conflict resolution in multi-agent based intelligent environments

lable at ScienceDirect

Building and Environment 45 (2010) 574–585

Contents lists avai

Building and Environment

journal homepage: www.elsevier .com/locate /bui ldenv

Conflict resolution in multi-agent based Intelligent Environments

Jaewook Lee*

Department of Convergence Technology, KT Central R&D Laboratory, 463-1 Jeonmin-dong, Yusung-gu, Daejeon 305-811, Republic of Korea

a r t i c l e i n f o

Article history:Received 1 April 2009Received in revised form2 July 2009Accepted 21 July 2009

Keywords:Intelligent EnvironmentsMulti-agent systemsCollaborative designEnvironmental conflictsConflict resolution

* Tel.: þ82 42 870 8199; fax: þ82 42 870 8679.E-mail address: [email protected]

0360-1323/$ – see front matter � 2009 Elsevier Ltd.doi:10.1016/j.buildenv.2009.07.013

a b s t r a c t

Intelligent Environments are able to support ever-changing environmental needs by automatically anddynamically adjusting their key parameters without explicit human intervention. However, the currentdevelopment of Intelligent Environments primarily focuses on the technical aspects of the physicalcomponents, and does not give sufficient consideration to the dynamic interrelationship between peopleand the built environment. As a result, environmental conflicts among users, activities, and physicalsettings are not properly resolved. To overcome this limitation, this article proposes a model for multi-agent based Intelligent Environments and a conflict resolution mechanism by applying the concept ofcollaborative design. To demonstrate the types of conflicts and their resolution method, a set of hypo-thetical cases is developed and tested. The result of the case study shows that the proposed model canenable the environment, as an organization of multiple agents, to intelligently perceive the user activityand efficiently handle setting conflicts, thus minimizing the burden to the users of controlling the setting,while maximizing their environmental satisfaction.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

Buildings and other inhabited environments are designed tosupport diverse human activities, yet they often fail to satisfy thisprimary role due to their static and rigid nature. That is, they havemobility and dynamics that are too limited to cope with thechanging needs of their users, activities, and contexts. Unlike suchconventionally built environments, Intelligent Environments areable to support ever-changing environmental needs by automati-cally and dynamically adjusting their key parameters (temperature,light, sound, etc.) without explicit human intervention.

Since Negroponte’s introduction of the concept of IntelligentEnvironments [1], attempts to make buildings ‘intelligent’ havebeen actively conducted in various domains, thanks in part to theadvent of affordable computer technologies. These attempts can becategorized into two approaches: (i) the development of individualdevices or agents that react to simple environmental changes,independently from other devices or agents [2–4]; and (ii) thedevelopment of multiple devices or agents that control variousbuilding components, responding to more complex environmentalchanges in collaboration with other devices or agents [4–11]. Anexample of the first approach is the i-Land project [12], whichcomprises a set of room-ware components, such as an interactivetable and wall for office workers. As an example of the secondapproach, a multi-agent system developed by Xerox PARC [13],

All rights reserved.

utilizes multiple temperature controllers to improve the energymanagement of an office building.

However, most of the attempts to make buildings ‘intelligent’have dealt primarily with the technical aspects of building compo-nents, largely ignoring the dynamic interrelationship betweenpeople and the built environment. Consequently, various environ-mental conflicts among users, their activities, and physical settingsare not completely resolved, which may lead to user dissatisfaction[14]. Specifically, in multi-agent based Intelligent Environments, inwhich multiple intelligent agents modify environmental settings bynegotiating with other agents [15–17], these environmentalconflicts should be properly and promptly resolved to ensure theconsistency of environment-wide setting modification [18].

To overcome the drawbacks of the current approach, this articleproposes a model of multi-agent based Intelligent Environments thatis rooted in the concept of collaborative design. The proposed modelcomprises a hierarchical organization to facilitate the collaborative(design) decision making of agents for the efficient resolution of theenvironmental conflicts that arise among objects, users, and (users’)activities. To validate the proposed model and demonstrate its conflictresolution mechanism, a set of hypothetical test cases is used.

2. Theoretical backgrounds

2.1. Multi-agent systems (MAS)

Multi-agent systems (MAS) have the potential to conceptualize,design, and implement complex systems involving multiple agents

mailto:[email protected]

www.sciencedirect.com/science/journal/03601323

http://www.elsevier.com/locate/buildenv

J. Lee / Building and Environment 45 (2010) 574–585 575

and mechanisms [15,19]. Theoretically, agents can be built in anyimaginable environment, and either a centralized, single-agent ora decentralized (or distributed) multi-agent system is possible.However, as the agent behavior is strongly dependent on the natureof the task environments, single-agent systems work well whentask environments are simple, small, and static, whereas MAS aremore appropriate for complex, large, and dynamic environments[16,19]. In MAS, multiple agents are ‘‘situated in some environment,and capable of flexible, autonomous action in that environment’’[20]. They are interconnected to function in a manner exceeding thecapability of any singular agent [21]. Human organizations,composed of multiple human agents, also share this fundamentalcharacteristic of MAS [22,23].

As ‘‘the most basic technique for tackling any large (orcomplex) problem is to divide it into smaller, more manageablechunks,’’ the power of MAS comes from the division of labor andthe cooperation of the agents [24]. Rather than employinga centralized, single agent to deal with a complex task, designerscan decompose a task into smaller subtasks and assign them todifferent agents, thereby obtaining a synthesized solution to theoriginal task from the partial solutions sought by the agents withtheir own interests and goals [19,25]. Within MAS, agents need tointeract and negotiate with one another to achieve their individualgoals, as well as common organizational goals. The majoradvantages of MAS include [17,19]: (i) representation of thedifferent (and possibly conflicting) interests and goals of differententities; (ii) robustness against failure with distributed controland responsibilities; (iii) scalability through easy agent additionand modification; and (iv) accommodation of uncertainty anddynamics of the task environment. However, conflicts arising fromthe different goals, perspectives, and interests of individual agentsmust be efficiently resolved in order to achieve the shared orga-nizational goals.

