-mation control strategy to guide the object to the goaldestination. The object approach technique is based onpotential fields, in which force vectors attract the robottoward the object and generally away from other robots.Song and Kumar proved [34] the stability of this poten-tial field approach for collective transport. Robots searchfor proper configurations around the object by represent-ing the problem as a path finding problem in configura-tion space. This work describes a necessary and sufficientcondition for testing for object closure. Later work [35]presents a fast algorithm to test for object closure. Ex-periments with 20 robots validate the proposed approach(see Figure 1.7).

A further enhancement of this vector-based controlstrategy was developed in [36], which can account forinter-robot collisions. This latter strategy implementsthree (3) primary behaviors – approach, surround, andtransport. In this variant of the work, robots converge toa smooth boundary using control-theoretic techniques.This work was implemented on a collective of physicalrobots, as illustrated in Figure 1.8.

1.3 Object Sorting and Cluster-ing

Collective object sorting and clustering requires robotteams to sort objects from multiple classes, typically intoseparate physical clusters. There are different types ofrelated tasks in this domain [37], including clustering,

1.3. OBJECT SORTING AND CLUSTERING 13

Figure 1.7: Illustration in simulation of object closure bytwenty robots. (Reproduced from [33].)

Figure 1.8: Demonstration of the use of vector fields forcollective transport via caging. (Reproduced from [36].)


segregation, patch sorting, and annular sorting. Earlydiscussions of this task in robot swarms were given byDeneubourg, et al., [38], with the ideas inspired bysimilar behaviors in ant colonies. The objective is toachieve clustering and sorting behaviors without anyneed for hierarchical decision making, inter-robot com-munication, or global representations of the environ-ment. Deneubourg, et al., showed that stigmergy couldbe used to cluster scattered objects of a single type, or tosort objects of two different types. To achieve the sort-ing behavior, the robots sensed the local densities of theobjects, as well as the type of object they were carrying.Clustering resulted from a similar mechanism operatingon a single type of object. Beckers, et al [39], achievedclustering from even simpler robots and behaviors, viastigmergic threshold mechanisms.

Holland and Melhuish [37] explore the effect of stig-mergy and self-organization in swarms of homogeneousphysical robots. The robots are programmed with simplerule sets with no ability for spatial orientation or mem-ory. The experiments show the ability of the robots toachieve effective sorting and clustering, as illustrated inFigure 1.9. In this work, a variety of influences were ex-plored, including boundary effects and distance betweenobjects when deposited. The authors conclude that theeffectiveness of the developed sorting behaviors is criti-cally dependent on the exploitation of real-world physics.An implication of this finding is that simulators must beused with care when exploring these behaviors.

Wang and Zhang [40, 41] explore similar aims, butfocus on discovering a general approach to the sortingproblem. They conjecture that the outcome of the sort-ing task is dependent primarily on the capabilities of therobots, rather than the initial configuration. This conjec-ture is validated in simulation experiments, as illustratedin Figure 1.10.

Other work on this topic includes that of Yang andKamel [42], who present research using three colonies ofants having different speed models. The approach is atwo-step process. The first step is for clusterings to bevisually formed on the plane by agents walking, pickingup, or setting down objects according to a probabilisticmodel, which is based on Deneubourg, et al, [38]. Thesecond step is for clusters to be combined using a hyper-graph model. Experiments were conducted in simulationto show the viability of the approach. The authors also

1.3. OBJECT SORTING AND CLUSTERING 15

Figure 1.9: Results of physical robot experiments in sort-ing. The first image shows the starting configuration,while the second shows the sorting results after 1.75hours. (Reproduced from [37].)


Figure 1.10: Results of simulations of sorting tasks, with8 robots and 40 objects of two types. (Reproduced from[40].)

1.4. COLLECTIVE CONSTRUCTION ANDWALL BUILDING17

discovered that having too many agents can lead to adeterioration in the swarm performance.

Martinoli and Mondada [43] implement another bio-inspired approach to clustering, in which the robot be-havior is similar to that of a Braitenberg vehicle. Theyalso discovered that large numbers of robots can causeinterference in this task, concluding that non-cooperativetask cannot always be improved with more robots.

1.4 Collective Construction andWallBuilding

The objective of the collective construction and wallbuilding task is for robots to build structures of a speci-fied form, in either 2D or 3D. This task is distinguishedfrom self-reconfigurable robots, whose bodies themselvesserve as the dynamic structure. This section is focusedon the former situation, in which manipulation is re-quired to create the desired structure. The argumentin favor of this separation of mobility and structure isthat, once formed, the structure does not need to moveagain, and thus the ability to move could serve as a li-ability [44]. Further, robotic units that serve both asmobility and structure might not be effective as passivestructural elements.

