Reflections on the role of the ‘users’: challenges in a multi-disciplinary context of learner-centred design for children on the autism spectrum

Technology design in the field of human–computer interaction has developed a continuum of participatory research methods, closely mirroring methodological approaches and epistemological discussions in other fields. This paper positions such approaches as examples of inclusive research (to varying degrees) within education, and illustrates the complexity of navigating and involving different user groups in the context of multi-disciplinary research projects. We illustrate this complexity with examples from our recent work, involving children on the autism spectrum and their teachers. Both groups were involved in learner-centred design processes to develop technologies to support social conversation and collaboration. We conceptualize this complexity as a triple-decker ‘sandwich’ representing Theory, Technologies and Thoughts and argue that all three layers need to be appropriately aligned for a good quality ‘product’ or outcome. However, the challenge lies in navigating and negotiating all three layers at the same time, including the views and experiences of the learners. We question the extent to which it may be possible to combine co-operative, empowering approaches to participatory design with an outcome-focused agenda that seeks to develop a robust learning technology for use in real classrooms.


Introduction
Participatory design (PD) of educational technology encompasses a range of approaches to inclusive research that has developed within the field of human-computer interaction (HCI) over the past 30 years (Coleman et al. 2012). Here we define inclusive research broadly and in the context of the specific call for papers of this special issue i.e.: . . . as research that seeks to involve those who tend to be the subjects or objects of research, such as learners, practitioners or parents, as agents in the conduct of research; it addresses issues that are important to them and includes their views and experiences. Such inclusive research tends to have a practical agenda of improving educational experiences as well as being concerned with democratization of the research process. (Special Issue Call for Papers) With the continued and current interest of how children use technologies in their everyday lives (Rideout, Foehr, and Roberts 2010), including how to support their learning both within and outside of school (Clark et al. 2009), the role of children within technology design processes has received increasing research attention (Druin 2002;Guha et al. 2005;Frauenberger et al. 2012). This increase in attention has also included greater involvement of stakeholders in such design processes, including multi-disciplinary research teams as well as teachers, other professionals and practitioners, parents and, of course, children themselves (Porayska-Pomsta et al. 2012). Thus, PD as a methodological orientation has moved well beyond the field of HCI to encompass a much wider range of views and disciplines, including education.
It is within this context that we aim to illustrate, and critically reflect on some of the methodological challenges within our own work which has focused on a particular group of learners: children on the autism spectrum. Our aim is to position this work within the context of understanding what inclusive research may look like in education, whilst at the same time delineating where the boundaries of inclusivity may lie. Specifically, we position our approach as 'learner-centred design (LCD)' (Soloway, Guzdial, and Hay 1994) and explore the challenges of trying to achieve this within a multi-disciplinary context. Firstly, we distinguish between different conceptualizations of how children have been included in technology design research, and draw some parallels between this and other discussions within the inclusive research literature. We then move on to discuss the conceptualization of our own work with children on the autism spectrum, before illustrating some of the challenges from a recent multi-disciplinary project.
Conceptualizations of designing technology with and for children Nesset and Large (2004) review the theories and applications of how children have been involved in informational technology design processes. They note that usercentred design (UCD) was the main approach used by technology developers through the 1980s-1990s as technology started to become more available to mainstream markets. UCD is characterized by potential users being brought in at a late stage of the technology development process to provide testing or evaluation of products that are ready for the market. As Nesset and Large (2004, 141) note: 'All research in UCD is focused on the impact of the technology on the child user'. Whilst this can be useful for designers in terms of including large numbers of children as 'testers' and in developing products swiftly (Nesset and Large 2004), the main difficulty with such an approach is that this leaves little room for end-users to influence design for the better. This means that the power relationship between designers and users is located firmly with the former; technology designers have already made most of the decisions and can choose to what extent they consider, or act, upon users' views (Scaife and Rogers 1999).