2.2. Layered agent structure

Hierarchical, layered structures have long been regarded asa natural way of organizing and solving any complex problem, andthus have been widely studied in diverse fields of research,including scientific computing, business data processing, andheuristic problem solving. In general, organizational structure isclosely related to the size and complexity of an organization. Thebasic idea of a hierarchical structure is that a large, complex orga-nization (or system) can be designed by decomposing it intosubgroups (or subsystems), which perform particular sub-func-tions. This successive partitioning of the organization typicallyforms a pyramidal, hierarchical authority structure, and the overallbehavior of the organization is largely determined by the interac-tion between its higher-level and lower-level subgroups. Theoret-ically, organizations built on a hierarchical structure require muchless information transmission among their constituent parts thando other types of organizations [26].

The characteristics of the hierarchical structure described aboveare also valid in the development of MAS. Minsky [25] suggests thathigher-level intelligence (i.e., mind or agency) can only be built onthe hierarchical structure of multiple agents, ‘‘because each agenthas only a single job to do: it needs only to ‘look up’ for instructionsfrom its supervisor (higher-level agent), then ‘look down’ to gethelp from its subordinates (lower-level agents).’’ Coen [27,28] alsoclaims that software agent systems benefit from layered systemarchitectures when dealing with complex and dynamic real-worldproblems that particularly require frequent agent addition, dele-tion, or alteration. The concept of the hierarchical, layered agentstructure has been widely applied to the development of differenttypes of MAS [17,19,29].

2.3. Conflict resolution in MAS

The need for decentralization of an organization is due to thecomplexity of its tasks as well as the limited information andcapability of the individual agents. Decentralization becomesimperative because it is impossible to gain a synoptic view of thenumerous factors that should be taken into account for organiza-tional decision making [30]. But once decentralization is necessary,it contributes to the rise of organizational conflicts due to goal orperception difference between agents. Consequently, to accomplishshared organizational goals, appropriate mechanisms for resolvingconflicts are required. These resolving mechanisms can be viewedas centralized coordinating processes [31]. In other words, the(micro-level) decentralization of tasks calls for a (macro-level)centralized decision-making process to generate organizationalactions. As such, the need for hierarchical control layers to resolveorganizational conflicts is implied.

When a conflicting situation arises within an organization, theparties involved in the conflict tend to seek a method of resolvingthe conflict that achieves their own goals. However, not everyconflict can be resolved by the conflicting parties themselves. If oneor both of the parties resist coming to an agreement by holding totheir original positions, the resolution of the conflict is impossible.In particular, in an organization with shared goals that need to beachieved within a limited time frame, a conflicting situation cannotbe held indefinitely. Therefore, timely action is required to resolvethe conflict, and a third-party mediator is often involved to help theparties move to a settlement [32,33].

During the process of mediation, the third-party mediatorclarifies the conflicting situations by identifying the source of theconflict and understanding the respective positions of the con-flicting parties [33]. Then, considering the impact of the outcomesof possible resolutions, the mediator assists the disputants inreaching an agreement. In resolving organizational conflicts,personnel at the managerial level generally play the role of medi-ators and make decisions on behalf of the conflicting individuals.Thus, the primary task of managers is to coordinate the behavior oftheir members for the successful accomplishment of organizationalgoals.

Like human organizations, in MAS (often combined with sensornetworks), agent interaction may call for hierarchical control layers,particularly when the task environment is dynamic, complex, anduncertain. Although there are many systems based on distributedconsensus or coordination [34–38] depending on the size and typeof the task environment, supervising or coordinating agents may berequired to improve system performance by efficiently handlingagent conflicts [28,39–41]. For example, Scerri’s robotic soccer [42]utilizes a higher-level agent to monitor the overall game state andresolve conflicts between lower-level robot players (i.e., agents)who directly interact with the game environment and other co-players. The primary advantage of this approach is that the timeand cost of processing the information required to pursue organi-zational goals can be reduced by allowing the lower-level agents topursue self-contained tasks under their own autonomy, with thehigher-level agents becoming involved only when conflicts arisebetween lower-level agents.

3. Proposed model

3.1. Application of the concept of collaborative design

In a broader sense, a task that an Intelligent Environment dealswith at a given point in time can be considered a dynamic designactivity that transforms a present situation into a desirable one[43–45]. The Intelligent Environment perceives user activities and

J. Lee / Building and Environment 45 (2010) 574–585576

the current state of the environment (problem identification),determines the goal state and constraints that reflects the needs ofthe users (goal formulation), finds potential environmental settingsthat can achieve the goal state and abide by the constraints (solu-tion synthesis), and evaluates the potential settings to select themost suitable one among them (evaluation).

As such, a new approach proposed by this study begins from theobservation that adjusting the physical parameters of IntelligentEnvironments (after they are built) can be compared to the processof designing environments in the first place (before they are built).This approach further proposes to replace the actions of humandesigners with intelligent software agents that will extend theongoing negotiation and collaborative decision making that char-acterizes the design phase of buildings.

In the proposed model, individual building components(walls, doors, furniture, HVAC systems, lights, etc.) are repre-sented as multiple intelligent agents that know how to behavegiven any activity by any user. They have the ability to perceivecontextual changes in the environment, and adjust their behavioraccording to their immediate context to support the context-specific activities of users. Furthermore, these multiple agents, asmembers of a team-like organization, resolve various environ-mental conflicts through collaboration and negotiation amongthemselves, similar to the design process of a human designteam. Specifically, the actions of User Identification, ActivitySensing, and Collaborative Setting Modification used by Intelli-gent Environments correspond to similar activities performed byhuman designers (Fig. 1).

