CGIV2020 Conference, Sep 2020, Victoria University
by Tim Law, Daniel Lam and Juliana Endang
Abstract— The advent of office workstations into architectural offices introduced CAD (Computer-Aided Design) into the workflow largely that disrupted hand-drawings, which had been the traditional method of design and communication for millennia. With Virtual Reality (VR) becoming mainstream, the architectural industry is poised for a new paradigm shift in the process of translating spatial thinking to a form of representation and communication. This paper presents contemporary VR technology as a viable tool to improve efficiency from the conceptual phase of architectural design, where the sense of scale and presence unique to VRAD (Virtual Reality Aided Design) can help designers make important late-stage decisions early. While VR in architecture has been more actively discussed in the context of project management, education, collaboration and communication/presentation, VR technology being used as a drawing tool is somewhat new and with few precedence. Thus, an experiment was conducted with architectural students to compare their sense of spatial acuity between two conventional modes of drawing against VRAD. Students were brought to observe a room, and then to 3 stations to replicate their observations; using as their drawing tools: (1) hand drawings, (2) 3D modeling by an operator, and (3) manipulation of walls in VR. Our study revealed that, at the VR station, the sense of proportion was most accurate and the level of confidence was highest despite distances consistently being underestimated. This proves great potential that VRAD can help designers accurately assess spatial quality, allowing architectural design decisions to be confirmed earlier in the design phase with much greater confidence, reducing required changes during the construction phase.
Keywords- Virtual Reality, CAD, spatial perception
I. Introduction
Virtual reality (VR) has the potential to generate a deeper understanding of unbuilt spaces thus enhancing the act of design. However, instead of exploring how VR could improve on an existing design, the authors have been deliberating if VR could fundamentally change the mode of drawing, by integrating it into the early stage of conceptual architectural design. This is considered in the context that in the architectural firm, where hand-drawing had been decisively replaced with 2D CAD (Computer-Aided Design), and that itself has nearly been superseded with 3D CAD or BIM (Building Information Modelling).
This paper contemplates how spatial perception is affected by an inside-out design experience in a 1:1 virtual environment, compared with an outside-in modelling method where the drawings are created at a much smaller scale to the user. The use of conventional outside-in design approaches, i.e. using plan/section/elevation views through third person perspective, rely on spatial assessment based on numerical cognition, as dimensions are used for accuracy, and spatial perception as it is perceived in the trained designers’ mind.
Since VR would afford an accurate internal viewpoint that would allow a virtual space to be referenced anthropometrically, the authors hypothesised that the estimation of space in VR could be more accurate than traditional modes of spatial representation, such as hand drawing, 2D- and 3D-CAD. If this were true it would make VRAD a viable design tool for internal designs which demanded ergonomic accuracy. This could be beneficial to inexperienced designers, who, in this instance, would manipulate the model whilst being inside the model, without having to transpose the model to conceptual space within the designer’s mind.
With the ability of designing from a first-person perspective combined with the depth perception that VR technology provides, a sense of presence and appreciation of space that is unique to this design methodology is created. Considering the spatial immersive experience and ability to work in real time that are offered by VR technology, the authors see remarkable potential for VRAD (Virtual Reality Aided Design) to benefit designers in their early architectural design phase.
II. VR as a Drawing Tool
Until the introduction of axonometric projections in the late 1850’s, the architectural design process relied solely on the 2-dimensional plan and section (Carreiro and Pinto, 2013). The reliance on these 2D drawing types has continued throughout architectural history. Despite the rapid development of computer aided drawing technology and even the development of BIM workflow providing the vehicle for parametric design (Quirk, 2012), the drawing process remained mostly limited to the 2 dimensional surface of computer screens.
Drawing in a virtual environment provides the potential to break through the limitations of 2D display, allowing the designer the opportunity to work in a fully virtual environment at 1:1 scale. This has the potential to generate a deeper understanding of unbuilt spaces, thus potentially enhancing the mode of design (Carreiro et al., 2013).
The concept of shaping an environment in VR is not new, and an early concept was explored in the 1990’s (Donath and Regenbrecht, 1996). Virtual environment allows architectural designers to work in space, no longer limited to the orthogonal surfaces of the screen or paper, which improves efficiency and allows flexibility to project thoughts directly into space (Carreiro at al., 2013). This suggests that 3D drawing in a VR environment for architecture design has potential to improve the efficiency and flexibility of the early design process by translating design ideas directly into 3D representations.
