How to improve group performances on colocated synchronous manipulation tasks? Jean Simard

Adrien Girard

Ana¨ıs Mayeur

Mehdi Ammi

CNRS-LIMSI Univ. Paris-Sud Orsay, 91400 Email: [email protected]

CNRS-LIMSI Univ. Paris-Sud Orsay, 91400 Email: [email protected]

CNRS-LIMSI Orsay, 91400 Email: [email protected]

CNRS-LIMSI Univ. Paris-Sud Orsay, 91400 Email: [email protected]

Abstract—Collaborative Virtual Environments provide new working methods to associate several users on the same problem through different space and time configurations (synchronous/asynchronous, distant/colocated). This new approach enables the simultaneous management of complex environments with several constraints for a more robust and efficient process. Previous studies investigate various features of synchronous Collaborative Virtual Environment but they mainly focus on configurations involving mainly 2 users. This paper presents an experimental study which investigates the collaborative work for more users. We propose to compare 2 users with 4 users to perform a closely coupled molecular deformation task. The results show that groups of 4 participants are more efficient than pairs, even more if they plan their actions during a brainstorming step. An important outcome of this study shows that the role of a leader improve the coordination between the participants and limit the conflicting communication.

I. I NTRODUCTION Nowadays, Collaborative Virtual Environments (CVE) are important in the evolution of working methods, allowing several users to create, exchange and manipulate information in shared spaces. This approach capitalizes creative abilities, associates complementary skills and distributes workloads. Some studies already proposed the association of partners with complementary skills without necessary direct and physical interactions between them. Jensen et al. [1] proposes collaborative scientific visualization for distant learning. Lidal et al. [2] allow a collaborative review of projects in geological resources prospection. de Oliveira et al. [3] designed a platform for industrial remote training. However, these applications only explored lowly coupled actions between partners. With recent advances in Information & Communication Technology (ICT) and reduction of communication delays between remote sites, several applications with closely coupled interactions were explored. Basdogan et al. are among the first to study the simultaneous manipulation of shared artefacts with the haptic feedback [4]. Salln¨as et al. [5] carry out several experimentations to understand the role of the haptic feedback to improve the presence and the awareness of partners in a CVE. More recently, Iglesias et al. [6] propose a complete system with haptic feedback for the collaborative assembly of 3D CAD models. Simard et al. [7] investigate synchronous strategies to manipulate flexible molecular structures. If these works explore with relevance closely coupled interactions for

configurations involving 2 partners, only few studies investigate configurations requiring more users. This field has already been explored in the domain of psychology [8]. Chan et al. [9] are among the first to investigate complex group dynamic for collaborative interaction on a custom turn-taking protocol to improve the communication. In this paper, we propose to explore closely coupled interactions in a CVE with groups of 4 participants and to compare it with pairs in similar experimental conditions. The objective of this study is to highlight the evolution of working efficiency, collaborative interactions, and communication processes. This study is based on a synchronous and colocated CVE: (1) synchronization keeps spontaneity of workers interactions and requires interactive coordination of actions; (2) colocalization avoids incomplete feedback in inter-subjects interactions (no video, bad audio quality or no real eye contact could be a constraint for workers) and requires mutual active supervision between partners. The context of the proposed study is the real-time deformation of huge molecules. Deformation of huge molecule is a complex process based on 3 levels: • Macromolecular level concerns the manipulation of 2 macromolecules with large amplitude displacements and orientation. • Molecular level consists in associating molecules from the macromolecular manipulation. Molecules will be deformed to match each other according to several criteria such as surface complementarity or electrostatic balance. • Atomic level is the refinement of molecular model through the manipulation of atomic groups and focusing on important biological and chemical interactions such as salt-bridges, hydrophobic effect and hydrogen bonding. Even if these procedures occur at different scales, they are based on a common set of elementary subtasks: 1 • Search consists in finding a target (residue , atom, etc.) according to several criteria (articulation, shape, hydrophilic regions, etc.). • Selection consists in accessing and holding the target. • Deformation concerns the deformation of molecular structures to minimize energy criterion. 1 Elementary

molecular structures constituted of several atoms

Indep Mac OS X). The VRPN client software, VMD in our case, is ran on another computer and send/receive informations to/from VRPN servers.