In addition, considering the complexity, dynamics, and uncer-tainty of Intelligent Environments where (i) diverse users andobjects may coexist; (ii) both of them may interact with each otherin various ways depending upon the activities of the users; and(iii) they may continue to change (i.e., users’ preferential changes,user/activity changes, and object addition/removal/alteration), theproposed model is built as a multi-agent system with hierarchicalagent layers. The agent layers include lower-level object agentswho directly control the physical setting of the environment basedon their own autonomy, and higher-level coordinating agents whoare in charge of resolving environmental conflicts from a levelbelow them.

3.2. Environment and profiles

An environment is composed of three basic elements, Users,Activity, and Setting, and the intersection of these three elementsdefines the environment (i.e., E¼ f (User, Activity, Setting)). Becausethese elements are naturally dynamic, they keep changingaccording to the type of activities, the preference of users, and thespatiotemporal context of activities. The major components thatcomprise the proposed model of Intelligent Environments

Intelligent Environment

User Identification

Activity Sensing

Setting Modification through Communication

Fig. 1. Intelligent Environmen

correspond to the three elements described above, and each ofthem is represented as a profile, a computational description.

First, individual objects (i.e., doors, windows, and lighting)that construct a physical setting are programmed to sense andrespond to users with embedded processors and mechanisms.Such programmed objects can be viewed as multiple intelligentagents, and their behavior descriptions are stored in ‘ObjectProfiles.’ Second, the initial environmental preferences of indi-vidual users are also programmed and collected in ‘User Profiles,’with each profile including a user ID, object IDs, property vari-ables, and their values (e.g., lighting level, temperature, humidity,and chair/desk height). Third, given that within a single envi-ronment users may have different setting preferences fordifferent activities at different times, these activity-specificpreferences need to be encoded into ‘Activity Profiles.’ The threetypes of profiles enable the Intelligent Environment to identifyusers and their activities, and modify the setting of the envi-ronment accordingly (Fig. 2).

In the proposed model, each user has only one User Profile, butcan have multiple Activity Profiles depending on the number ofactivities that the user may perform. Both the User Profile and theActivity Profile(s) can be encoded in a badge that is carried by theuser and read by objects through wireless communication (e.g.,Bluetooth or RFID).

3.3. Layered agent structure of the model

For the efficiency of environmental conflict resolution and theconsistency of environment-wide modification, the proposedmodel incorporates a layered structure to organize the intelligentagents, which determines the setting of the environment. Themodel is comprised only of three levels of agents arranged ina hierarchical structure (Fig. 3) in consideration of the structure ofthe task environment, which is hierarchical in its composition (i.e.,Building Level – Zone Level – Object Level). Kulkarni’s ReBa(Reactive Behavioral System) employs a similar layered structure tobuild a multi-agent based Intelligent Environment [46,47]. Itsegments user activities into different layers (e.g., meeting andmovie) in order of priority, and organizes a set of correspondingreactions (i.e., action rules) to each activity layer. Consequently, itpays more attention to the functional hierarchy than the spatialone, which is an important consideration in the proposed model.The following describes the three agent levels and their roles.

OA (Object Agent): the behavior of each object is controlled byan OA. The OA of an object identifies an activity of the user whooccupies or uses the object, and responds to it based on the Object,User, and Activity Profile. OA is the lowest-level agent that directlymanipulates the environmental setting.

BMA (Behavior Management Agent): each BMA controls thebehavior of an assigned zone (a room, a lobby, etc.) that includes

Design Team

Problem Identification

Solution Search

Confirmation through Communication

t vs. human design team.

Activity: axxx

Preferences: Music - No Lighting - 500 lux Temp. - 60

AudioControl /

CommunicationObject Profile

User ID: David

Preferences: Music - Classical Lighting - 400 lux Temp. - 65 Humidity - 50% Chair Height - 21" Desk Height - 30" …

LightingControl /


DeskControl /


...

User-Activity Profiles

(User Preferences)

Setting Modification

(Task Environment)

Object Profiles

(Object Agents)Load Apply

Activity: a001

Preferences: Music - No Lighting - 500 lux Temp. - 60

User Profile

Activity Profiles

Fig. 2. Profiles and setting modification.


a number of OAs. By summarizing user activities based on the datareceived from the OA(s), the BMA can handle conflicts betweenOAs. BMAs are intermediate agents in the hierarchical organization.

EMA (Environment Monitoring Agent): the EMA is the top-levelagent that controls the overall behavior of a whole environment. Itcan identify the context of each zone and deal with zone-levelconflict (i.e., conflict between BMAs).

4. Conflict resolution in the model

4.1. Types of conflicts

In the proposed model of Intelligent Environments, interactionsbetween users, their activities, and physical settings can be repre-sented as a relationship between respective profiles (Fig. 4). Whentwo or more profiles are present in the same zone, environmentalconflicts may arise due to a perceptual difference between OAs, ora preferential difference between users. Of the six different types ofconflicts in the environment, the three most prominent typesof conflicts are: ‘Object Profile Conflict,’ ‘User Profile Conflict,’ and‘Activity Profile Conflict.’

Control

Module

Knowledge

Module

Comm.

Module

Control

Module

Knowledge

Module

Comm.

Module

Control

Module

Knowledge

Module

Comm.

Module

OA1 OA- nOA2

Knowledge

Module

Comm.

Module

Control

Module

Knowledge

Module

Comm.

Module

OA1

BMA1

EMA

Messages

Knowledge

Module

Comm.