Werfel and Nagpal have extensively explored thistopic, developing distributed algorithms that enable sim-plified robots to build structures based on providedblueprints, both in 2D [45, 46, 47] and in 3D [44]. Intheir 3D approach, the system consists of idealized mo-bile robots that perform the construction, and smartblocks that serve as the passive structure. The robots’job is to provide the mobility, while the blocks’ role is toidentify places in the growing structure at which an ad-ditional block can be placed that is on the path towardobtaining the desired final structure. The goal of theirwork is to be able to deploy some number of robots andfree blocks into a construction zone, along with a singleblock that serves as a seed for the structure, and thenhave the construction to proceed autonomously accord-ing to the provided blueprint of the desired structure.

Several simplifying assumptions are made in this work[44], such as the environment being weightless and therobots being free to move in any direction in 3D, includ-


Figure 1.11: Experiments for a variety of 3D structures,built autonomously by a system of simple robots andblocks. (Reproduced from [44].)

ing along the surface of the structure under construction.This work does not address physical robot navigationand locomotion challenges, grasping challenges, etc.

In this approach, blocks are smart cubes; they cancommunicate with attached neighbors, they share a globalcoordinate system, and they can communicate with pass-ing robots regarding the allowability of block attach-ments to exposed faces. Once robots have transported afree block to the structure, they locate attachment pointsin one of three ways: random movement, systematicsearch, or gradient following. A significant contributionof this work is the development of the block algorithmthat enables the blocks to specify how to grow the devel-oping structure with guarantees, and with only limitedrequired communication. More specifically, the commu-nication requirement between blocks scales linearly inthe size of the structure, while no explicit communica-tion between the mobile robots is needed.

Experiments using this approach have shown the abil-ity of the system to build a variety of structures in simu-lation, as illustrated in Figure 1.11. A proof-of-principlephysical robot experiment using a single robot in the 2Dcase [46] is illustrated in Figure 1.12.

Werfel [48] also describes a system for arranging in-ert blocks into arbitrary shapes. The input to the robotsystem is a high-level geometric program, which is thentranslated by the robots into an appropriate arrange-ment of blocks using their programmed behaviors. Thedesired structure is communicated to the robots as a listof corners, the angles between corners, and whether theconnection between corners is to be straight or curved.Robots are provided with behaviors such as clearing,doneClearing , beCorner, collect, seal, and off. Figure 1.13shows some example structures built using this systemin simulation.


Figure 1.12: Proof-of-principle experiments for 2D con-struction, using a single robot. (Reproduced from [46].)

Figure 1.13: Geometric structures built by a team of 30robots, in simulation. (Reproduced from [48].)


Figure 1.14: Experiments with prototype hardware de-signed for multi-robot construction tasks. (Reproducedfrom [49].)

Hardware challenges of collective robot constructionare addressed by Terada and Murata [49]. In this work,a hardware design is proposed that defines passive build-ing blocks, along with an assembler robot that constructsstructures with the robots. Figure 1.14 shows the proto-type hardware completing an assembly task. In princi-ple, multiple assembler robots could work together tocreate larger construction teams more closely alignedwith the concept of swarm construction.

Other related work on the topic of collective construc-tion includes the work of Wawerla, et al., [50], in whichrobots use a behavior-based approach to build a linearwall using blocks equipped with either positive or neg-ative Velcro, distinguished by block color. Their resultsshow that adding 1 bit of state information to commu-nicate the color of the last attached block provides asignificant improvement in the collective performance.The work by Stewart and Russell [51, 52] proposes adistributed approach to building a loose wall structurewith a robot swarm. The approach makes use of a spatio-temporal varying light-field template, which is generatedby an organizer robot to help direct the actions of thebuilder robots. Builder robots deposit objects in loca-tions indicated by the template. Figure 1.15 shows theresults from one of the experiments using this approachon physical robots.

Another type of construction is called blind bulldoz-ing, which is inspired by a behavior observed in certain


Figure 1.15: Experimental trial demonstrating a swarmbuilding a loose wall via a spatio-temporal varying tem-plate. (Reproduced from [52].)