Evaluation consists in calculating energies (potential, electrostatic, etc.) of a given molecular interface or structure to evaluate stability of the molecular complex. After exploring the search and selection elementary subtasks in previous works [7], we will focus on the deformation procedure. This subtask enables the creation of new conformations, at several scales, which are involved in subsequent assembly processes. Beyond the role of this procedure in the molecular design process, this task provides working configurations where several constraints (kinematic, physical, etc.) must be managed in the same time. These constraints should highlight closely coupled manipulations involving simultaneous manipulations of shared artefact (atoms, residues, structures, etc.) through investigated collaboration strategies. This paper is structured as following: Section II expose the objectives and formulate hypothesis; in Section III, we detail the hardware setup and apparatus; we show the experimental protocol in the Section IV; we present the results and corresponding analysis in the Section V; and finally, we summarize the study and present some prospects for CVE in the Section VI. •

II. O BJECTIVES AND HYPOTHESIS Previous work showed that pairs are more efficient than single participants for tasks during the molecular design process [10]. In order to go further in the reflexion of the impact of the number of participant, the present research explores the collaborative activity of 4 users. In fact, some tasks during the molecular design process can go beyond the ability of 2 participants. They require the simultaneous control of complex structures during the deformation process: the global workload should be shared between several partners. Each partner would manage one or several constraints (DoF, attractive field, etc.) according to the actions of other partners. We propose to study this group configuration with the following factors: (1) complexity of the task with several levels of complexity and constraints are proposed to participants; (2) role of the tasks’ planning through a brainstorming session, to improve the efficiency of the group [8]. Hypothesis are the following: H1 A group (4 users) will be more efficient than pairs. H2 Actions planning provides better structure for the group and improve communication between partners.

Shared visualization is assumed with a beamer on a large display screen. Participants can not modify the point-of-view of theone virtual environment. an experimental choiceas to system; VRPN server forThis eachis haptic tool and VMD avoid confusion of a potential competition between particithepants VRPN client. Sony camera (HDR-SR11E) set up if we giveAthe ability to every participantswas to modify in order to record the global view.audio and video.

Figure 1: 1: Illustration ofofthe Figure Illustration theexperimental experimentalplatform. platform. Finally, a Sony camera (HDR-SR11E) was set up in order to record audio and video. IV. M ETHOD

A. Participants 4. Method

16 adults (4 women and 12 men) with a mean age of Participants µ 4.1. = 26.5 years old (σ = 5.1) served as subjects. They 16 adults womens and 12 mens) with a mean age of were recruited(4from a research laboratory and were linguistic, µ = 26.5 years old (σ = 5.1) served as subjects. They fluid mechanic, virtual reality, audio-acoustic researcherswere or recruited from a research laboratory and were linguistic, webfluid development engineer and audio-acoustic all French native speaker. mechanic, virtual reality, researchers or Participants have all normal or and optically corrected vision and web development engineer all French native speaker. had allThey normal or naive optically corrected vision no Participants audio deficiency. were to the purposes but and all no audio deficiency. They were naive to the purposes of the have at least a previous 1 hour experience on this platform. experiment and were given a full explanation of the experi-

procedures. A written informed consent was obtained B. mental Variables

before participation with the option to withdraw from the study at any time. Each participant had variables at least a are: 1 hour of 1) Independent variables: Independent previous experience on this platform.

Vi1

tools per participant (one for each hand), whereas groups had 1 haptic tool per participant. 4 groups of 4 participants and 8 pairs have been tested (there is more pairs than groups). The type of molecule (within subject variable). 2 different molecules were proposed to the participants : “Prion” vs. “Ubiquitin”. The first molecule called Prion (see 2KFL on Protein DataBase) has 1779 atoms. This molecule is not strongly constrained and can easily be divided into subtasks. The second molecule called Ubiquitin (see 1UBQ on Protein DataBase) has 1231 atoms. This molecule is strongly constrained and should require at least 2 participants to deform. The Prion molecule has always been the first tested molecule. The possibility to do or not a brainstorming before each experiments (between subject variable). The Vi3 has 2 modalities: “brainstorm” vs. “no brainstorm”. Based on the supposition that groups will have

submitted to COMPUTER GRAPHICS Forum (4/2011).