When the potential market for the end consumer is children, there was recognition from some researchers (Scaife and Rogers 1999;Druin 2002) that traditional approaches to UCD were not very effective in developing appropriate and enjoyable technologies, leading to the suggestion that children should be included earlier in the design process. As well as the end-product not being very effective in meeting needs, there was also recognition that the methods used in UCD approaches may not be very accessible or meaningful for children (Nesset and Large 2004). For example, UCD methods may include tasks where children are observed, or questionnaires where children are asked about likes/dislikes of the technology, and these may be not be very engaging for children or easy for them to understand. The growing awareness that children could provide insightful inputs into technology design in the early to mid-1990s (summarized in Scaife and Rogers 1999) led to the development of PD approaches, which were intended to include children as design partners rather than as design evaluators in technology development and research. This meant a shift in the positioning of children within the research with a movement towards greater sharing of power and control over what was developed and why. Scaife and Rogers (1999) describe the importance of the relationships between members of the research team, including children, and how these relationships vary between different members. This creates complexity about whose voices contribute at which points in the process, requiring 'weighing up and integrating the different contributions' by the research team (Scaife and Rogers 1999, 31). Ultimately, whilst children are involved in the early stages of the design process, it is the adults in the research team who make the decisions. Scaife and Rogers (1999) called their approach 'informant design', recognizing that different people -including children -can act as useful informants at different stages of the design process because they can offer unique perspectives on different stages of the research problem. So, for example, teachers can discuss learning needs of their pupils as well as difficulties they may have in teaching particular concepts or ideas; whilst children can provide helpful insights into how learning can be motivating and fun and these ideas can be incorporated early on. Nesset and Large (2004) argue that informant design is positioned between UCD and PD perspectives because whilst children have greater involvement in the process compared to UCD approaches, their involvement is nevertheless planned and organized by the adult researchers. Druin's (2002) approach was to try to embed a more equal partnership with children in technology design moving beyond the child as user, tester and informant, to include them as design partner. She provides a detailed account of the historical developments of the inclusion of children in technology use and design, arguing for the importance of children's more equal involvement because: While a child cannot do everything that an adult can do, they should have equal opportunity to contribute in any way they can to the design process . . . They too have special experiences and viewpoints that can support the technology design process that other partners may not be capable of contributing. (Druin 2002, 19) Druin (2002) defines her own work in this area as co-operative inquiry, drawing upon collaborative design approaches in Scandanavian workplaces in the 1970s which recognized the inherent 'messiness' and complexity of such places and, therefore, the need to include many stakeholders in the design of appropriate technological tools. Co-operative inquiry essentially seeks to implement: ' . . . a new "power structure," in which neither adults or [sic] children are completely in charge. Both must begin to work together toward common goals ' (22). This means planning together the work to be done, with children and adults devising questions, collecting data and developing and testing new prototypes. Methods used to support these processes have been adapted from more traditional PD approaches to make them more 'child-friendly', for example, children (rather than adults) using video to make observations of other children in the contexts where technology is being used; taking notes using video, photographs, words and drawings rather than relying on words or transcripts of discussions only; using 'low-tech' prototypes (e.g. using lego, drawings, cardboard mockups) to illustrate what the eventual technology prototype could be like.
Such methods are not the unique preserve of co-operative inquiry, however; for example, Neale, Cobb, and Kerr (2003) discuss the development of a 'toolbox' of methods, including a combination of 2D and 3D prototyping methods as important for PD with users with intellectual disabilities and adults on the autism spectrum. Similarly, Brown et al. (2011) describe a range of methods, including the use of pictorial story-boards, and observation and feedback from adults with learning disabilities, in a 'co-discovery' process. Co-discovery encourages pairs of users to explore unfamiliar technology together and give feedback during use rather than reflecting on experiences and views after using the technology. Nevertheless, it was the researchers in each case who made the decisions about development through a synthesis of the so-called 'expert' and 'user' views from the range of methods employed.
Thus, the main difference in the design approaches including children briefly described above is not necessarily the methods used to elicit children's views and experiences, but rather the timing at which children's views are sought and the relationships with children through the project over time. That is not to trivialize the impact of these differences on the process, experience and outcomes for children, but rather to emphasize that it is when and how the methods are deployed that tends to characterize particular kinds of approaches to inclusive PD. Notably, then, in HCI research, there has been a continuum of children's participation along which researchers have travelled that closely mirrors discussion and examples in other fields (Beresford and Evans 1999;Kellett et al. 2004;Porter, Parsons, and Robertson 2006;Thomson and Gunter 2006;Lewis et al. 2008;Fleming 2013;Nind and Vinha 2012). The thread that weaves through such accounts, whether explicitly or implicitly, is the extent to which empowerment of the 'users' takes place in relation to decision-making, contributions, confidence and relationships. Rodríguez and Brown (2009) characterize this as participants moving from 'voice' to 'agency', with a concomitant shift from the 'expert' position of university researchers to a more shared notion of power and decision-making in research.
Similarly, along the continuum of approaches discussed in HCI research -defined by Druin (2002) as the role of the child shifts from user, tester, informant to partnerthere is an increasing emphasis on empowerment, either as a targeted and intended outcome or beneficial corollary of involvement. For example, Bleumers et al. (2012) discuss the idea of digital empowerment in relation to supporting social inclusion of marginalized groups through their use, as well as co-design and co-development, of technologies. This recognizes the importance of power and involvement in inclusive design, and in societal inclusion more broadly, and aligns with the views of Abascal and Nicolle (2005) who emphasize that social and political will is required for inclusive design in HCI to be socially and ethically aware. Their argument focuses on accessibility considerations for disabled people and suggests that to promote accessibility and inclusivity of technologies, disabled users should be involved from the 'first steps' (486) of development in order to aim for more universal design.
