Improving collaboration efficiency via diverse networked mobile devices

Collaboration is the combination and exchange of different core competencies and expertise, with the goal of creating a joint outcome in agreement, considering ideas and objectives of all participants. Based on a thorough literature review, we identified the main task phases involved in collaborative processes, depicted in Fig. 1.

Every collaboration session starts with the assignment of tasks and roles to each actor, which is facilitated by a software tool or performed in a group meeting setup beforehand (phase 1). On the one hand, assignment of tasks and roles is necessary to coordinate the work and team member in order to manage dependencies between activities[28]. Members without a task or a role can simply not participate in the group work. Passive or even no participation of actors lowers the individual’s satisfaction that is linked with motivation [29], engagement [30], and self-perceptions [31]. On the other hand, task assignment might lead to workload equality, which increases individual’s satisfaction as stated in [32]. This individual satisfaction may influence the performance of the complete team, which has been investigated in [33] but also the individuals willingness to continue the cooperation, what could be observed in [34]. Therefore, the first requirement (R1) on a collaboration framework is that member can coordinate activities.

In the next phase, users create drafts of the desired goals and the approach for reaching these goals (phase 2). This task can be performed individually or in a joint session. In this phase the outline of the work is created and the assigned tasks refined. This phase is accompanied by continuous comments and feedback loops, which are leading into group discussions (phase 3). The outcome of this phase is a draft in which ideas and expertise from all participants are considered and integrated. Communication between members across phases is crucial in order to synchronize the approaches to guarantee the progress towards the agreed joint goal. The communication between members is another requirement (R2) to support collaboration.

Afterwards the working style changes from group task to single task performance where the actual execution of the assigned tasks is performed (phase 4). The actual execution of the assigned tasks requires the active performance of an activity (R3). This phase is closely linked with the iterative reviewing and revision task (phase 5) in which all ideas, comments, and suggestion are discussed and incorporated. The exchange between members is necessary to communicate individual ideas and the satisfaction with the performance state. The decision about accepting the output is the outcome of phase 5 that leads to task establishment (phase 6).

In regards to computer support of teamwork additional attributes like the participants’ location needs to be considered. Therefore, the task performance can be identified upon the attributes location; indicating the physical location of the participants and division of labor. Thus, we can distinguish among four types of working styles between which actors switch most likely during:

Changes made by the independently and simultaneously operating team-members have impact on the overall system and the tasks being performed. Johnson et al. defined interdependence in the context of joint activity as follows: “Interdependence” describes the set of complementary relationships that two or more parties rely on to manage required (hard) or opportunistic (soft) dependencies in joint activity [6]. Our view of interdependence generalizes their definition. The existence of one task is not necessarily dependent on the existence or completeness of another task; there is no necessary “relies-on” relationship, but there may be a “can-be-positively-or-negatively-influenced-by” relation. The performance of one task can influence another task. Tasks can be interdependent through dependencies, but not as a consequence of merely existing. Dependencies and influences in only one direction can exist, which is not equivalent to interdependence describing a bi-directional dependence. Therefore, we use the term inferdependency, which refers to the combination of influence and dependence between two elements, one- or bi-directional.

We explain our notion of inferdependency via a simple example: Consider the scenario of biocenosis, where organisms coexist in the same habitat and interactions are evident in food or feeding relationships. In this scenario we have 4 actors: 1) a flower, 2) a butterfly, 3) a bee, and 4) a bear. A butterfly depends on a flower for nectar (food); the flower depends on the butterfly to pollinate and make seeds for reproduction. A direct interdependence between the butterfly and the flower exists. In addition to the butterfly, a bee coexists in the same habitat. The bee depends on the flower to produce honey; the flower depends on the bee to cross-pollinate. A direct interdependence between the bee and the flower exists. Bee and butterfly coexist without influencing each other. However, in reality both species influence each other by cross-pollinating flowers, thereby accelerating reproduction. The task performance of both species makes their jobs (pollinating flowers) easier and leads to an overall improved outcome (higher reproduction of flowers), which is also the precondition (food) for task performance. Consider a third participant, e.g., a bear eating honey (produced by the bees). We find a direct relation between bees and bears and indirect relation between butterflies and bears. Thus, influences and (indirect) dependencies, implying inferdependencies, between all participants exist, see Fig. 2.