Vi2

III. H ARDWARE SETUP AND APPARATUS Experiments were conducted on a collaborative platform coupling standard desktop workstations and providing shared view on a large shared screen display supported by a projector (see Figure 1); the point-of-view is fixed. The experimental platform is a desktop system for molecular deformation. The highly configurable graphics renderer (VMD [11]) is connected to a real molecular simulation (NAMD [12]) through IMD [13] for a real-time simulation. R 4 haptic tools (PHANToM Omni from SensAble ) interacted with the simulation through VRPN [14] as a client/server

The number of participants (within subject vari-

A larger would have better for able). sample Two number’s set of been participants werestatistical chosen analysis. However, the cost to synchronize each group of : “2” (c.f. “pair”) vs. “4” (c.f. “group”). Pairs and 4 participants was very high. Moreover, the 2 independent groups used thewesame hardware to perform the variables imply that should recruitsetup 4 additional groups that pairs participants. had at their disposal 2 haptic which task; meansnote 16 additional

Vi3

(Vi1 ) Tw "pa sam pai (on par ipa tha (Vi2 ) fer "P (se mo vid (se mo lea the (Vi3 ) exp 2m the din at t bra an can exp

4.2.2.

Depe

(Vd1 ) nee ite (Vd2 ) nu (Vd3 ) ber the • •

(Vd4 ) ob po (Vd5 ) Ne to

On Figure 2, there is a view of what is shown to participants. At the top of the screen, the RMSD score is displayed with green/red bar. The green bar is growing when the RMSD is becoming better (the RMSD score is decreasing). A RMSD threshold is defined for each task.

difficulties to coordinate their actions, a short time period (max. 1 minute) at the beginning of each task allow participants to make a brainstorming in order to elaborate a strategy or establish an action plan. During this short time period, participants can not manipulate the virtual environment but they can explore the visual area with their cursors. 2) Dependent variables: Dependent variables are: Vd1 The completion time. Completion time is the time needed for participants to achieve the task. Time was limited to 10 minutes. Vd2 The number of selections. The number of selections is the mean number of atoms selected by each participant per minute. Vd3 The verbal communication. Each verbal exchange between participant is classified into 2 categories: • the participant is ordering subtasks to others • the participant is providing information about its position or about the environment Vd4 The average speed of the cursor. Average speed is obtained with a simple trapezoidal integration of logged positions (each 50 ms). Vd5 The force applied by the participant (in Newton) is scaled and relayed to the simulation through VRPN. The force is non-null when an atom is selected and a participant is pushing/pulling the atom. C. Task Participants are experimented on molecular deformation in a CVE. They deform a molecule to reach a targeted molecule which is displayed with transparency (like a ghost). To deform this molecule, participants are able to select atoms and apply forces on them. These forces are transmitted to the simulation (NAMD) which applies the deformation and returns all new atoms positions to the visualisation software (VMD). The similarity between the deformed molecule and the target molecule is evaluated by the RMSD score. Root Mean Square Deviation (see equation (??equ-RMSD)) is one of the score used to evaluate the difference between molecules in biology. v u N u1 X 2 kci − ti k (1) RMSD (c, t) = t N i=1

where N is the number of atoms and ci , ti are respectively the current and the targeted position of the atom i. On Figure 2, there is a view of what is shown to participants. At the top of the screen, the RMSD score is displayed with green/red bar. The green bar is growing when the RMSD is becoming better (the RMSD score is decreasing). On the deformable molecule, all atoms are shown while only the Ribbon (bezier curve on backbone atoms) is shown on the ghost molecule. When participants are selecting an atom, the entire residue is highlighted (see deformed residue on Figure 2). Moreover, when an atom is selected, the targeted