Extending this idea further, Harrison, Sengers, and Tatar (2011) argue for a 'third paradigm' in HCI research based on standpoint epistemology, which draws upon feminist philosophies of research to argue for the importance of understanding 'embodied meaning-making' and the 'social, cultural, and physical situatedness of both users and analysts' (385). However, in contrast to Abascal and Nicolle's (2005) argument aiming for universal design, Harrison, Sengers, and Tatar (2011) suggest that more local, nuanced approaches to design are needed that provide detailed insights into the needs of particular users. In other words, their argument about inclusive design is for particularization rather than universality. Specifically, they suggest that: 'Since all possibilities cannot be designed for, one strategy is to design an interface with respect to its intended embodied location' (388). Harrison, Sengers, and Tatar (2011), therefore, argue against methodological determinism and in favour of 'situated knowledges' that require sensitivity, awareness and flexibility of a range of possible methods, and epistemological assumptions, in order to make methods 'locally meaningful' (391). Furthermore, they call upon researchers to be reflexive and critical in their approaches to their research as well as in reporting processes and outcomes. Specifically, they ask that researchers in HCI report their 'intellectual and political commitments' within particular projects, and describe and justify the methods used (392). Similarly, but writing from a different (non-HCI) context, Nind and Vinha (2012) emphasize that 'doing research inclusively' encompasses a diversity of approaches and that there is no one way of doing it; instead they suggest that it may be ' . . . productive to work flexibly and reflexively with an expansive vision and various models' (7). It is within this spirit of transparency and reflection that we now turn to discuss our own work involving children on the autism spectrum in the design and development of innovative technologies. We position our own work as 'LCD' and this is discussed next before illustrating the challenges that we faced with some examples.
Learner-centred design LCD characterizes a particular approach to PD with children in the sense that it is specifically educational technologies that are the focus alongside careful consideration of educational aims as well as learner characteristics (Soloway, Guzdial, and Hay 1994;Nesset and Large 2004;Good and Robertson 2006). In other words, it is not just children's views as children that matter, but also their experiences and needs as learners; in particular, focusing on the role that technology can play in supporting constructivist approaches to learning (Soloway, Guzdial, and Hay 1994). Thus, LCD is very much about providing appropriate and effective scaffolding for children's cognition through the use of technologies. Crook (1991) argued that educational computer use needed to be considered a socially and culturally constructed and embedded activity in which interactions, both within and around the computer, can be supported and evaluated. That is, consideration of the wider context is essential and this includes the pedagogic, facilitative role taken by teachers, as well as peer interaction and dialogue. Soloway, Guzdial, and Hay (1994) reflected some of this wider context in their TILT model (Tools, Interfaces, Learner's needs, Tasks) for conceptualizing the LCDapproach, placing the learner at the centre and connecting them to other factors that should guide the design of learner-centred software. For example, 'Tasks' relates to the use of coaching as a scaffolding technique; 'Interfaces' to the need to support different modes of communication; and 'Tools' should be adaptable depending on the learner's needs. This is a helpful initial orienting framework for thinking about LCD, but provides limited detail about how such factors can be achieved and how they might interact with each other. Additionally, Good and Robertson (2006) argued that whilst Soloway, Guzdial, and Hay (1994) TILT model described the features of a system designed using LCD (i.e. was product orientated), it did not describe or take into account the process of how researchers could implement the LCD approach in their work. Therefore, Good and Robertson (2006) proposed their conceptualization of the LCD process which they termed CARRS: Context, Activities, Roles, Stakeholders, Skills.
In line with Scaife and Rogers' (1999) informant design, the CARRS model recognizes the importance and value of including a range of 'stakeholders' in the design process, including children, and also the different relationships and responsibilities that members of the design team will have in relation to the project being undertaken. Again, this is a helpful framework for considering a range of influential factors, but the model (as intended) focuses only on the learner needs and activities within the technology development process, and does not consider the wider context in which design projects may take place. Notably, for example, the role of theory in guiding design activities is not included. This is an important omission, not least because the ways in which learners, and learning, can be supported varies dramatically depending on the theoretical position adopted.
For example, such differences in approaches to learning are particularly contentious in the field of autism education -the focus of the project example included belowwhere there is considerable debate about the effectiveness and desirability of learning approaches informed by behaviourism compared with those informed by other learning theories (see Parsons et al. 2011a for a summary). Moreover, there is much discussion about the specific learning needs of children on the autism spectrum and the extent to which these require special or particular pedagogic approaches (see Norwich 2005 andJordan 2005 for discussion). The aim here is not to arbitrate between, or debate, particular positions but to note that the conceptual position taken on these issues really matters in terms of where and how research contributes to knowledge in the field, as well as where and how children are involved in, and contribute to, research. This is especially noticeable when considering the position of children in the field of autism research which, compared to other fields, is relatively late in recognizing and researching the views of people, including children, on the autism spectrum directly (see Pellicano and Stears 2011 for discussion).