Module

User/Activity


Fig. 3. Multi-agent based Intelligent Environm

4.1.1. Object-side conflicts (object profile conflicts)This type of conflict generally results from perception or goal

differences between OAs. An OA, as a spatiotemporally and ratio-nally bounded entity, can only perceive a (small) part of the envi-ronment, about which the OA has subjective knowledge as well asa limited reasoning capacity. As a result, different OAs may inter-pret the same user activity differently. Furthermore, each OA has itsown goal, which may be different from that of other OAs. Theseperception and goal differences are a major source of Object ProfileConflict.

4.1.2. User-side conflicts (User–Activity Profile conflicts)When two or more users are in the same zone or room at the

same time, there might be differences in their preference regardingthe settings of the zone or room. That is, an overall room settingthat one user prefers may conflict with that of other user(s). Thispreferential difference between users may result in User ProfileConflicts in the Intelligent Environment.

Similarly, while a single user may normally perform a singleactivity for a certain time period, two or more users may performdifferent activities in the same zone or room at the same time,

Control

Module

Knowledge

Module

Comm.

Module

Control

Module

Knowledge

Module

Comm.

Module

OA- nOA2

Knowledge

Module

Comm.

Module

BMA2

CO1

O3

O2

L

K

H

T

EMA(Building Level)

BMA(Zone Level)

OA(Object Level)

Task

Environment

ent: agent structure and communication.

User

Profiles

Object

Profiles

Activity

Profiles

User Profile Conflict

Object Profile ConflictActivity Profile Conflict

User - Activity Profile Conflict

User - Object Profile Conflict

Activity - Object Profile Conflict

Fig. 4. Three types of profile conflicts.


which may also lead to an Activity Profile Conflict. For instance, inan office, the lighting and sound preference for the resting activityof one user may conflict with the preferences of another user forhis/her working activity. These types of conflicts arise due topreferential differences between users, and may not be easilypredicted in advance.

4.2. Conflict resolution: case study

To validate the setting modification of the proposed model ofIntelligent Environments, three hypothetical test cases are devel-oped. These cases aim to demonstrate how the proposed model canbe applied to the implementation of Intelligent Environments. Assuch, the major concern of this case study is the process of theagents’ conflict resolution, and each of the three cases is designedto explain a specific conflict resolution of the proposed model. Theobjects of the test environment satisfy the following basicconditions:

- Each object has a set of sensors that can detect users and theiractivities.

- Each object has a set of actuators that can modify its setting forthe current user(s).

- The setting modification of objects is based on User, Activity,and Object Profiles.

When a user is in a zone or a room, the possible activities of theuser can be represented as a set U¼ {u0, u1, u2.}. At any given point

User ID: u001

object-1: setting_0 object_prop_1 = a object_prop_2 = b ... object_prop_n = x

Activity: a001

object-1: setting_1 object_prop_1 = c

Activity: a002

object-1: setting_1 object_prop_2 = d …

User-Activity Profile

user - 1

object-1: o001

user-1: u001

Plan View

object-2: o002

Fig. 5. Case 1: plan v

in time, the user is assumed to perform one of these activities. Foreach activity of the user, there exists an object setting that the userprefers for the activity. This again forms a set of preferred objectsettings for the user and his/her activities, O¼ {o0, o1, o2.}. Oncethe object agent (OA) perceives the user’s ID and his/her activity,the agent searches for the corresponding object setting andexecutes an action that modifies the current object setting (i.e.,setting modification) to the desired one for the perceived useractivity. Thus, the capability of the OA can also be represented bya set of actions, A¼ {a0, a1, a2.}. This process of agent actiongeneration can be represented as the following function:

action: U�O / A (or an ¼ action (un, on))The above function maps a user’s activity and object setting to

an object action. In the proposed model, a set of ‘‘User Activity –Object Setting’’ pairs forms a ‘‘User-Activity Profile.’’ Similarly, a setof object actions constructs an ‘‘Object Profile.’’ Therefore, theprimary task of an OA is to perceive a user activity, and map thispercept to an object action. The following three cases are based onthis process of agent-action generation.

4.2.1. Case 1: one user and two objects (resolution ofobject-side conflicts)

A room or a zone commonly contains multiple objects. To servethe user properly, for a given user activity, each object needs to beset in accordance with the setting of the other objects. That is, thesetting of one object is closely related to that of other objects in thesame room or zone in which the user activity occurs. For example,the height of a chair is dependent on the height of a desk that a useruses for her/his office work. Therefore, each OA must perceive theuser and his/her current activity correctly, and this agent perceptshould also be matched to the percept of other OAs for successfulsetting modification. However, because the agent perception isspatiotemporally bound, individual agents may perceive the sameuser activity differently according to their own sensory mechanism,which may cause an agent conflict.

As shown in Fig. 5, this case is composed of a single user and twoobjects. When the presence of the user, User-1, is detected by theOAs of Object-1 and Object-2 through their built-in sensor(s), theOAs identify User-1’s ID and load her/his User-Activity Profile intotheir memory. The OAs then adjust the properties of the objects tothe initial setting for the user. Thereafter, as the user’s activitychanges, the OAs modify the object setting based on the loadedUser-Activity Profile. At this moment, as Fig. 6 depicts, while OA-1(o001) perceives a user activity (activity-1) correctly as a001, OA-2(o002) perceives the same activity incorrectly as a002. This raises

object - 1

Object ID: o001


object_prop_1: set_action_op_1() object_prop_2: set_action_op_2() ... object_prop_n: set_action_op_n()

Object Profile

object - 2

Object ID: o002



Object Profile

iew and profiles.