We wil efficien

5.1. Im

RSMD score indicator fixed residue deformed residue ghost molecule

ghost residue

deformed molecule

Figure molecule. Figure 2: 2: The The deformed TRP-ZIPPER molecule. On the deformable molecule, all atoms are shown while

residue (see ghost(bezier residuecurve on Figure 2) is displayed all only the Ribbon on backbone atoms) is with shown atoms. displayed residue help participants to pull on theThis ghost molecule. Whenshould participants are selecting an atom,in thethe entire is highlighted (see deformed residue atoms rightresidue direction. Finally, some atoms have been Moreover,(see when an residue atom is on selected, on Figure fixed in the2). simulation fixed Figurethe 2)tarand geted (see ghost on Figure is displayed acts likeresidue an anchor. If not,residue the molecule can2)move freely in with all atoms. and Thiscould displayed help view. particithe environment moveresidue outsideshould the screen pants to pull atoms in the right direction. Finally, some atoms

D. Experimental design Participants are split in groups depending on 2 conditions: (1) “pair” vs “group” (see Vi1 ) and (2) “brainstorm” vs “no brainstorm” (see Vi3 ). In both conditions, 3 tasks are proposed. The first is a training task based on TRP-CAGE molecule (see 1L2Y on Protein DataBase). The second and third tasks are respectively Prion and Ubiquitin molecule (see Vi2 ). The number of participants factor Vi1 is crossed with the brainstorm factor Vi3 . The experimental design avoid the order effect by counterbalancing the number of participants factor and the brainstorm factor in 4 groups. Possible effects like practice or fatigue are controlled. A larger sample would have been better for statistical analysis. However, the cost to synchronize each group of 4 participants was very high. Moreover, the 2 independent variables imply that we should recruit 4 additional groups which means 16 additional participants. V. R ESULTS AND ANALYSIS Statistical analysis of variance (Kruskal-Wallis test with χ2 critical value) is proposed with an alpha level of 0.10 due to the low number of participants [15]. A. Improvement of work efficiency Table I: Evolution of completion time and number of selections depending on participant number Vi1 and molecule Vi2 Prion

time (s) # of selections / mn

Ubiquitin

Pair

Group

Pair

Group

189

238

409

386

6

7.8

6.3

6.8

Figure number

Figu pending Particip Prion m ence be

Figu minute number selectio 0.319).

On F

Table I shows the mean completion time Vd1 and the mean number of selections Vd2 according to the number of participants Vi1 and the tested molecule Vi2 . Participants are faster on the Prion molecule than on the Ubiquitin molecule (χ2 (1) = 5.606, p = 0.018) with no difference between pairs and groups (χ2 (1) = 0.004, p = 0.951). Moreover, participants present a similar number of selections for both molecules (χ2 (1) = 0.454, p = 0.501). These two results clearly show that these two configurations present similar working efficiency. H1 is not validated according to these first measures. Previous works [16] show that the bimanual mode, in which 2 end-effectors are controlled by the same user, is less efficient than collaborative configurations, in which end-effectors are controlled by different users. In fact, the separate control of end-effectors enables a better simultaneous management of constraints than the bimanual mode. By transposing this result in this study, one would think that group of 4 users can manage more constraints and more complex structures. However, as developed in section 5.2, the increase of the number of users introduces many communication and interaction conflicts which limit the efficiency of this working configuration. Next results (see Section V-B) will show that the activity planning structures the group and provides a better organization of communication which improve the group efficiency.

to establish the required actions and to define the role of each partner during consecutive steps. Next sections will show that a common strategy (with a brainstorming) decreases conflicting actions and thus improves the communication since each partner is aware about his role and about the role of his partners. Groups without a defined strategy (no brainstorming) try to define an action plan during the progress of the task. However, they must consider simultaneously the evolution of the molecular structure under the actions of other partners, which disturbs the global progress of the task. This situation leads to conflicting actions which increases the level of communication to solve contradictory actions. C. Role of brainstorming in groups Table III: Evolution of completion time, number of selections and verbal communication depending on the brainstorming Vi3 and participant number Vi1 Pair Brainstorm time (s)