In technology development, there have been some recent examples of work involving children on the autism spectrum in design (Davis et al. 2010;Benton et al. 2012;Porayska-Pomsta et al. 2012;Frauenberger et al. 2013). Our own work involving children with autism in educational technology development and research within multi-disciplinary teams began over a decade ago (Parsons et al. 2000;Beardon, Parsons, and Neale 2001;Cobb et al. 2002a). We applied a 'toolbox' of methods to support children's involvement (Neale 2001;Neale, Cobb, and Kerr 2003) and demonstrated successful learning outcomes for young people on the autism spectrum (Mitchell, Parsons, and Leonard 2007). We have continued with this work in a more recent project: Communication and Social Participation: Collaborative Technologies for Interaction And Learning (COSPATIAL), which was funded by the European Commission FP7 Programme to develop prototypes of collaborative technologies to help teach children on the autism spectrum about collaboration and social conversation (Cobb et al. 2010b;Gal and Weiss 2011;Parsons et al. 2011b;Weiss et al. 2012). We draw upon our experiences from this work to provide a meta-perspective on the range of factors that can influence how 'users' are included in technology development research. We argue that the complexity of factors can be conceptualized as a triple-decker 'sandwich', which only produces a good 'product' when all layers are appropriately aligned.
The conceptual triple-decker: theories, technologies and thoughts The top slice of the sandwich represents the top-down, theory-driven influences on the development of the educational tasks. As noted earlier, this is a layer that has tended to be under-represented in previous conceptualizations of the LCD process. In the context of academic research, however, theory should be omnipresent and, in the context of multi-disciplinary research in particular, will be understood and defined in very different ways (Harrison, Sengers, and Tatar 2011). The COSPATIAL project was informed by a constructivist understanding of the learner as an active agent in their own learning and, relatedly, the importance of working collaboratively with peers to support and develop understanding, for which there is an encouraging evidence-base for a range of learners both using (Higgins, Xiao, and Katsipataki 2012) and not using (Davis and Florian 2004;Nind and Wearmouth 2006;Aronson and Patnoe 2011) technology. We were also informed by theoretical accounts of the specific needs of the learners; in this case the well-documented socio-cognitive difficulties experienced by children on the autism spectrum relating specifically to collaboration and reciprocity in behaviour and communication (McConnell 2002;Williams White, Koenig, and Scahill 2007) and the good practice of targeting such core areas of need in educational intervention (Prizant and Rubin 1999). Moreover, we adopted a position informed by the principles of cognitive behavioural therapy (CBT) that effective support for social communication needs to focus both on actions as well as thinking, and also provide a means to reflect on experiences and learning (Bauminger et al. 2009). Such approaches have demonstrated beneficial outcomes on social cognition for young people with high-functioning autism (Bauminger 2002(Bauminger , 2007a(Bauminger , 2007b. There is a close alignment between these theoretical orientations, in that the CBT principles (at least as we operationalized them, see Bauminger et al. 2009) and constructivist approaches to learning require scaffolding and facilitation from adults and/ or peers to take place as part of the learning process, and so such opportunities for facilitation need to be factored in from the start (cf. Crook 1991). The emphasis on revisiting and reflecting on experiences and learning, with facilitation, therefore required that our target learners would need to have a reasonably good level of communication and understanding i.e. that they would need to be on the autism spectrum, but without a significant learning disability. In addition, we wanted to focus on under-represented groups of children in the research literature i.e. older children and adolescents (Parsons et al. 2011b;Edwards et al. 2012), and so targeted children aged 8-14 with high-functioning autism as our research participants.
The filling of the sandwich: technologies and their affordances The middle part of the sandwich, representing the filling, relates to the type of technology being used or developed, the specific affordances that they may offer for learning and, relatedly, the nature of the learning tasks and objectives developed. It has always been our contention that technology use should never just be for technology's sake and that we must, instead, carefully consider what it is that technology may uniquely offer to the mix. For us, this means being clear about the specific affordances of the technology and how we choose to exploit those both in relation to the theory in the top slice of the sandwich, as well as the needs of the context and users represented in the bottom slice of the sandwich (see below). In the COSPATIAL project, the technologies chosen were ones that supported the use and collaboration of more than one user at the same time: collaborative virtual environments (CVEs) and shared active surfaces (SAS) (see Figure 1); given our theoretical orientation outlined above, this particular affordance was essential. Our own work focused on the CVE and so this is what we will concentrate on here.