Object Agent-2User-1Object Agent-1

u001

o001

User Perception: u001

User-Activity Profile Load: u001.p

Setting Modification: set_action_o001(u001.p.setting_0)

Activity Perception: u001.p.a001



o001: setting_0


t=0

u001: activity-1


o001: setting_1

User Perception:u001

Activity Identification: u001.p.a002

t=1

o002

o002: setting_0

o002: setting_2

Object-Side Conflict

Fig. 6. Case 1: object-side conflict.


an ‘object-side’ setting conflict. If this object(-side) conflictbetween o001 and o002 is not successfully resolved, the objectsetting, setting_2, which has been modified by o002, may notsuccessfully support the user activity, a001; and the user, u001, maynot be satisfied with the current environmental setting.

User-Object Agent-1

u001

o001

User Perception: u001



Activity Perception: u001.p.a001


o001: setting_0t=0

u001: activ

o001: setting_1t=1

Conflict Resolution: O

BMA: Conflict Id

a001, a0

BMA: Conflict R

a001

Fig. 7. Case 1: resolution o

One easy way of resolving the object conflict is to allow the userto correct the object setting by and for him/herself. However, for anIntelligent Environment, this should be the last option to beconsidered. Instead, an appropriate resolution mechanism isrequired for successful setting modification by intelligent objects.

Object Agent-21



ity-1




o002

o002: setting_0

o002: setting_1

bject-Side Conflict

entification

02

esolution

f object-side conflict.

User ID: u001


Activity: a001


Activity: a002



user - 1 object - 1

Object ID: o001



Object Profile

object-1: o001

user-1: u001

Plan View

User ID: u002


Activity: a001


Activity: a002



user - 2

user-2: u002

Fig. 8. Case 2: plan view and profiles.

User-2User-1

t=0

Object Agent-1

User Perception:u001, u002

u001 u002

o001

User-Activity Profile Load:u001.p, u002.p

Setting Modification:set_action_o001(u001.p.setting_0) o001:

u001.p.seting_0

User-Side Conflict

Fig. 9. Case 2: user-side conflict.


In the proposed model, a ‘‘Behavior Management Agent (BMA),’’ asa coordinator, resolves these object-side conflicts resulting from theperceptual difference of agents. The process of conflict resolutionby the BMA is shown in Fig. 7. After identifying the perceptualdifference between OA-1 (o001) and OA-2 (o002), the BMA queriesthe activity history and pattern (stored in the database) of User-1(u001), and resolves the conflict by correcting o002’s percept froma002 to a001. Then, o002 modifies its object setting to setting_1. Assuch, to properly modify the setting of objects, OAs require bothagent communication and conflict resolution.

4.2.2. Case 2: two users and one object (resolution ofuser-side conflicts)

Unlike the previous case, this case comprises multiple users, andshows how a single object can serve multiple users simultaneously.The basic condition for each user and object is the same as before.However, the interaction between the users and the object isdifferent according to the object type and the users’ activity. Ingeneral, objects that compose the built environment can be cate-gorized into two types, according to their temporal serviceability:(i) objects that can only serve a single user at a time, and (ii) objectsthat can serve multiple users at the same time. The first categoryincludes personal equipment or furniture, such as chairs and desks,whereas the second category includes most shared equipment orfurniture, such as lighting devices, HVAC units, and sofas.

In the case of objects in the first category, there is no preferentialconflict between users (i.e., user-side conflict). As an object can onlybe occupied and used by a single user at a time, the OA can simplymodify the object setting for the current user (however, object-sideconflicts may still occur, as in the previous case). However, in thecase of objects in the second category, as two or more users sharethe same object, there may be preferential conflicts between theusers in terms of the setting of the object. These are ‘user-side’setting conflicts that arise due to preferential differences betweenusers.

As shown in Figs. 8 and 9, after OA-1 (o001) perceives thepresence of both User-1 (u001) and User-2 (u002), it loads theirUser-Activity Profile (i.e., u001.p and u002.p). Next, o001 selectsu001’s profile for its setting modification. Here, because the currentsetting of the shared object, o001, is based on u001’s profile, theother user, u002, may not be satisfied with the current objectsetting, which may result in a user (-side) conflict.

To resolve this type of conflict, the OA needs to consider thesetting preferences of two users at the same time. This is the pointthat requires a mechanism to combine two sets of user preferencesover the shared object. In the proposed model, a new type of user

profile, ‘Group-Activity Profile,’ is created and managed by BMAs toresolve the user (-side) conflicts between multiple simultaneoususers.

As Fig. 10 describes, when the two users are present in thesame room at the same time, the OA-1 (o001) perceives theirpresence and retrieves a Group-Activity Profile (i.e., g001.p)through communication with the room’s BMA. Thereafter, as theuser group (i.e., u001 and u002) changes its activity, o001 modifiesits setting based on the loaded Group-Activity Profile, g001.p. ThisGroup-Activity Profile is created by the BMA utilizing a set ofresolution methods, such as politics, consensus, bargaining, and/orlearning.

4.2.3. Case 3: two users and two objects (combinedconflict resolution)

In general, building environments contain multiple users, whoperform multiple activities with multiple objects that support theseactivities. Therefore, compared to the previous cases, this case ismore like a real-world environment, in which both object (-side)and user (-side) conflicts can be observed. The basic assumption forthis case is that one or both of the two objects serves at least oneuser at a time. This assumption not only simplifies the case, but alsoguarantees at least one user–object interaction in the environment.According to the serviceability of the objects, this case can havedifferent sub-cases. As shown in Table 1, these sub-cases can begrouped into four case groups according to the type of agentconflicts and their resolution.