Having interactive teamwork in mind, data and information exchange between experts has to be performed and inferdependencies must be connected. Inferdependent activities imply the presence of conflicting interests, which have to be coordinated to capture discrepancies before they become serious in order to achieve common goals with the help of common grounds [35]. Common ground is supported by continually informing others about changes that have occurred outside their views [35]. The determination of other’s activities (R5) is therefore crucial. Johnson et al. stated that not all team members must be fully aware of the entire scope of an activity; all must be aware of the interdependence in-between their activities [36]. Awareness of tasks and activities influences the coordination and task performance in a positive manner. Due to the establishment of shared knowledge and impact awareness (R6), team members can work together effectively and adjust their activities (R7) as necessary [37].

In this work it is not our aim to deploy an all-embracing catalog for collaborative systems. Thus, to state necessary aspects that will be covered in this work. Our aim is to establish a work environment that allows one to perform individual tasks as well as group work task in a natural manner including the consideration if inferdependencies. Based on the assumptions we state to support collaboration the requirements for collaborative work can be stated as follows:

To support a collaboration environment, we use a setup similar to the IN2CO (Intuitive and Interactive Collaboration) framework described by [38] that is been enhanced in order fulfill the requirements stated above. In analogy to Overview and detail views, a large display device presents an over-view of the complete data in form of a public viewport (see Fig. 3).

Supplementary, mobile devices are used to provide detailed information of task-driven aspects of the data and they also act as input and control interface. In a user study performed for a typical, simplified factory-planning problem, it was demonstrated that team members could focus on the problem-solving task itself, instead of concentrating on interaction issues. By using the intuitive interaction capabilities provided by the smart devices, focusing on the actual task at hand was made possible. A virtual representation of the data on the shared viewport facilitated communication and decision making in a team-oriented manner.

Simultaneous work development is an important aspect to support efficient real time collaboration. Likewise different aspects and interpretations of the data matter of analysis are also important. Team-members have different interests, and, consequently, the data must be shown in multiple views. The IN2CO framework combines different task models together with visualizations for a shared public view on a large display devices and interaction and visualization capabilities for mobile devices. Elements of mobile devices are used to support particular tasks, while the public view visualization enhances the existing visualizations combined in one view. Impacts caused by changes of another autonomous team-member are considered in the public viewport, where changes of the entire system are visualized. Impacts on particular tasks are hidden in the complete system view and cannot be identified by a single team-member. Overview and Detail technique as well as Context and Focus technique are not sufficient to support collaborative work including consideration of inferdependencies. Three different views are necessary to give insights in group performance, individuals’ performance, and the visualization data. Our aim is to overcome the limitations by proposing a general framework holding three views: Simulation view, status update view, and status report view indicating overview, performance view, and detail view. The simulation view enact as Overview, holding a Virtual Reality (VR) application that generates realistic images and depictions of the processes. The status report view provides insights in individuals’ aspect of the data. And the status update view indicates individuals’ performance. The simulation view is located on a public large display device, thus all team-member are able to observe the same view and have the same base knowledge on which they can investigate. Both status views are private elements visualized on smart devices. In this way detailed information of single processes are removed from the public screen in order to not overwhelm the user with unnecessary information or even occlude more relevant information beneath.

According to van der Veer and Van Welie [39] it is necessary to include descriptions of many aspects of the task world, not just the tasks alone in order to design groupware systems. Their framework structures task models. Task models for complex situations have three different aspects: agents, work, and situation. These aspects are further decomposed into five main foci, as described in [40]:Accordingly, each focus describes the task world from a different viewpoint; tasks have specific relationships. For the design of the task-supporting tools, designers can read and design from different angles, thus assuring consistency and completeness. An overview of the aspects and relations involved is sketched in Fig. 4.

As powerful communication technology has become increasingly pervasive, collaboration between people has moved in the direction of computer-supported cooperative work, where computers are now additional actors in collaborative processes. Roles can be exchanged easily between actors, and activities can be delegated to systems [41].

To cover the design of collaborative task models and a supporting system, activities to be performed must be clearly defined. Who is performing what activities? What objects are needed? How should one represent the information to user-groups? How can one enable interaction with the systems? What are the dependencies involved? These are the most important questions that one must answer.

It is necessary to have a clear understanding of the requirements for the intended system. Services, users, environment, and associated constraints, for example, must be defined and connected. Users in the system are actors performing tasks using the task world ontology. One must include dedicated task models into the system together with rules and rights of data access and functionalities. Participants must be able to choose profiles that are connected with tasks when they register smart devices in the main system. Users can coordinate their activities and assign tasks (R1). The implemented visualization and interaction techniques are adequately realized for different scaled devices providing differing ranges of capabilities. These procure, that participants are enable to perform activities (R3). Setting up the co-located environment as depicted above, team-members share the same location and make use of a combined shared viewport on a large display device. Team members can determine others’ activities (R5) and are self-evidently enabled to communicate with each other (R2) in a natural manner. However, the postulated requirements R4 (Member can switch between collaboration styles), R6 (Member can understand the impact of changes made), and R7 (Member can make adjustments based on impact) are not ensured and will be tackled with the distribution of the user interface across different scaled devices.