The bespoke CVE was a 3D interactive virtual space, which could be navigated in real-time, and allow more than one user to interact with the scene, and with each other, concurrently. The use of a CVE represents a significant advance on existing published research in the use of virtual reality (VR) technologies for autism (Parsons and Cobb 2011) by moving well beyond the restrictions of one-user/one-computer configurations (Parsons, Mitchell, and Leonard 2004) and didactic learning approaches common in earlier computer-based studies (Silver and Oakes 2001;Bosseler and Massaro 2003). There are unique affordances of virtual environments (VEs) for supporting learning generally including ' . . . improved contextualisation of learning and richer/more effective collaborative learning as compared to tasks made possible by 2-D alternatives' (Dalgarno and Lee 2010, 10), as well as for supporting learners with autism specifically. For example, Parsons and Mitchell (2002) argue that the visual, navigational and interactive features of (single-user) VEs can incorporate the effective educational intervention components from both the behavioural and cognitive research traditions in autism through role-play, rehearsal and reflection in a safe space that shares some similarities with the real world, but which does not require face-to-face interaction. In addition, the level and number of non-verbal and verbal features of communication, as well as contextual cues and features, can be directly controlled and manipulated and adapted for individual needs (Parsons and Mitchell 2002).
Collaborative VEs incorporate these affordances, but offer additional benefits too, particularly that communication (verbal, spoken) and responding can be more naturalistic between users, compared to a single-user platform where interactions with a programme by a user will receive a computer-programmed, rather than human-generated, response. Moreover, and of particular relevance to the cognitive difficulties documented in autism regarding socio-communicative perspective-taking (Williams White, Keonig, and Scahill 2007), CVEs provide a virtual platform where different users share the same virtual space, but will always experience this from their own first-person viewpoint. This means that each user inevitably has a different perspective on any scene or interaction compared to other users and so this is an important feature that can be worked with. Finally, the presence of other users both within the virtual space as well as supporting a child who sat next to them in the classroom (for example) provides opportunity for facilitation of understanding from both peers and teachers (Yelland and Masters 2007). Again, given our emphasis on the importance of collaboration and facilitation, it was essential that such interactions were a core part of what the CVE technology could provide.
The bottom slice of the sandwich: thoughts and territory of the users The bottom layer of the sandwich represents the 'bottom-up' processes that capture the needs and views of target user groups as well as the contexts in which they operate; for the COSPATIAL project this was primarily children on the autism spectrum and their teachers. Previous work in the design and development of VR applications informed our approach (Cobb et al. 2002b;Crosier, Cobb, and Wilson 2002), including working with teachers and schools from the beginning of projects. We also respected and reflected the importance of the local, or 'intended embodied location' (Harrison, Sengers, and Tatar 2011, 388), using school contexts as an essential part of technology development activities (Cobb et al. 2014). The eventual final 'output' of the project needed to be relevant and useable for teachers and pupils in the real world and so this required taking into account social and organizational factors of schools (Neale, Cobb, and Kerr 2003), including existing technology infrastructure (which can create significant challenges for the use of new technologies (Newbutt 2013)).
Millen's doctoral research Patel 2011a, 2011b) describes the methods and processes used for engaging and working with children, both with and without autism, over time. Given the particular communication preferences and needs of students on the autism spectrum, supporting their involvement and feedback requires the adaptation of some PD approaches (Frauenberger et al 2013). For example, low-tech prototyping (e.g. building models from lego, card, story-boards) is commonly used in many PD approaches with children (Nesset and Large 2004) but making the imaginative leap between low-tech and actual prototype may be difficult for some children on the spectrum. Nevertheless, in line with our own epistemological positions regarding the value of including user views in the design process , we sought to place user views at the centre of our approach in order to promote their meaningful participation in the development of COSPATIAL prototypes (Parsons et al. 2011b). The following brief analysis based on a small part of one of our prototypes highlights how hard this is to do in the complex context of a multi-disciplinary project i.e. the 'user voice' is just one element to be considered amongst many.
Making the sandwich: illustrations from the COSPATIAL project As noted above, an important objective for us was that the technology developed in the project should be useful for, and useable in, school settings and so this required a good understanding of the target learners as well as the context in which the technology was to be used. To gather this knowledge, we established three levels of schools engagement in the project: (1) a core design team of five teachers from three different schools (one mainstream and two autism specialist schools) plus project researchers who were closely involved in early decisions and iterative ideas generation about how to use CVE technology for student learning, as well as later formative evaluation of prototypes; (2) three additional schools who, together with teachers from two of the schools engaged in level 1, reviewed concept design and prototypes of the CVEs that were developed (40 teachers in total); and (3) a further four schools who were involved in final evaluation studies to assess use and suitability of the CVEs for student learning. Figure 2 illustrates the different levels of involvement of these schools, the number of design activities (development and/or evaluation) that took place, and the scope for potential to influence the design and development of the prototypes at these different stages.