“To sum up, new tools emerge for 1:1 scale vision of a wholeness design, as if one has access to the architect’s mind. However, its ability to change the modus operandi of the creative process, surpassing the sketching and two-dimensional outlook is yet unknown. Still, one can ask: does digital offer new advancements to already existing tools, pushing up reorganization, or is it completely surpassing the already existing tools?” (Carreiro at al., 2013, p34)
It was from this point that the experiment was proposed to compare the accuracy with which a mental image can be represented in VR (as seen in Figure 1) as compared with other methods of drawing.
III. Methodology
The experiment was designed to test the accuracy with which a mental image of a space can be replicated on different modes of drawing by a designer. To standardise the particular mental image, the conductors selected a specific observation room (Figure 2), in which each vertical plane had been identified and measured.
Prior to conducting the experiment, an ethics application was sought through the regular university process, and risks such as photosensitive epilepsy, motion sickness and trip hazards (over the VR tether) were identified and mitigated against in the selection of participants and conduct of experiment.
First year students were invited to participate from the Bachelor of Architecture at the University of Tasmania. They were identified as the ideal candidates as they would not have yet developed strong inclinations toward particular design drawing tools, thus minimizing bias toward any particular drawing methods. These participants were not allowed to precisely measure up the room, but were encouraged to move around the room freely to experience and understand its scale and size.
Each participant would then be required to reproduce the room by drawing in the following three methods: hand-drawing on paper (Figure 3), 3D modelling on SketchUp (Figure 4) and VRAD (Virtual Reality Aided Design, Figure 5). The starting point and sequence for each participant were shuffled to minimize any influence caused by participants’ previous drawing stations. These three methods were implemented due to their similarities to recreate measurable scaled output drawings based on the spatial experience of the participants, which could be compared against actual dimensions.
Participants at the hand drawing station were expected to complete the task without assistance due to its simplicity. They are provided with a pencil and an eraser, which is to be used on a piece of paper with a square grid (Figure 3). The room was to be replicated in plan view to scale. The completed drawing was then measured.
At the SketchUp station, participants were tasked to produce the room in full scale within the 3D modelling software. Inside the software environment, the task began with a template containing a human avatar as a scale and a roofless box with five basic planes representing the floor and four walls of the space (Figure 4). During the drawing process, the participant was to give instructions to the appointed conductor who drove the software on the participant’s behalf. This was to avoid a situation where the participant felt compelled to accept a result because of a lack of software proficiency. There were no limitations to what instructions could be given, however, most instructions that were taken would only involve just basic SketchUp tools (i.e. push and pull tools). After the participant left, the completed drawing was dimensioned and recorded (Figure 7).
As for the VRAD station, participants utilized the HTC Vive headset and controllers to produce the space within the Unreal Engine 4 (UE4) environment in 1:1 scale. The “Editor” mode in UE4 used a user interface system that is intuitive and very similar to the Gumball interface in Rhino 3D modelling software. The accuracy and scale of the VR Editor in UE4 were tested against reality by holding the two controllers apart and measuring them both in reality as well as in VR. Sight lines, vanishing points and the field of view were tricky to verify, but the participants who used the VR headset agreed that the perspectives in the VR environment were accurate and seemingly realistic.
During the experiment, a 2m x 2m room template was framed using planar and cubic objects within the VR environment (Figure 8). Each of these vertical planes, which were identified as walls in this experiment, can be moved around by the participant and still intersect one another without gaps. Once the task was complete, the experiment conductors would generate a plan view of the finished space for dimensioning and recording (Figure 8).
At the end of the experiment, participants were to fill out the post-experiment questionnaire to provide a qualitative assessment based on their experience of using VRAD. Statistical analysis was undertaken and graphed in R.