Figure 1.16: Experiments of blind bulldozing for siteclearing using physical robots. (Reproduced from [55].)

ant colonies. Rather than constructing by accumulatingmaterials, this approach achieves construction by remov-ing materials. This task has practical application in siteclearing, such as would be needed for planetary explo-ration [53]. Early ideas of this concept were discussedby Brooks, et al., in [54], which argues for large numbersof small robots to be delivered to the lunar surface forsite preparation. Parker, et al. [55], further develop thisidea by proposing robots using force sensors to clear anarea by pushing material to the edges of the work site.In this approach, the robot system collective behavioris modeled in terms of how the nest grows over time.Stigmergy is used to control the construction process, inthat the work achieved by each robot affects the otherrobots’ behaviors through the environment. Figure 1.16shows some experiments using this approach on physi-cal robots. The authors argue that blind bulldozing isappropriate in applications where the cost, complexity,and reliability of the robots is a concern.

Bibliography

[1] C.R. Kube, H. Zhang: Collective robotics: From so-cial insects to robots, Adaptive Behavior 2(2), 189 –218 (1993)

[2] D. Stilwell, J. Bay: Toward the Development of aMaterial Transport System using Swarms of Ant-like Robots, IEEE International Conference onRobotics and Automation, Atlanta, GA 1993, 1993)766 – 771

[3] P.J. Johnson, J.S. Bay: Distributed control of sim-ulated autonomous mobile robot collectives in pay-load transportation, Autonomous Robots 2(1), 43 –63 (1995)

[4] Z. Wang, E. Nakano, T. Matsukawa: Realiz-ing cooperative object manipulation using multi-ple behaviour-based robots, Proc. of the IEEE/RSJInternational Conference on Intelligent Robots andSystems (IROS), 1996) 310 – 317

[5] K. Kosuge, T. Oosumi: Decentralized controlof multiple robots handling an object, Proc. ofIEEE/RSJ International Conference on IntelligentRobots and Systems (IROS) 1996, 318 – 323

[6] N. Miyata, J. Ota, Y. Aiyama, J. Sasaki, T. Arai:Cooperative transport system with regrasping car-like mobile robots, Proc. of IEEE/RSJ Interna-tional Conference on Intelligent Robots and Systems(IROS) 1997, 1754 – 1761

[7] C.R. Kube, E. Bonabeau: Cooperative transport byants and robots, Robotics and autonomous systems30(1), 85 – 101 (2000)

23

24 BIBLIOGRAPHY

[8] R. Groß, M. Dorigo: Towards group transport byswarms of robots, International Journal of Bio-Inspired Computation 1(1), 1 – 13 (2009)

[9] Y. Mohan, SG Ponnambalam: An extensive reviewof research in swarm robotics, World Congress onNature and Biologically Inspired Computing 2009(IEEE, 2009) 140 – 145

[10] D. Nardi A. Farinelli, L. Iocchi: Multirobot sys-tems: a classification focused on coordination, IEEETransactions on Systems, Man, and Cybernetics,Part B 34(5), 2015 – 2028 (2004)

[11] L. E. Parker: ALLIANCE: An Architecture forFault Tolerant, Cooperative Control of Heteroge-neous Mobile Robots, Proc. of the IEEE/RSJ/GIInt’l Conf. on Intelligent Robots and Systems, Mu-nich, Germany 1994 (IEEE, 1994) 776 – 783

[12] L. E. Parker: Lifelong Adaptation in HeterogeneousTeams: Response to Continual Variation in Individ-ual Robot Performance, Autonomous Robots 8(3)(July 2000)

[13] B. P. Gerkey, M. J. Mataric: Sold! Auction methodsfor multi-robot coordination, IEEE Transactions onRobotics and Automation 18(5), 758 – 768 (2002)

[14] S. Yamada, J. Saito: Adaptive action selectionwithout explicit communication for multirobot box-pushing, Systems, Man, and Cybernetics, Part C:Applications and Reviews, IEEE Transactions on31(3), 398 – 404 (aug 2001)

[15] B. Donald, J. Jennings, D. Rus: Analyzing Teamsof Cooperating Mobile Robots, IEEE Interna-tional Conference on Robotics and Automation 1994(IEEE, 1994) 1896 – 1903

[16] R. Simmons, S. Singh, D. Hershberger, J. Ramos,T. Smith: First results in the coordination of het-erogeneous robots for large-scale assembly, ISERSeventh International Symposium on Experimen-tal Robotics, Honolulu 2000 (Springer-Verlag, NewYork 2000)