Dividing the simulation view in order to enable Overview and detail techniques in one viewport is not sufficient. First, virtual reality is used for the simulation to generate realistic images and imagination of the process. Splitting the view in two parts would decrease the level of immersion and the perception of being physically present in a non-physical world [42]. Second, positioning a second view in the simulation view leads to a reduction of the visualization area.

Mobile devices enable the implementation of many interaction metaphors, leading to more natural and intuitive interaction. Tablet computers as well as smart phones offer input capabilities due to touch input mechanisms and other sensors, and we can use them as secondary displays. Tablet computers offer a big screen compared to smart phones, where the status update and status report views can be juxtaposed in a split-view. A split-view is not sufficient on a small display like those on smart phones.

We use a smart watch to re-distribute the status update view, while the status report view is shown on the smart phone, as depicted in Fig. 5. Locating the status update view on a smart watch is analogous to the usage of a usual watch, where users capture time information.

Using a real-world scenario in production control, we illustrate the impact and interdisciplinary collaboration requirements, and discuss the implemented system components in detail.

Companies in high-wage countries must be highly efficient and innovative in manufacturing to remain competitive. Market competition increased through opening economic regions, e.g., in Eastern Europe or the Far East [60]. Mass products are offered for lower prices by countries in these regions. Companies in high-wage countries have to adapt in this evolving competitive setting, as they might become obsolete and be destroyed otherwise [61]. The domain of factory planning tackles these problems. A strategy to address this challenge is a shifted focus on highly specialized, customized products [60]. By moving to such a more individual customer-oriented production setting a company follows a make-to-order processing paradigm, where the production of a parts starts with the arrival of an order [62].

Nevertheless, customers still require products with high quality to be offered at a low price and a short delivery time. For a company this means that it must be flexible in offering individual products in a short amount of time while still remaining economical [63]. High flexibility within a production process makes necessary a concentration on production planning and control. A production process can be planned but only in rare cases the production is planned. Events that lead to such a deviation within the production process, are, for example, machine breakdown, missing parts or manufacturing of unusable parts/products [64]. In such circumstances, a planned optimal production process cannot be realized.

A company must react as quickly as possible to events (new customer orders or deviation from a planned optimal production process) to minimally deviate from a stated production plan. Such events cause a gap between the current state of a production process and the planned state of production. An alternative production plan must be created in this situation, leading to the concept of “production control.” Production control regulates such conditions within the order processing, i.e., it determines the sequence of sub-processes that should be executed [65]. Should changes in the production plan arise, the production control will be responsible to carry out modifications. The first step identifies the current state of production. This step must be executed rapidly to minimize, as much as possible, the deviation between the planned and the actually observed current state of production [63]. To react rapidly a continuous view of the state of production is necessary, and all related information must be continually recorded and available. Such a set-up makes it possible to adapt a production plan quickly once a disturbing event is recognized. Production control must be an automatic process to detect undesired events and adapt the production plan accordingly.

To meet these requirements an event-driven production control (EDPC) was developed. The EDPC system uses an extended bill of materials in order to shorten the reaction time in case of occurrence of an (undesired) event. The bill of materials is extended with additional information for each part (e.g., required production station, size, mass, set-up time and production time). This extension makes it possible to store information within the bill of materials that is necessary for the production control. The system approach taken by Kasakow et al. [66] uses the production of a turbo charger as an example. By placing an order, the EDPC uses the content of the order and creates an appropriate extended bill of materials. Necessary production tasks to be done to satisfy the customer order derive from this bill of materials. According to Kasakow et al. [66] have shown that the EDPC can realize an automated production control. It is possible to derive all necessary actions to be taken within a production process (e.g., creation of production orders, arrangement of the production sequence), based on a customer’s bill of materials and the information of the current state of production.