Notably, the learning context was very well represented by significant involvement of teachers and children in the early stages of the project. During this time, the core design team decided on two key areas of social interactions that teachers said were difficult (and important) to teach -collaboration and social conversation -following workshops to illustrate the key features of the CVE technology, as well as the main conceptual approach of the project. Consequently, early design decisions were made in a context where we were simultaneously negotiating all three layers of the sandwich within the core design team. There appeared to be a good alignment between the learning needs of the target users, the pedagogic wishes of the teachers, the specific affordances of the CVE technology in being able to support the interaction of more than one user at the same time, and the theoretical approach taken. However, tensions between the different layers of the sandwich arose as we tried to narrow down the design ideas and then develop them within the technology. Through a series of design review activities involving discussion groups and annotated visual storyboards with the core design team, 12 initial design ideas were eventually whittled down to three main concepts that we then developed to produce working prototypes. To ensure that the learning framework incorporated cognitive behavioural principles, each scenario was also described using a 'CBT template' which includes features important to our conceptual approach, as well as relevant characteristics of the users and technology. To illustrate this approach, we use one of our main concepts as an example: the COSPATIAL Block Challenge (full details of this scenario and its iterative development can be found in Cobb et al. 2010a).
This scenario concept was for a two-player problem-solving game where each player has different but interdependent objectives to achieve. The objective was for the players to work together to build a tower of blocks that had different patterns when viewed from each side (see Figure 3). Each player was presented with information about the pattern that should be viewed from their side of the tower, but no information about the target pattern for the other player. In order for both players to achieve their target patterns, they would need to communicate with each other to find out what colour pattern their partner required and they would need to work together to jointly rotate individual blocks until the correct colours were showing on each side of the tower. The CBT scenario grid for this scenario is shown in Table 1 (from Weiss et al. 2010).
This CBT grid illustrates that the main social components of collaboration that were targeted by the development of Block Challenge were mutual planning (cognitive), mutual performance (behavioural) and choosing (cognition and behaviour). Problemsolving -a specific CBT-informed learning technique (Aberson, Shure, and Goldstein 2007) -was suggested as being particularly useful for supporting the cognitive aspects of these social goals via the technology (rather than via a human facilitator). Non-technology-based problem-solving techniques usually require the child to take a meta-cognitive perspective on a social scenario or task by thinking through and generating different solutions to problems before trying them for real (Shure and Spivack 1982;Bauminger et al. 2009). One of the potential strengths of the CVE technology was that we could combine these cognitive aspects of planning and choosing with the behavioural techniques of aspects of rehearsal/practice. In other words, children The screen on the left shows the viewpoint for player 1. They can see the avatar for player 2 (wearing a green top) on the other side of the room. The target for player 1 is to build a two-block tower with a yellow block in the bottom and a grey block on the top. The screen on the right shows the viewpoint for player 2. They can see the avatar for player 1 (wearing a red top) on the other side of the room. The target for player 2 is to build a two-block tower with a red block in the bottom and a brown block on the top. The first block has been selected and is placed in the centre of the room between the avatars. The players need to rotate the block to match their respective block tower patterns.
could concurrently think about, plan and discuss the problem whilst acting it out within the CVE. In addition, children received visual reinforcement for successfully completing different levels of the game through being awarded stars on their toolbar (see Figure 4). Thus, there was a strong push factor from the top theory-driven slice of the sandwich that guided the development of this scenario. In addition, this concept was liked by the teachers and was premised on targeting key skills of collaboration and turn-taking utilizing some of the main identified affordances of the CVE (perspective-taking, non-face-to-face communication). However, in translating the concept into the CVE software, there were many instances of misunderstanding and differences in expectations leading to compromises that had to be made.
A specific example was the degree to which (and how) facilitation, to support concept clarification, should take place within the CVE. In the case of the Block Challenge activity, this was instantiated through modelling an example of a visual perspective-taking task which requires participants to mentalize what someone else can see Table 1. Description of the COSPATIAL Block Challenge scenario, using the CBT template.

Scenario description.
Block Challenge: A joint problem-solving game where each player has different but interdependent objectives to achieve. This activity revolves around two or more players engaging in a problem-solving task in a virtual environment. The aim of the task is to match the pattern on the target image to the pattern on your column by twisting and turning blocks in collaboration with the other players. This scenario could be greatly simplified in order to cater for a wider range of abilities. Players could have to build a block tower collaboratively to match a target pattern. For example, 'Build a tower of three red blocks' -the players would have access to a number of colour blocks and they must choose three red blocks and pick them up and move them together to match the target pattern.
What is the intervention population of focus or target (low functioning Autism Spectrum Disorder (ASD), high functioning ASD, typically developed)?