As in the previous cases, the objects in Category 1 serve a singleuser at a time (i.e., personal objects), whereas the objects in Cate-gory 2 can serve multiple users simultaneously (i.e., shared

User-2User-1

t=0

Object Agent-1

User Perception: u001, u002

u001 u002

o001

Group-Activity Profile Load: g001.p

Setting Modification: set_action_o001(g001.p.setting_0)

Activity Identification: g001.p.a001: u001=a001, u002=a002

u001: activity-1 u002: activity-2

BMA: Group-Activity Profile

g001.p: u001, u002

o001: g001.p.setting_0

Setting Modification: set_action_o001(g001.p.setting_1) o001:

g001.p.setting_1t=1

Group ID: g001

User: u001, u002


Activity: a001

u001:a001,

u002:a002


Activity: a002

u001:a002,

u002:a002


Group - Activity

Profile

group - 1

Conflict Resolution: User-Side Conflict

Fig. 10. Case 2: Group-Activity Profile and resolution of user-side conflict.


objects). Each sub-case represents a specific conflict type or agentbehavior pattern in relation to the user activity. In Category 1,conflicts can be observed only in Case Group 2, but in Category 2,conflicts can arise in both Case Groups 3 and 4. For example, if oneof the two users occupies both of the two pieces of personalfurniture (e.g., a chair and a desk), a (object-side) conflict of CaseGroup 2 may occur in a manner similar to the first use case, Case 1.On the other hand, if both of the two users occupy one of the two

Table 1User–object interactions and conflict-resolution types.

pieces of shared furniture (e.g., a sofa and a bed), a (user-side)conflict of Case Group 3 may arise in a manner similar to the seconduse case, Case 2.

Unlike the other cases, in Case Group 4, because both of the twoobjects interact with either or both of the two users at the sametime, both an object- and a user-side conflict can arise. This is themost complex case of all, and its agent interaction and conflictresolution are further discussed below.

User ID: u001


Activity: a001 object-1: setting_1 object_prop_1 = c

Activity: a002 object-1: setting_1 object_prop_2 = d …


user - 1 object - 1

Object ID: o001

Setting Modification object_prop_1: set_action_op_1() object_prop_2: set_action_op_2() ... object_prop_n: set_action_op_n()

Object Profile

object-1: o001

user-1: u001

Plan View

User ID: u002


Activity: a001 object-1: setting_1 object_prop_1 = c

Activity: a002 object-1: setting_1 object_prop_2 = d …


user - 2

user-2: u002

object-2: o002

object - 2

Object ID: o002

Setting Modification object_prop_1: set_action_op_1() object_prop_2: set_action_op_2() ... object_prop_n: set_action_op_n()

Object Profile

Fig. 11. Case 3: plan view and profiles.

user-1 user-2

object-1 object-2

: User-Object Interactionuser-side conflict

object-side conflict

: Conflict

Fig. 12. User–object interaction and conflict types.


As shown in Fig. 11, in the room, two users (u001 and u002)share a single object (o001), and one (u002) of the users occupiesand uses the other object (o002). As in the previous cases, theindividual User-Activity Profiles of the users are active, and the OAof each object sets the object setting with its Object Profileaccording to the loaded User-Activity Profiles.

o001User-1Object Agent-1

u001

User Perception: u001, u002

Group-Activity Profile Load: g001.p


Activity Identification: g001.p.a001: u001=a001, u002=a002


t=0

t=1

Conflict Resolution: O


BMA: Group-A

g001.p: u0

u001: activity-1 u

BMA: Conflict

u002: a00

BMA: Conflict

u002:


Conflict Resolution:

Fig. 13. Case 3: resolution of us

Because object-1 (o001), which is a Category-2 object, serves thetwo users at the same time, a user-side conflict can arise betweenthe two users in a manner similar to Case 2. Similarly, according tothe perceptual difference between the two objects that simulta-neously interact with user-2, an object-side conflict between theobjects can also occur, as in Case 1 (Fig. 12).

In this case, to resolve the user- and object-side conflict at thesame time, two different methods are applied in combination. Thatis, the user-side conflict is resolved by a Group-Activity Profilecontaining the combined environmental preference of the twousers, whereas the object-side conflict is resolved through theBMA’s coordination, based on accumulated user activity data.

As shown in Fig. 13, the OA of o001 detects the two users andloads their Group-Activity Profile from the database managed bythe BMA of the room, and it modifies the setting as the activities ofthe users change. Although a user-side conflict can be resolved

Object Agent-2






o002

o002: u002.p.setting_0

o002: u002.p.setting_2

bject-Side Conflict

User-2

u002

ctivity Profile

01, u002

002: activity-2

Identification

2, a003

Resolution

a002

User-Side Conflict

er- and object-side conflict.

OA1 OA2

Knowledge Module

Knowledge-base

Message Processing

User/Group

Data

Object

Data

Comm. Module

Comm. Channel

Control

Communication

Protocol

User/Group

Activity

Data

Behavior Management Agent (BMA)

Conflict Detection

Conflict Resolution

Control

Module

Knowledge

Module

Communication

Module

Control

Module

Knowledge

Module

Communication

Module

Object-side

Conflict

Conflict Management

o001 o002

u001 u002

User ID: u001

Preferences: …

User ID: u002

Preferences: …

User-side

Conflict

Sensing Acting

Badge: User-Activity Profiles

Object & User

Conflict

Coordination &

Group Profile

Fig. 14. Case 3: system configuration for conflict resolution.


through this resolution process, there still remains an object-sideconflict between o001 and o002 for the current activity of theuser, u002. As both objects interact with u002’s activity, theperceived user activity by one OA may not be identical to that ofthe other OA. Here, again, the BMA intervenes and coordinates thetwo objects to resolve the conflict as described in Case 1. All ofthese resolution processes are done in real time, and thereby theenvironment can support the users and their activity in a timelyand appropriate manner. Fig. 14 shows the overall systemconfiguration of the proposed model for the resolution of envi-ronmental conflicts.