A disadvantage of this EDPC is the acceptance by users. The acceptance of a system depends on the experience of a user. Here, a user is a planner of a production. The more experience a planner has with automation errors, the more she/he wants to supervise and monitor the system. But a planner trusts automation only when it is fully reliable. The reliability of an automated system leads to the required or desirable amount of supervision and monitoring effort. A lack of reliability reduces the acceptance of an automated system [67]. In case of misbehavior or breakdown of the automated EDPC, a planner needs to have access to an uncomplicated solution to interfere with this production control to ensure a smooth operation of the production process or to carry out tasks that are not part of the EDPC (e.g., implementation of a rushed order) [66, 68]. An ideal system allows the planner to monitor the status of the current production and provides all necessary information to optimize production, to ensure a smooth operation of production.

Task model inferdependency Factory planning, as domain field of EDPC, is characterized by the parallel consideration of multiple aspects such as production resources, production process and technology, and products while anticipating uncertainty and future developments over the factory life-cycle [69]. These aspects usually result in different partial-models with specific information content (e.g., layout model, process model) and components of the factory (e.g., building, machinery, foundation, media), which need to be analyzed in combination. The different partial solutions are usually developed by various stakeholders, but typically interfere and require each other [70]. The major tasks regarding collaborative factory planning are [71]:In order to implement a real-world collaboration process, two domain-specific tasks of EDPC and layout planning were chosen, both tasks being relevant for factory planning. Two different task models were created using the task world ontology introduced in [40], included in the system: Factory layout planning and event-driven production control.

The concur task tree (CTT) diagram [72] depicts the simplification of both task models and their inferdependency. It is shown in Fig. 10 ¹. All shown tasks are further refined as task world models, formulated as tasks performed by an actor having a specific role. Tasks can use objects and trigger, or are triggered, by events, see Fig. 4. The CTT diagram shown in Fig. 10 shows a refined definition of the tasks based on the task world ontology. For better readability, the detailed and graphical notation of the task models, based on the task world ontology, is not covered in this paper. The CTT diagram depicts a simplification of the tasks that are performed for factory planning. Inferdependencies are found in sub-tasks incorporating the combination of influence and dependence between two elements, one- or bi-directional. Adjusting the model as sub-task in EDPC changes the underlying dataset for EDPC and factory layout planning, either due to changes in the simulation or manipulation of the model itself. Manipulating objects in the course of a factory layout planning sub-task has a direct influence on the production flow and simulation within EDPC, clarifying the inferdependencies between both tasks. Layout planning concerns the task of ”deciding on the best physical arrangement of all resources that consume space within a facility,” which is performed when there is a change in the arrangement of resources [73]. Improvements of the overall production performance can be achieved concerning the following parameters:EDPC provides information regarding the following parameters and provides explanations related to their interplay:Other processes and tasks also influence these parameters. Impact on material flow and overall production performance results when changes of the order are performed, but also between material flow and layout changes (two differing task models).

The dynamic model in the existing prototype combines the two task models of event-driven production control and factory layout planning. Based on the listing of improvement parameters of both tasks above, it is easy to see, that both tasks have several inferdependencies between each other, e.g., changing machine positions and paths has an direct impact on material flow, production time, transportation time, waiting time, and path utilization. Bellow we describe the information/data that is displayed in each of the views and devices for this case study.

The simulation view of the framework provides an overview of the underlying manufacturing system, consisting of the building, storage areas, machines, human resources, and conveyors, see Fig. 3, right. The smart-devices are used to control the scene and execute the functionalities in the large screen setup. The following functionalities are currently implemented for a desktop and CAVE setup, and for mobile devices. Supported functionalities include:In order to identify different users, the following roles are defined and associated with the implemented functionalities. Each functionality is associated with exactly one role, implying existence of distinct roles. One or several roles can be associated with one user/actor.

A possible collaboration session within factory planning is sketched following. Two users are jointly located in the CAVE-System and register with the system, as depict in Fig. 15:

One user is assigned to the task of layout planning (further called planner) and associated the roles of Factory layout planner basic, Manipulator, and Collaborator; the other one is assigned to the task of production control (further called controller) and is associated with the roles EDPC basic, Manipulator, Collaborator, and Navigator (R1). The joint goal is to find a layout of the facilities interior in which the resources are optimal arranged and the production program is optimized. While the planner has information of spacial constraints and environmental conditions like illumination, temperature, etc., the controller has information about the machining sequence, production program, and inventory. For both users it is important to find the shortest paths between the single machining steps. Both users are able to move machines and the controller can initiate product orders further more (R3). The orders are visualized in the simulation view, which enables the user to track the material flow with the current layout. It can be easily identified if machines have to be rotated or if machines are not used and should be probably removed. At any time, production parts can be missing, leading to waiting-phases of the product in the machine. In the simulation view it can just be recognized, that a product is in the machine, but not if it is getting produced. The controller gets a haptic feedback on the watch, if parts are missing. With a quick sight on the watch, he can identify problems in the processing and reorder parts to counteract the potential delay (R4) (R7). Both users could detect, that waiting products in machines blocks the machine. If following orders are delayed too, the machine describes a bottleneck initiating a demand of action. In joint discussions, in which the ideas and thoughts of both users are considered (R2) (R4), they decide to install a waiting area next to ”bottleneck-machines” and an alternative conveyor system, so that missing parts can delivered to the machine. The planner performs the installation of the new resources, which is commented and observed by the controller (R5) who can track the impacts of this changes as positive or negative influences on the production program on the watch (R6).