Group

Without

With

Without

With

320

279

475

149

# of selections / mn

6.4

5.8

10

4.6

# of verbal comm.

124

132

110

55

B. Planning of actions through a brainstorming Table II: Evolution of completion time and number of selections depending on the brainstorming Vi3 and molecule Vi2 Prion Brainstorm

Ubiquitin

Without

With

Without

With

time (s)

279

132

464

339

# of selections / mn

7.5

5.7

7.7

5.1

Table II shows the mean completion time Vd1 and the mean number of selections Vd2 according to the possibility to do or not a brainstorming Vi3 and to the molecule Vi2 . These results include both pairs and the groups of 4 users. Participants who plan their actions greatly decrease the completion time with a nearly significant effect (χ2 (1) = 2.709, p = 0.099). The decrease is of 53 % for Prion molecule and of 27 % for the Ubiquitin molecule. The measure of the number of selections presents also a significant decrease according to the brainstorming factor (χ2 (1) = 5.603, p = 0.018). The decrease is of 24 % for the Prion molecule and of 34 % for the Ubiquitin molecule. These two results clearly show that the activity planning significantly improves the efficiency since groups provide a better time performance with less actions (number of selections). These results validate H2 . In fact, during the brainstorming session, participants (1) analyze the different features of the task (complexity factors, simultaneous constraints, etc.), and (2) establish an action plan to perform the task. The brainstorm allows the group

Table III shows the mean completion time Vd1 and the mean number of selection Vd2 according to the possibility to have or not a brainstorming Vi3 and according to the number of participants Vi1 (4 or 2). Groups (4 users) who plan their actions greatly decrease the completion time (decrease of 69 % with χ2 (1) = 3, p = 0.083) while pairs do not present a significant decrease (χ2 (1) = 0.224, p = 0.636). The measure of the number of selections follows the same evolution. Groups with actions planning greatly decrease the number of selections (decrease of 54 % with χ2 (1) = 5.333, p = 0.021) while pairs do not present a significant decrease (χ2 (1) = 0.893, p = 0.345). Table III show the total number of communication Vd3 according to the possibility to do or not a brainstorming Vi3 and to the number of participants Vi1 . Groups with actions planning greatly decrease the number of communication (decrease of 50 %) while pairs do not present a significant increase (increase of 6 %). These results show that the brainstorming session mainly improves the effectiveness of the group (4 users) but have not effect on pairs. In fact, the group configuration presents an important level of inter-subjects interaction (up to 3) while pairs configuration enables the interaction with only one partner. During the progress of the task, subjects in groups can simultaneously: (1) collaborate effectively with some partners and (2) come into conflicts with other partners leading to abundant communication. The brainstorming avoids such conflicting situations (which greatly decrease the verbal communication) since the tasks and the working spaces are

20 15

P1

P3

P2

P4

19

previously partitioned. The pair have not such conflicting sit10 uations since the face-to-face communication is more natural: 5 the brainstorming has a very limited effect. 3

0

2

2

(G11)

D. Group structure

speed speed (m/s) (m/s)

A thorough analysis of previous measures and the analysis Figure 9: Number of orders (Vd3 d3) depending on each particofipant measures related to communication between (P1, P2, P3 and P4) of the group on (G11). partners, we can highlight two categories of groups according to the possibility to do or not a brainstorming. In following sections: G1 refers to groups with a leader and G2 refers to groups without P3 P1 P2 P4 a leader. 1 0.88

Table IV: Evolution of number speed 0.72 0.7 and average 0.66of orders depending on each participant (P1, P2, P3 and P4) of the group G1 0 Participant

P1

(G11) P2

P3

P4

force force (N) (N)

Figure 10: partici# ofAverage orders speed (V19 3 2 on each 2 d4 ) depending d4 pant (P1, Average P 2, P 3 speed and P(m/s) 4) of the group on (G ). 11 0.88 0.66 0.70 0.72