5. Discussion

This study presents a method and process of conflict resolutionwithin the proposed model of Intelligent Environments througha set of hypothetical test cases. However, as the test cases aredesigned for verifying the conceptual feasibility of the model sug-gested in this article, there are more dimensions that need to beconsidered in order to successfully implement the proposedmethod in real-world environments. Of these additional dimen-sions, the major considerations are as follows:

- Although the test environment used in the case studiescontains two users and two objects at most, the interactionamong the users and objects is already sufficiently complex.Therefore, in an environment where a large number of occu-pants exist simultaneously, the processing of all environmentalconflicts among occupants may not be practical; even in anenvironment where such a modification is made, it is signifi-cantly difficult to completely meet the environmentalrequirements of all occupants. It is, therefore, expected thatsome degree of modification to the proposed method may benecessary when it is applied in reality, depending on thebuilding or the zone. In a large office, for example, the entirespace could be divided into several small zones, each of whichwould be assigned a group profile to manage. On the otherhand, in a conference hall, where identical or similar activities(e.g., concert and party) are held for a limited period of time,the group profile can be subdivided depending on the type ofactivities, and appropriate environmental settings can beapplied.

- The responsiveness of Intelligent Environments depends ontheir ability to obtain an accurate perception of user activities.Therefore, to improve this accuracy, the number of sensors and


actuators required may reach hundreds or thousands,depending on the type and size of the environment. This raisesanother issue of energy consumption, which needs to beconsidered in the following two respects. First, for the powerneeded to drive sensors and actuators, which may not berequired for the conventional built environment, studies toreduce such additional energy consumption have been activelyconducted in a range of fields [34,48]. On the other hand, theenergy generally required for elements of the building infra-structure, including HVAC systems, lights, and elevators, can bereduced by introducing the proposed model for the IntelligentEnvironment. That is, intelligent agents can optimize thebuilding operation, thereby eliminating unnecessary energyconsumption [5,13,37].

- In this research, not much discussion is given to the controlprocess by the Environment Monitoring Agent (EMA) due tocurrent technological limitations and economic infeasibilities.However, to endow a building-level intelligence, in which theobjects, rooms, and the building that contain them as a wholeare dynamically responsive to the users, more study on thissubject will be required for the future development of Intelli-gent Environments.

6. Conclusion

The resolution of environmental conflicts is a complexprocess, and is a critical one in multi-agent based IntelligentEnvironments. Furthermore, there is no single method that isapplicable to the resolution of all of the different types ofconflicts [33,49]. Just as human designers solve a design problemby communicating with each other and resolving conflictsamongst themselves, in the proposed model the object agentsmodify the setting of the environment through communicationand collaboration with other agents, as well as with their higher-level agents. The set of test cases demonstrates that user iden-tification, activity sensing, and communication of agents areessential for building Intelligent Environments. Moreover, theproposed model can enable the environment, as an organizationof multiple agents, to perceive the user activity intelligently,modify the settings, and handle setting conflict efficiently, tominimize the user burden of controlling the environmentalsetting. This, in turn, can overcome the drawbacks of conven-tional approaches to the development of Intelligent Environ-ments, and broaden its application to various building types toimprove the quality of the built environment.

Acknowledgements

The author would like to express his appreciation to ProfessorsYehuda Kalay, Galen Cranz, and Alice Agogino at the University ofCalifornia at Berkeley for their support and contribution to thisstudy.

References

[1] Negroponte N. Soft architecture machines. Massachusetts: MIT Press; 1975.[2] Kidd CD, Orr RJ, Abowd G, Atkeson CA, Essa I, MacIntyre B, et al. The aware

home: a living laboratory for ubiquitous computing research. In: Proceeding ofthe second international workshop on cooperative buildings. Lecture notes incomputer science 1670. Berlin: Springer-Verlag; 1999. p. 191–198.

[3] Krogh P, Grønbæk K. Roomware and intelligent buildings - objects andbuildings become computer interfaces. In: Architectural research and infor-mation technology. Nordic Association for Architectural Research. ÅrhusSchool of Architecture; 2001. p. 63–8.

[4] The Media Lab. Research groups, <http://www.media.mit.edu/research/>;2009 [accessed 13.3.09].

[5] Boman M, Davidsson P, Skarmeas N, Clark K, Gustavsson R. Energy saving andadded customer value in intelligent buildings. In: Proceedings of the thirdinternational conference on practical application of intelligent agents andmulti-agent technology; 1998. p. 505–517.

[6] Brumitt B, Meyers B, Krumm J, Kern A, Shafer, SAN. EasyLiving: technologiesfor Intelligent Environments. InProceedings of the second internationalsymposium on handheld and ubiquitous computing; 2000. p. 12–29.

[7] Coen MH. A prototype intelligent environment. In: Proceedings of the firstworkshop on cooperative buildings. Lecture notes in computer science 1370.Berlin: Springer-Verlag; 1998. p. 41–52.

[8] Colley M, Clarke G., Hagras H, Callaghan V, Pounds-Cornish A. Intelligentinhabited environments: cooperative robotics & buildings. In: Proceedingsof the 32nd international symposium on robotics and automation; 2001.p. 745–750.

[9] Holmes A, Duman H, Pounds-Cornish A. The iDorm: gateway to heteroge-neous networking environments. In: Proceedings of international ITEAworkshop on virtual home environments; 2002. p. 30–37.

[10] Mozer MC. Lessons from an adaptive house. In: Cook D, Das R, editors. Smartenvironments: technologies, protocols, and applications. New Jersey: J. Wiley& Sons; 2005. p. 273–94.

[11] Sharples S, Callaghan V, Clarke GS. A multi-agent architecture for intelligentbuilding sensing and control. Int Sensor Rev J 1999;19:135–40.

[12] Streitz NA, Geißler J, Holmer T. Roomware for cooperative buildings: inte-grated design of architectural spaces and information spaces.. In: Proceedingsof the first workshop on cooperative buildings. Lecture notes in computerscience 1370. Berlin: Springer-Verlag; 1998. p. 4–21.