The method of property checklist is a structured way to do an evaluation, in which the expert goes through a checklist of design goals for different product properties. In our case, the product properties correspond to the seven postulated requirements in chapter 3. As described in Sect. 5.4, all requirements are realized and the users are facilitated to perform collaborative work in an efficient and natural manner.

Nestler et al. [76] proposed a qualitative assessment approach used when a comparative evaluation cannot be performed, and when baseline efficiency values/rates cannot be used to benchmark the system. No other existing setups could serve for benchmarking when less functionality and interaction capabilities are provided. If one were to compare the usability of the proposed multi-modal interface with the usability of the earlier EDPC system, the new approach would be inferior. Therefore, the evaluation method of Nestler et al. is a viable alternative, based on a reliable questionnaire for assessing technology acceptance. According to Nestler et al., general considerations useful for qualitative usability evaluations are: (1) most usability problems are detected with three to five subjects. (2) It is unlikely that additional subjects reveal new information. (3) Most severe usability problems are detected by the first few subjects. Due to the limited number of available experts in our domain, we involved three participants for the qualitative usability evaluation.

In close collaboration with three EDPC experts, we performed an experimental user study. The experts performed the collaboration session as described in Sect. 5.4. Each expert performed the experiment three times—two times with two participants involved, and once with three participants involved, leading to four experiments. The number of participants was not sufficient to gather significant results concerning efficiency or effectiveness of the prototype. Nevertheless, we were able to gain insight into general users’ experiences with our system. We measured a general usability score U as introduced in [76]. After performing the experiments, we solicited expert feedback using questionnaires as proposed in [76] concerning four categories, see [75]: (1) ease-of-use; (2) user satisfaction; (3) usefulness; and (4) intention to use of our system. For the qualitative evaluation the interviewer used open-ended questions and did not interrupt the subject. The aim of the interview was to discuss all perceived problems with the subject in order to detect usability issues.

According to [77] the qualitative results of the assessment provide a performance quantification basis, resulting in a scalar usability value U. The usability categories are adjusted on a three-point scale: (a) positive comment (1.0); (b) neutral comment (0.5); and (c) negative comment (0.0). The mean of the values is calculated. We obtain a quantitative rating of all categories on a scale from 0.0 to 1.0. Moreover, the usability categories are weighted, summing up to 100, to express the importance of a category in a specific application. Formally, the usability score is calculated by multiplying all weights w(s) with the quantitative scores v(s), with the value of U being between 0.0 and 1.0:

During the experiment execution, a brief look at the watch provided insight into the production process, allowing one to determine whether actions were needed to interfere or not. At the same time, the team-members could observe the material flow in the simulation view, detect potential bottlenecks, and make suggestions for an optimized factory layout. As both task models of EDPC and factory layout planning are integrated in the prototype, the experts were able to adjust the layout and track the impact on production on the watch (at a glance), and establish priorities and initiate re-ordering of parts with the smart phone application. It was simple to stop/pause the simulation view and balances the current outcome. Furthermore, the experts could leave the physical setup and track changes on the watch for monitoring purposes due to the WIFI connection with the server and database system.

(3) Usefulness The system supported the users to perform both group and individuals’ tasks of both use cases and facilitated the recognition of impacts and the cooperation between the users. We have identified several questions that could be answered by using our system in this specific application domain. Example questions are:The usefulness of the system for the specific use case and for collaboration activities is verified.

The resulting quantitative scores and weights leading to a usability value \(U = 0.902\) are summarized in Table 1:

The qualitative and quantitative assessment results shown are very good. It is possible to conclude that our framework leads to a more efficient and successful collaboration in the case of factory planning. Due to the modular concept of our framework, different scenarios and task models can be integrated, facilitating collaborative work and decision-making in other domains.