Medium to high functioning children with ASD, and typically developing (TD) pupils. What is the intervention focused general social goal?
Cooperation (based on Hamilton, Brindley, and Frith 2009). This was developed as a training activity to be completed prior to performance of the collaborative tower-building activity (main task). With hindsight, we realized that, whilst we had put considerable thought into the details of the main task, this was not the case for planning of the training activity. The concept elaboration storyboard included only images for the training activity (see Figure 5), depicting the idea that a two-colour 3D block would be placed in the VE and the student would move around the block to see it from different sides. The intention was that moving around the block would facilitate student understanding of how the block would look (differently) from different perspectives. However, whilst this was agreed by the core development team at the teacher workshops, implementation in the VE revealed the need for additional design decisions and revisions which were added through a series of prototype review sessions involving different participants including members of the core design team, and children with, and without, an autism spectrum condition (ASC). The first iteration of the VE training environment replicated the storyboard; the 3D two-coloured block was placed in the centre of the room and a CVE character 'Professor Blocks' was available to provide prompts (facilitation/scaffolding) about how to interact with the VE (see Figure 6). Initial review of this prototype by teachers and the typically developing (TD) children revealed that it was not immediately obvious to the user how to move around the room to see the block from different sides. To overcome this, 3D objects were placed in different corners of the virtual room and Professor Blocks provided prompts to encourage the user to move to the object in order to explore the 3D space.
Through a series of iterative walkthroughs of CVE prototypes, involving teachers and students, further design changes were made to the training environment (see Figure 4). These were useful changes to make based on user feedback but in making these changes a major tension was revealed between the different preferences of users (children vs. teachers), the desirable affordances of the CVE technology, and the implementation of theory. Specifically, the need to facilitate student interaction within the CVE, both to overcome navigation control difficulties, but also in order to guide them through the learning activity in a structured sequence, restricted use of  some of the natural affordances of the technology -in this case, free navigation to explore the 3D space. The constraint on exploration was driven by the need to focus student learning and was experienced again in the Block Challenge main activity; initially the design of the CVE allowed students to move their avatars around the virtual space to explore the blocks in the room from which to choose the block required. During testing, it was observed that the students enjoyed this feature but got distracted by chasing the avatar of the other player around the room. In order to focus attention on the learning objective, teachers requested that the free navigation feature was removed. This resulted in restricted movement control, wherein the students could only turn their avatars around to see what blocks were in the room, but not move around the room.
In addition, the guided task requires progression through a linear sequence of decisions and interactions that are repeated many times over. Whilst this constrained interactivity and repetition in task execution was beneficial for learning in less able students, the result was that we deviated from using some of the affordances of CVEs that were initially considered to be attractive in the application of this technology in education viz: free navigation and interaction with the CVE (allowing the user to make the own decisions about what they want to do and in what order) and flexibility (to allow teachers to adapt the learning experience to suit specific needs of individual learners). This resulted in two major consequences: firstly, building constraints into the VE was very time consuming and difficult and, secondly, some of the more able students were bored and frustrated by the constrained activity in the evaluation of the technology, leading to reduced motivation to continue or complete the task.
This did not mean that the Block Challenge scenario was unsuccessful (an unappealing 'sandwich'); three important affordances of CVEs were utilized in this scenario (3D visualization, perspective-taking, and collaborative interaction) and feedback from teachers during formative and summative evaluation studies was positive (Parsons et al. 2012). Rather, what we seek to illustrate is that our experience throughout the design process was messy, and became impaired by tensions arising from the conflicting requirements, and differences in expectations of these from different stakeholders. These tensions were not experienced by our design team alone; the team developing the SAS scenarios also experienced considerable tensions and 'constructive misunderstandings' (Zancanaro 2012). Thus, through a series of decisions made with good reasons via different user groups at different stages of development, we ended up with a CVE training scenario that was boring (for some) and difficult to use and seemed to jar with many of our initial intentions. Crucially also, because of the significant time investment in creating this scenario, we were then very reluctant to rethink and revise our plans.

Conclusions
This is an example from just one small part of one of the COSPATIAL prototypes and, in this paper we can only provide glimpses into the complexity of the processes involved. Nevertheless, by examining this one small part, it becomes possible to imagine how magnified and multiplied such challenges could become throughout a much larger, multi-disciplinary project like this. In the example of the training environment, we see a sandwich that might have started off looking and sounding like a good product but, in trying to make the sandwich, we ended up with fillings that did not necessarily align well with either the top or bottom slice of our triple-decker.