[13] Huberman BA, Clearwater S. A multi-agent system for controlling buildingenvironments. In: Proceedings of the first international conference on multi-agent systems (ICMAS-95); 1995. p. 171–176.

[14] Arens E, Xu T, Miura K, Zhang H, Fountain M, Bauman F. A study of occupantcooling by personally controlled air movement. Energy Build 1997;27(1):45–59.

[15] Jennings NR. On agent-based software engineering. Artif Intell J2000;117(2):277–96.

[16] Russell S, Norvig P. Artificial intelligence: a modern approach. 2nd ed. NewJersey: Prentice Hall; 2003.

[17] Weiss G, editor. Multiagent systems: a modern approach to distributed arti-ficial intelligence. Massachusetts: MIT Press; 1999.

[18] Shafer SAN, Brumitt B, Cadiz JJ. Interaction issues in context – aware intelligentenvironments. Human Comp Interact (HCI) J 2001;16:363–78.

[19] Stone P, Veloso M. Multiagent systems: a survey from a machine learningperspective. Auton Robot 2000;8(3):345–83.

[20] Wooldridge M. Agent-based software engineering. IEE Proc Softw Eng1997;144(1):26–37.

[21] Nwana H. Software agents: an overview. Knowledge Eng Rev 1996;11(3):1–40.[22] Galbraith JR. Organization design. Massachusetts: Addison-Wesley; 1977.[23] Pfeffer J. Organizations and organization theory. Massachusetts: Pitman

Publishing Inc.; 1982.[24] Booch G. Object oriented analysis and design with applications. 2nd ed.

Benjamin/Cummings Publishing Company Inc; 1994.[25] Minsky M. The society of mind. New York: Simon and Schuster; 1986.[26] Simon HA. The new science of management decision. New York: Harper &

Row; 1960.[27] Coen MH. Building brains for rooms: designing distributed software agents.

In: Proceedings of the ninth annual conference on innovative applications ofartificial intelligence (IAAI’97); 1997. p. 971–977.

[28] Coen MH. Design principles for intelligent environments. In: Proceedings ofthe 15th national conference on artificial intelligence (AAAI’98); 1998.p. 547–554.

[29] Kolp M, Giorgini P, Mylopoulos J. Multi-agent architectures as organizationalstructures. Auton Agents Multi-Agent Syst 2006;13(1):3–25.

[30] Hayek FA. The road to serfdom. Chicago: University of Chicago Press; 1944.[31] Devons E. Planning in practice. Cambridge: Cambridge University Press; 1950.[32] Boulding KE. Conflict and defense: a general theory. New York: Harper; 1962.[33] Tidwell A. Conflict resolved?: a critical assessment of conflict resolution. New

York: Pinter; 1998.[34] Akyildiz IF, Kasimoglu IH. Wireless sensor and actor networks: research

challenges. Els Ad Hoc Netw 2004;2(4):351–67.[35] Carley KM. Computational and mathematical organization theory: perspective

and directions. Comput Math Org Theory 1995;1:39–56.[36] Giampapa J, Sycara K. Team-oriented agent coordination in the RETSINA

multi-agent systems. In: Proceedings of the first international jointconference on autonomous agents and multiagent systems. Bologna (I):ACM Press; 2002.

[37] Sandhu JS, Agogino AM, Agogino AK. Wireless sensor networks forcommercial lighting control: decision-making with multi-agent systems.In: Proceedings of the AAAI-04 workshop on sensor networks; 2004.p. 88–92.

[38] Vinyals M, Rodrguez-Aguilar JA, Cerquides J. A survey on sensor networksfrom a multi-agent perspective. In: Proceedings of the second internationalworkshop on agent technology for sensor networks (ATSN-08); 2008.

[39] Alshabi W, Ramaswamy S, Itmi M, Abdulrab H. Coordination, cooperation andconflict resolution in multi-agent systems. In: Proceedings of the IEEE inter-national joint conferences on computer, information, and systems sciences,and engineering (IEEE CISSE’06); 2006.

http://www.media.mit.edu/research/


[40] Bensaid N, Mathieu P. A framework for cooperation in hierarchical multi-agentsystems. In: Proceedings of the 11th international conference on mathemat-ical modeling and scientific computing; 1997.

[41] Park S, Sugumaran V. Designing multi-agent systems: a framework andapplication. Expert Syst Appl 2005;28:259–71.

[42] Scerri P. A multi-layered behavior based system for controlling RoboCupagents. In: Proceedings of the first robot world cup soccer games andconferences (RoboCup-97); 1998. p. 464-–74.

[43] Kalay YE. Architecture’s new media: principles, theories, and methods ofcomputer-aided design. Massachusetts: MIT Press; 2004.

[44] Rittel HWJ, Webber MM. Dilemmas in a general theory of planning. Policy Sci1973;4:155–69.

[45] Simon HA. The structure of ill-structured problems. In: Cross N, editor.Developments in design methodology. New Jersey: J. Wiley & Sons; 1984. p.146–66.

[46] Kulkarni A. Design principles of a reactive behavioral system for the intelligentroom. Bitstream: The MIT Journal of EECS Student Research 2002.

[47] Kulkarni A. A reactive behavioral system for the intelligent room. Technicalreport. MIT AI Lab 2002.

[48] O’Hare GMP, Marsh D, Ruzzelli A, Tynan R. Agents for wireless sensor networkpower management. In: Proceeding of the international workshop on wirelessand sensor networks (WSNET-05); 2005. p. 413–418.

[49] Tessier C, Chaudron L, Muller HJ, editors. Conflicting agents. KluwerAcademic; 2001.

conflict resolution in multi-agent based intelligent environments

Documents