[17] Russell G. Brown, James S. Jennings: APusher/Steerer Model for Strongly Cooperative Mo-bile Robot Manipulation, Proceedings of 1995 IEEE

BIBLIOGRAPHY 25

International Conference on Intelligent Robots andSystems (IROS ’95), Pittsburgh 1995 (IEEE, 1995)562 – 568

[18] K. Bohringer, R. Brown, B. Donald, J. Jennings,D. Rus: Distributed robotic manipulation: Exper-iments in minimalism, Experimental Robotics IV,11 – 25 (1997)

[19] D. Rus, B. Donald, J. Jennings: Moving Furniturewith Teams of Autonomous Robots, Proceedings ofIEEE/RSJ International Conference on IntelligentRobots and Systems, Pittsburgh 1995, 1995) 235 –242

[20] C. Jones, M. J. Mataric: Automatic Synthesisof Communication-Based Coordinated Multi-RobotSystems, Proceedings of IEEE/RJS InternationalConference in Intelligence Robots and Systems,Sendai, Japan 2004 (IEEE, 2004) 381 – 387

[21] Antonio Bicchi: On the closure properties of roboticgrasping, The International Journal of Robotics Re-search 14(4), 319 – 334 (1995)

[22] T.G. Sugar, V. Kumar: Control of cooperating mo-bile manipulators, Robotics and Automation, IEEETransactions on 18(1), 94 – 103 (2002)

[23] A.J. Ijspeert, A. Martinoli, A. Billard, L.M. Gam-bardella: Collaboration through the exploitation oflocal interactions in autonomous collective robotics:The stick pulling experiment, Autonomous Robots11(2), 149 – 171 (2001)

[24] A. Martinoli, K. Easton, W. Agassounon: Model-ing swarm robotic systems: A case study in col-laborative distributed manipulation, The Interna-tional Journal of Robotics Research 23(4-5), 415 –436 (2004)

[25] F. Mondada, L.M. Gambardella, D. Floreano,S. Nolfi, J.L. Deneuborg, M. Dorigo: The cooper-ation of swarm-bots: Physical interactions in collec-tive robotics, IEEE Robotics and Automation Mag-azine 12(2), 21 – 28 (2005)

[26] R. Groß, E. Tuci, M. Dorigo, M. Bonani, F. Mon-dada: Object transport by modular robots that

26 BIBLIOGRAPHY

self-assemble, IEEE International Conference onRobotics and Automation 2006 (IEEE, 2006) 2558 –2564

[27] S. Nouyan, R. Groß, M. Bonani, F. Mondada,M. Dorigo: Group transport along a robot chain ina self-organised robot colony, Proc. of the 9th int.conf. on intelligent autonomous systems, 433 – 442(2006)

[28] R. Groß, M. Dorigo: Evolution of solitary and grouptransport behaviors for autonomous robots capableof self-assembling, Adaptive Behavior 16(5), 285 –305 (2008)

[29] A. Campo, S. Nouyan, M. Birattari, R. Groß,M. Dorigo: Negotiation of goal direction for co-operative transport, Ant Colony Optimization andSwarm Intelligence, 191 – 202 (2006)

[30] S. Berman, Q. Lindsey, M.S. Sakar, V. Kumar,S.C. Pratt: Experimental Study and Modeling ofGroup Retrieval in Ants as an Approach to Collec-tive Transport in Swarm Robotic Systems, Proceed-ings of the IEEE 99(9), 1470 – 1481 (sept. 2011)

[31] J.M. Esposito: Distributed grasp synthesis forswarm manipulation with applications to au-tonomous tugboats, IEEE International Conferenceon Robotics and Automation 2008 (IEEE, 2008)1489 – 1494

[32] Z.D. Wang, V. Kumar: Object closure and manipu-lation by multiple cooperating mobile robots, IEEEInternational Conference on Robotics and Automa-tion (ICRA) 2002 (IEEE, 394 – 399

[33] Z. Wang, V. Kumar: Object closure and manipu-lation by multiple cooperating mobile robots, IEEEInternational Conference on Robotics and Automa-tion 2002 (IEEE, 2002) 394 – 399

[34] P. Song, V. Kumar: A potential field based ap-proach to multi-robot manipulation, IEEE Interna-tional Conference on Robotics and Automation 2002(IEEE, 2002) 1217 – 1222

BIBLIOGRAPHY 27

[35] Z. Wang, Y. Hirata, K. Kosuge: Control a rigidcaging formation for cooperative object transporta-tion by multiple mobile robots, Proc. of IEEE Inter-national Conference on Robotics and Automation(ICRA) 2004, 1580 – 1585