For example, in prioritizing the perspective, while taking aspects of the Block Challenge task that aligned strongly with underlying theoretical concepts relating to 'theory of mind' difficulties, we ended up diminishing the key affordance of flexibility (of exploration and interaction) that can be achieved in CVEs. This, in turn, led to reduced enjoyment of the use of the CVE for some of the users because their movements were constrained within the task. The feedback regarding the impact of the constraints on exploration and movement through the CVE came too late for us to make changes to this aspect of this task. Therefore, even though the bottom slice of the sandwich -the experiences and views of the users (teachers and children) -was well covered in our project design activities, this was not sufficient for avoiding problems in developing a useable end-product.
In addition, it was interesting that some of our tensions arose not necessarily between the layers of our sandwich but within one of the layers; namely, the bottom layer representing the users. In the example discussed above, we ended up prioritizing teachers' views (need for control and constraint over actions) over the children's (desire for exploration). This raises important challenges about which groups of users or stakeholders are involved in making key decisions about the design of the technology and how 'we' -as the design team -incorporate and work with those views. This challenge may be particularly sharply brought into focus when it is the LCD approach that is being attempted with novel technologies. In other words, because the development of CVE technologies for learning (generally and for children on the autism spectrum) is still a relatively rare pursuit, it is not clear (yet) who is best placed to advise on how the technology could and should be developed, and the timing at which such views should be sought. In our desire to be inclusive of both teachers and children in our LCD approach, we may have created a situation where it was more difficult to meet learning needs effectively because there was a difference in views. It could be that future projects need to specify and differentiate the 'central users' from other stakeholders and work differently with these distinct groups.
The particular challenge of which users to prioritize and when highlights that, without due reflection, it is easy to take a rather rose-tinted perspective on the involvement of users in projects like this as a way of achieving important ethical and epistemological objectives; namely, taking a more inclusive approach to educational technology development in the belief that it is the 'right thing to do' and that, as a consequence, this will result in better learning outcomes for children . Additionally, we have even claimed that by taking a learner-centred approach, it is possible to navigate and possibly ameliorate some of the tensions that arise within complex educational technology projects: agenda is an overly-simplistic claim. The aim in the LCD process is not necessarily one of empowerment for the users in the same way as Druin's (2002) co-operative inquiry approach. However, we would argue that we made attempts at greater democratization of user roles within the project, especially through being involved from the very early stages of the project (cf. Abascal and Nicolle 2005). Therefore, we were inclusivebut only up to a point; user voices sometimes were not given due prominence (and maybe at times, given too much prominence) and this leaves us with many questions for the future.
There remain significant challenges as to how the integration of user views can be more effectively achieved and there is also a need for much clearer conceptualization of the nature of the project and what is being aimed for -very much in agreement with Harrison, Sengers, and Tatar's (2011) plea for researchers to be transparent about their political and intellectual commitments. In other words, we need to be clear about the extent to which users may or may not be involved in decision-making and to ensure that they understand the reasons why/why not. Ultimately, we had a strong social justice agenda -we wanted to create something useable and available for schools, to make outputs that could make a difference to children's learning outcomes in the real world rather than simply being a research exercise. Perhaps this was too ambitious because it could be that in trying to value and include the user, whilst simultaneously aiming to produce a 'finished product' we were working on two competing or even opposing objectives. Martin and Sherington (1997) make a distinction between 'research driven' and 'development driven' projects, the former reflecting the pursuit of a research agenda and the latter focusing more on the processes of empowering people to make decisions and changes. They argue that these different drivers necessitate different relationships between the (traditionally defined) researchers and the researched, with development projects requiring more collegiate and collaborative relationships and research projects involving more 'contractual' or consultative relationships, where the power remains with the researchers. In other words, in development-driven projects, it is the relationships and processes that matter most, whilst for research-driven projects, it is the eventual outcome (answering of questions) that takes priority. Druin (2002, 19) makes a similar distinction in her research: 'With this role of design partner, the impact that technology has on children may not be as significant as the impact children can have on the technology design process'.
This reflection makes us consider that we perhaps vacillated between these objectives at different points in our project, trying to make the strong involvement of the user align with an outcome-focused agenda; making the best of both worlds. This could be where some of our tensions arose i.e. we were not always clear enough about where and how our users voices were contributing to the project and therefore when and how to prioritize them. This further leads us to question whether an outcome-focused agenda (eventual impact of the technology on children) is ever likely to be compatible with more co-operative, empowering approaches to PD. One possible 'middle way' in thinking about this could be to consider the learning outcomes that emerge through the processes of engagement and participation along the way, as well as via use of the 'learning object' that is eventually produced. It remains an open question though (and potentially, very messy territory) as to whether such objectives can be effectively and concurrently sustained. We do not know the answers but share our experiences with the aim of helping other researchers to navigate this complex arena; we hope other researchers will do the same so that the challenges we all face can be used as opportunities for exploring and understanding the methods and boundaries of inclusivity in LCD research.