Figure 11 shows force profiles about participants P1 and in (G11). On Figure 11a, the participant P1 has a chaotic force profile with numerous different selections (11 selecA complementary the verbal communication’s tions longer than 1analyse s). OnofFigure 11b, the participant P2 content of requests, type replies, longer etc.) shows made(type only few selections (4 of selections than 1that s). the

different requests of P1 correspond to (1) indications of actions The leader is the participant who will coordinate actions to fulfill (direction of deformation, amplitude, begin and stop, of every participant in the team. He could become the leader etc.), or (2) designations of working spaces (target space, if he is able to manage the team (knowing the objective, he residues or subtasks structuretotoevery manipulate, etc.). The and large number can give other participant) is accepted of by short selections (Figure 3a) with an important mean speed others. If nobody accept the leader, everybody will do (Table V), corresponds to the different actionsfor of actions. exploration anything he wants; there is no coordination and designation to support these requests. The long selections Onpartners Figure 9, P 1 is clearly a leader; he gives indications to of other (Figure 3a) and the lower mean speed of their other participants. Moreover, Figure 10 shows that P1 is the gestures, correspond to actions of deformation which require more active participant. The force profile confirms the result holding the manipulated structures for a long time. In this case, on Figure 11a where P1 is selecting a important number of the speed is constrained by the low dynamic of the molecular atoms. The leader explores and analyzes the environment to environment (high viscosity) theand precision needed the decide what are the priorityand tasks other tasks lessfor impordeformation task. tant. He can also occasionally help another participant. According to these analyses and the nature of the communiOncontent, the opposite, 11b the profile cation’s we can Figure consider thatisP1 is typical the teamforce leader. His of a follower participant (lead by the leader). He receives role concerns mainly the coordination of actions of the group subtasks to do and selected only atoms needed without exaccording to the defined strategy (i.e., brainstorming step). ploring. The follower participant pull the selected atom to its These different requests are mostly accepted (i.e., affirmative final destination without releasing which explains long phase reply or no reply). of selection and small mean speed on Figure 10 (see P2). of selection and small mean speed on Figure 10 (see P2).

5 4 3 2 1 0

Table V: Evolution of number of orders and average speed 5.4.2. Groups without leader depending on each participant (P1, P2, P3 and P4) of the group G1 At the opposite of groups with a leader, there is groups where interactions are anarchic.

0

20

40

60 time (s)

80

100

120

P1from (a) Force profile (VVd5 ) for P fromG1(G11) d5 (a) Force profile for the the participant participant P1 d5

force force (N) (N)

7

P2

5 4 3 2 1 0

selection

0

20

40

60 time (s)

80

100

120

P2from (b) Force profile (VVd5 ) for P fromG1(G11) d5 (b) Force profile for the the participant participant P2 d5

Figure 3: of of participants P1 Pand P2 from G1 . Figure 11:Force Forceprofiles profiles participants 1 and P 2 from (G11). 1) Presence of a team leader: Table V shows, for G1 , the number of orders given by each participant during the verbal communication, and the average speed for each participant. We observe that P1 presents more requests and provides a higher mean speed than the 3 other participants. Figure 3 shows the force profiles (and thus selection phases) for P1 and P2 in G1 . We observe in Figure 3a that P1 presents an important number of selections with short durations (11 selections during one experiment). Figure 3b shows that P2 presents fewer selections but with more important durations submitted COMPUTER GRAPHICS Forum (4/2011). submitted COMPUTER GRAPHICS than P1toto(4 selections > 10Forum s). (4/2011). The three other participants present a similar force profile. These different measures highlight the high level of activity (i.e., number of selection, requests, etc.) and the specific role of P1 in the group.

Participant

P1

# of orders P 1

P2 12

5

0.49

0.34

20 Average speed (m/s) 15 12 10

## orders orders

## orders orders

Jean Simard, Mehdi Ammi & Anaïs Mayeur / How to improve group performances on collocated synchronous manipulation tasks?