[36] J. Fink, N. Michael, V. Kumar: Composition of vec-tor fields for multi-robot manipulation via caging,Robotics: Science and Systems (2007)

[37] O. Holland, C. Melhuish: Stigmergy, self-organization, and sorting in collective robotics, Ar-tificial Life 5(2), 173 – 202 (1999)

[38] J. L. Deneubourg, S. Goss, N. Franks, A. Sendova-Franks adn C. Detrain adn L. Chretien: The dy-namics of collective sorting robot-like ants and ant-like robots, From Animals to Animats, Proc. of SABConference, Cambridge, MA 1990 (MIT Press, Paris

[39] R. Beckers, O. Holland, J. Deneubourg: From Lo-cal Actions to Global Tasks: Stigmergy and Collec-tive Robotics, Proceedings of the 14th InternationalWorkshop on Synthesis and Simulation of LivingSystems 1994, ed. by R. Brooks, P. Maes (MITPress, Cambridge, MA 1994) 181 – 189

[40] T. Wang, H. Zhang: Multi-robot collective sort-ing with local sensing, IEEE Intelligent AutomationConference 2003 (IEEE, 2003)

[41] T. Wang, H. Zhang: Collective Sorting with Mul-tiple Robots, IEEE International Conference onRobotics and Biomimetics (ROBIO) 2004, 716 – 720

[42] Y. Yang, M. Kamel: Clustering ensemble usingswarm intelligence, IEEE Swarm Intelligence Sym-posium 2003 (IEEE, 2003) 65 – 71

[43] A. Martinoli, F. Mondada: Collective and coopera-tive group behaviours: Biologically inspired experi-ments in robotics, Experimental Robotics IV, 1 – 10(1997)

[44] J. Werfel, R. Nagpal: Three-dimensional construc-tion with mobile robots and modular blocks, In-ternational Journal of Robotics Research 27(3-4),463 – 479 (2008)

28 BIBLIOGRAPHY

[45] Justin Werfel: Building patterned structures withrobot swarms, In Proceedings of the 19th Interna-tional Joint Conference on Artificial Intelligence (IJ-CAI05 2005 (Morgan Kaufmann Publishers, 2005)1495 – 1502

[46] J. Werfel, Y. Bar-Yam, D. Rus, R. Nagpal: Dis-tributed construction by mobile robots with en-hanced building blocks, IEEE International Con-ference on Robotics and Automation 2006 (IEEE,2006) 2787 – 2794

[47] J. Werfel, R. Nagpal: Extended Stigmergy in Collec-tive Construction, IEEE Intelligent Systems 21(2)(2006)

[48] J. Werfel: Building blocks for multi-robot construc-tion, Distributed Autonomous Robotic Systems 6,285 – 294 (2007)

[49] Y. Terada, S. Murata: Automatic modular assem-bly system and its distributed control, InternationalJournal of Robotics Research 27, 445 – 462

[50] J. Wawerla, G.S. Sukhatme, M.J. Mataric: Collec-tive construction with multiple robots, IEEE/RSJInternational Conference on Intelligent Robots andSystems 2002 (IEEE, 2002) 2696 – 2701

[51] R.L. Stewart, R.A. Russell: Building a loose wallstructure with a robotic swarm using a spatio-temporal varying template, IEEE/RSJ Interna-tional Conference on Intelligent Robots and Systems2004 (IEEE, 2004) 712 – 716

[52] R.L. Stewart, R.A. Russell: A distributed feed-back mechanism to regulate wall construction bya robotic swarm, Adaptive Behavior 14(1), 21 – 51(2006)

[53] et al. T. Huntsberger: Robotics challenges forrobotic and human Mars exploration, Proceedingsof Robotics 2000 2000, 2000) 340 – 346

[54] R. A. Brooks, P. Maes, M. J. Mataric, G. More:Lunar Based Construction Robots, Proceedings ofInternational Conference on Intelligent Robots andSystems 1990, 1990)

BIBLIOGRAPHY 29

[55] C.A.C. Parker, H. Zhang, C.R. Kube: Blindbulldozing: multiple robot nest construction,IEEE/RSJ International Conference on IntelligentRobots and Systems 2003 (IEEE, 2003) 2010 – 2015

Collective Manipulation and Construction -...

Documents

Transcript of Collective Manipulation and Construction -...