IV. Discussion
A total of 19 first year architectural students participated in the experiment. Upon completion of the experiment, the measurements were resolved to 0.1m, and collated for each station and compared to the actual measurements in Table 1.
| Wall | A | B | C | D | E | F |
| Length(m) | 5.8 | 6 | 3 | 0.8 | 2.8 | 6.8 |
The time taken by tasks were reflected in boxplots in Figure 9. In the hand-drawing exercise where the field of view of the user is traditionally from a third person perspective, the participants spent the least time in reproducing the floor plan of the test room. There were two possibilities that accounted for this: (1) participants already had experience using pencil and paper as a drawing tool, and (2) there was less decision making in drawing single lines on a flat 2D surface, and negligible amount of time making changes. It was noted that in the time spent doing this task, only one drawing (i.e. floor plan) was produced. Observations from this task showed that participants used numerical cognition (i.e. memorising an estimated length of each wall) in doing this task.
The test on SketchUp as a drawing tool was, on average, the second fastest test for participants to complete. Unlike the hand-drawing exercise which produces a single drawn outcome, the 3D SketchUp model can be used to generate floor plan, elevation, section and perspective drawings. In terms of productivity, this is more efficient as the extra time taken to generate the model potentially allows multiple drawings to be generated subsequently. The SketchUp operator (one of the experiment conductors) observed that similar to the hand-drawing experiment, most participants used numerical cognition in replicating the room into the output drawing. It was observed that only 2 out of 19 participants used their sense of scale using the human avatar on screen and size of the 3D model as a means of spatial interpretation. The rest of the participants gave numbers to the experiment conductors as instructions to input dimensions into the 3D model. Although participants were allowed to change and pan the view into different perspectives inside SketchUp, all participants chose to stay in the same viewport as when the SketchUp software started.
Participants spent the most time in completing the drawing task using VRAD. This may be due to: (1) VRAD being an unfamiliar drawing methodology for the participants, and/or (2) the novelty to the immersive aspect of the virtual environment was itself an incentive to stay and experience the environment longer. Although VRAD can be set to use a third person perspective (i.e. scaled up/down or looking in bird’s eye view), the field of view in VRAD for this experiment was restricted to first-person perspective (inside-out) and the scale was set to 1:1 mode. This was done to create a drawing environment unique to the VRAD test while participants were still allowed to walk around. VRAD uses an intuitive method for manipulating the test space, but it still requires initial instruction and time to learn how to use it. Akin to SketchUp, VRAD drawing can also generate floor plans, elevations, sections and perspective drawings with the additional benefit of allowing a 1:1 scale immersive walkthrough. The use of this design platform therefore has the most potential for multiple outputs, which reflects an efficient use of the extra time taken to generate the initial model.
Figure 10 shows the lengths of walls that have been normalised (divided by original wall length) and colour coded by each drawing method. Between the wall lengths, it can be observed that Wall D has the highest error margin for all drawing modes, which could stem from it being the shortest wall length (at 0.8m). Figure 11 plots the frequency distribution of differences, showing general conformity between the observations and a normal distribution curve, on which basis, a valid t-test can be taken on. Paired t-tests were undertaken, revealing statistically significant differences (p-value less than 0.05) when comparing hand-drawing and VRAD (p-value = 5.1 x10-7), and between Sketchup and VRAD (p-value = 0.02). Of interest is that if wall D is isolated, the null hypothesis (that there is no statistically significant difference between drawing modes) cannot be rejected when adopting a p-value <0.05, the observations are: hand drawing vs VRAD (p-value = 0.19), Sketchup vs VRAD (p-value = 0.65). In short, there were differences between VRAD dimensions and those of other drawing modes that could not be simply a random event.
In Figure 12, the frequency distribution of each wall is plotted, with Wall D partially omitted by assigning it a dashed linetype. With VRAD, the very similar frequency distribution for 5 wall dimensions (A, B, C, E and F), and the narrow standard deviation of mean values indicate a consistency with which wall lengths are perceived in VR space by different users, something that we do not observe with the other modes of drawing. With hand drawing and Sketchup, it is common to see multiple peaks (multimodal distribution), indicative that certain perceptual biases commonly occur across individuals. In contrast, with VR, there is some evidence that when a user is placed in virtual space, there is a consistency with which the relative lengths of walls are perceived against one another.