P2

P3 P 35

0.44

P4 8 P4 0.39

2) Groups without a team leader: Table V shows the 8 5 number of orders given by each in G2 during 5 participant 5 the verbal communication and the average speed for each 0 (Gpresent 44) participant. We note that the 4 partners similar numbers of requests and similar mean speeds (with a small advantage Figure 12: Number of orders (Vd3 d3) depending on each parfor P1). These measures show that there is no users that emerge ticipant (P1, P2, P3 and P4) of the group on (G44). in this group. Furthermore, we observe that the mean number of requests 12 increases shows the slightly number of given by each parof theFigure group (15orders %) compared to the ticipant during verbal communication for (G44 ). Every parprevious configuration. However, the mean speed significantly ticipant has given few orders to others. P1 has given a little decreases (55 %) which increases the global completion bit more orders than P4 who has given more orders than P2 time (see previous results). These two results highlight the and P3. diminution of the global effectiveness of this group since they shows average speed for each participant on Figure 13more communicate forthe a more important completion time and (G ). Average speed of each participant is similar. Note that 4 a low4 level of activity (low mean speed). average speedofonthe (Gverbal "no brainstorm" content (see Figure 10) 11) withcommunication’s The analysis shows is more important than average speed of (G ) with "brain4 4 that participants give simultaneously different requests (by storm". talking louder to overcome the other conversations). These exchanges corresponds to parallel andparticipants local tasks is involving a On Figure 12, we see that every giving orders to other. of In participant this case, every propose own limited number (not participants the 4 partners in theits same time). In fact, there is no global strategy since the group did not plan the activity before the beginning of the experiment. According to these results, we can conclude that this group does not present a leader who can coordinate actions of the participants and structure the global activity and the communication flow of the group. At the opposite of the previous

configuration, this group presents a less established group structure. Table VI: Evolution of completion time depending on the presence of a leader and molecule Vi2 Prion

Ubiquitin

Leader

Without

With

Without

With

time (s)

380

96

424

348

3) Synthesis: Table VI shows the completion time Vd1 according to the leader’s presence or not, and the molecule Vi2 . According to the previous results, 2 groups are considered with a leader (with brainstorming) and the 2 others without a leader (without brainstorming). We observe that groups with a leader are significantly faster than other groups (decrease of 75 % for the Prion molecule and of 18 % for the Ubiquitin molecule). We observe that the improvement level is linked with the molecule type, and thus with the nature of tasks to fulfill. The Prion molecule provides several independent subtasks without a strong coupling. Thus, the leader has not difficulties to share these subtasks between the involved participants. The brainstorming step enables the definition of a suitable strategy which does not evolve during the deformation process. The Ubiquitin molecule provides several closely coupled subtasks with a strong link between them. The leader has more difficulties to share the activity between the involved participants and should manage simultaneously the interference between the actions. Even if the brainstorming enables planning the required actions, the strategy must evolve to integrate last modifications and constraints of the environment. However, the brainstorming step enables the identification of a leader (to coordinate actions and exchange between the partners) and defines a first strategy. This results explains a less important improvement than for the Prion molecule. VI. C ONCLUSION AND PROSPECTS To conclude this study on molecular deformation, we show that increasing the number of participants improves the efficiency [17]. However, unorganized groups misexploit the potential of the number of participants and decrease their efficiency compare to smaller groups. Fortunately, groups with a coordinator is one solution. Groups should be structured as following: • 1 coordinator who should be able to manage the group and occasionally help every other participants. He should be chosen according to all the participants. The coordinator will decide the global strategy so he should be able to explore the entire environment to chose the best strategy. • other participants will be manipulators. They should trust the coordinator and accept decisions from him. We saw that the brainstorming before each task improves the efficiency. The coordinator should be charged of this mission. The difficulty is to adapt the strategy during the