Through these experiments it was found that wall lengths were consistently underestimated by an average factor of 0.82. When compared to other modes of drawing in Table 2, it can be said that hand drawing is more accurate and less precise, whereas the converse occurs with VRAD being more precise and less accurate, the difference in terminology being illustrated in Figure 13. We hypothesise that this is because VR technology has the advantage of enabling a first person presence with immersive sense of depth in an interior space, however, the tethered headset and fear of colliding into a physical wall when one is physically “blind” imposes a restraint on new users from moving too freely.
| A | B | C | E | F | AVE | S.D. | |
| HAND | 1.05 | 0.92 | 1.08 | 1.01 | 0.96 | 1.00 | 0.065 |
| SKP | 0.93 | 0.87 | 0.97 | 0.89 | 0.89 | 0.91 | 0.040 |
| VRAD | 0.80 | 0.85 | 0.80 | 0.79 | 0.88 | 0.82 | 0.037 |
To test this hypothesis further, the ratio of the overall width and depth of the room is compared in Figure 14. Although VRAD has a narrower distribution (better precision) than the other modes, the underestimation is not consistent, with the shorter dimension (width of room in relation to the door) being underestimated more than the depth of the room. It might be noteworthy that in the experiment, the VR space places the user in the same orientation, facing the “back” of the room as they would have experienced entering through the door. This presents an interesting conundrum and could suggest that we perceive, or remember, depths and widths differently in a space. In comparison we see that the SketchUp median coincides with the actual ratio of those two wall lengths. At this station, participants all preferred the external perspective view to an internal perspective view, and that could be a more exact way of judging proportions.
The results of the experiment survey, seen in Figure 15, show that the majority of the participants initially felt a lack of competence in using VRAD. However, they also felt more confident in their drawn outcome using VRAD. In addition, most participants expressed an interest in applying the use of a well-established VRAD environment into the architectural design process. These survey results therefore display the presence of a potential market who are willing to explore VRAD as a design tool.
V. Conclusion
The results from this experiment consist of two data sets: quantitative data obtained from the measurements of the experiment tasks, and qualitative data from the participant survey.
The quantitative data shows that all participants undersized the room using the VRAD drawing method. However, the fact that room was undersized using VRAD could arise from flaws in the experimental design, or may relate to discrepancy in the display within the HMD, or the graphics within the software used.
The qualitative survey data reflects a sense of confidence in the design outcome using VRAD, despite the fact that all test subjects actually undersized the test room using this drawing method. The fact that subjects felt more confident about the outcome using VRAD likely reflects that the method of spatial visualisation in VRAD more closely aligns with how they are used to interpreting space, that is in a first person perspective. In terms of spatial interpretation, the participants had enough experience working in 2D plan view to intuitively ‘know’ that their drawing matches with their mental image, or as in this case a real space.
From the results of the experiment it is possible to say that the initial hypothesis is at least in part supported, as the first person perspective provided by VRAD is facilitating the subject’s ability to assess spatial qualities in terms of proportion, which bolsters spatial confidence thus enabling a designer to proceed with other design decisions.
Due to the limited sample size of the test group, this experiment was not intended to be conclusive. However, the experiment does provide a proof of concept that there is merit in the precision with which space is perceived in VR as opposed to other modes of drawing. The hope is that with further research and development of the software, architectural education and the wider design industry will eventually see a new radically different method of design.
VI. Acknowledgements
This project was the outcome of the fifth-year architectural unit “Advanced Design Research” at the University of Tasmania. The experiment and report that underpin this paper were undertaken by a group of industrious and determined students: Simon U’Ren, Wei Ma, Juliana Endang, Raymond Wong, Daniel Lam. Co-supervision for this project was offered from Dr Winyu Chinthammit (Human Interface Technology Laboratory, UTAS) and Mr Peter Booth. The experiment was conducted under UTAS ethics application H0016599.
- Carreiro, M. and Pinto, P. (2013) The evolution of representation in architecture. Future Traditions. University of Porto, Faculty of Architecture (Portugal), pp. 27–38.
- de Kok (2015) Are you Confusing Precision for Accuracy? https://www.linkedin.com/pulse/you-confusing-precision-accuracy-stefan-de-kok/
- Donath, D. and Regenbrecht, H. (1995) ‘VRAD (Virtual Reality Aided Design) in the Early Phases of the Architectural Design Process’, in The Global Design Studio: Proceedings of the Sixth International Conference on Computer-Aided Architectural Design Futures. Singapore: National University of Singapore (CAAD Futures), pp. 313–322.
- Quirk, V. (2012) A Brief History of BIM, ArchDaily. Retrieved from: http:// www.archdaily.com/302490/a-brief-history-of-bim/ (Accessed: 18 April 2017).