task, depending of new and unexpected events. It should be interesting to make pauses during the task to adapt the strategy. Further studies should be conducted to confirm these results in other tasks’ context. In future works, we will propose haptic tools to assist collaborative work. The main objective will be to help participants to coordinate between themselves by encouraging one of the participant to begin a coordinator. R EFERENCES [1] N. Jensen, S. Olbrich, and W. Nejdl, “Building a collaborative virtual environment for scientific visualization,” Learning Lab Lower Saxony, Tech. Rep., 05 2007. [2] E. M. Lidal, T. Langeland, C. Giertsen, J. Grimsgaard, and R. Helland, “A decade of increased oil recovery in virtual reality,” in IEEE Computer Graphics and Applications, vol. 27, no. 6. IEEE Computer Society Press, 2007, pp. 94–97. [3] J. C. de Oliveira, X. Shen, and N. D. Georganas, “Collaborative virtual environment for industrial training and e-commerce,” in Virtual reality technologies for future telecommunications systems, A. Pakstas and R. Komiya, Eds. San Francisco, CA, USA: Willey, 11 2002, pp. 79– 102. [4] C ¸ . Bas¸do˘gan, C.-H. Ho, M. A. Srinivasan, and M. Slater, “An experimental study on the role of touch in shared virtual environments,” ACM Transaction on Computer-Human Interaction, vol. 7, no. 4, pp. 443–460, 12 2000. [5] E.-L. Salln¨as, K. Rassmus-Gr¨ohn, and C. Sj¨ostr¨om, “Supporting presence in collaborative environments by haptic force feedback,” ACM Transaction on Computer-Human Interaction, vol. 7, no. 4, pp. 461– 476, 12 2000. [6] R. Iglesias, S. Casado, T. Guti´errez, A. Garc´ıa-Alonso, W. Yu, and A. Marshall, “Simultaneous remote haptic collaboration for assembling tasks,” in Multimedia Systems, vol. 13, no. 4. Springer, Heidelberg, Germany, 01 2008, pp. 263–274. [7] J. Simard, M. Ammi, and M. Auvray, “Study of synchronous and colocated collaboration for search tasks,” in Proceedings of Joint Virtual Reality Conference (JVRC), D. Fellner, Ed. Eurographics Association, 09 2010, pp. 51–54. [8] B. Tuckman, “Developmental sequence in small groups,” Psychological bulletin, vol. 63, no. 6, pp. 384–399, 06 1965. [9] A. Chan, K. MacLean, and J. McGrenere, “Designing haptic icons to support collaborative turn-taking,” International Journal HumanComputer Studies, vol. 66, pp. 333–355, 05 2008. [10] J. Simard, M. Ammi, and M. Auvray, “Collaborative strategies for 3D targets search during the molecular design process,” Transaction on Systems, Man Cybernetics (part C), 2012. [11] W. F. Humphrey, A. Dalke, and K. Schulten, “VMD: Visual Molecular Dynamics,” Journal of Molecular Graphics, vol. 14, no. 1, pp. 33–38, 02 1996. [12] J. C. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid, E. Villa, C. Chipot, R. D. Skeel, L. Kal´e, and K. Schulten, “Scalable molecular dynamics with NAMD,” Journal of computational chemistry, vol. 26, no. 16, pp. 1781–1802, 05 2005. [13] J. Stadler, R. Mikulla, and H.-R. Trebin, “IMD: a software package for molecular dynamics studies on parallel computers,” International Journal of Modern Physics, vol. 8, no. 5, pp. 1131–1140, 06 1997. [14] R. M. Taylor II, T. C. Hudson, A. Seeger, H. Weber, J. Juliano, and A. T. Helser, “VRPN: a device-independent, network-transparent VR peripheral system,” in Proceedings of the ACM symposium on Virtual reality software and technology, ser. Virtual Reality Software and Technology 2001. New York, NY, USA: ACM, 11 2001, pp. 55– 61. [15] P. R. Peres-Neto and J. D. Olden, “Assessing the robustness of randomization tests: examples from behavioural studies,” Animal Behaviour, vol. 61, no. 1, pp. 79–86, 2001. [16] J. Simard, M. Ammi, and A. Mayeur, “Comparative study of the bimanual and collaborative modes for closely coupled manipulations,” Elsevier International Journal of Human-Computer Studies, 2012. [17] ——, “How to improve group performances on collocated synchronous manipulation tasks?” in Proceedings of Joint Virtual Reality Conference, ser. JVRC – EuroVR-EGVE. Nottingham, United-Kingdom: Eurographics Association, 09 2011, pp. 91–94.