Space Planning of Project-Based University Learning Spaces Using Generative AI

A. Background of the Study

Recent higher education has shifted from instructor-centered knowledge transmission toward student-centered education emphasizing active participation, collaboration, problem-solving, and project-based activities. Accordingly, university teaching and learning spaces need to evolve from uniform lecture-oriented settings into flexible and interconnected environments that support diverse learning activities and user needs.

Previous research developed a restructuring process for student-centered university teaching and learning spaces by identifying required unit spaces and spatial characteristics based on teaching and learning methods and activities, and analyzing spatial hierarchy and interrelationships. It also proposed five cluster types—Learning Support, Learning Community, Project-Based Learning, Technology-Based Learning, and Convergence Learning—with spatial relationships and prototypes for each type [1].

However, these prototypes primarily present basic spatial configurations and relationships. Further development is needed to incorporate specific planning elements, including learning activity flows, spatial connections, and zoning.

Generative AI has recently been used as a design-support tool to analyze design conditions and explore planning alternatives. This study therefore applies generative AI to the existing Project-Based Learning Cluster prototype to progressively refine space-planning conditions and examine its applicability.

B. Purpose of the Study

This study aims to progressively refine the space planning of a project-based learning cluster prototype proposed in previous research using generative AI and examine its potential and limitations as a space-planning support tool.

C. Research Method

The study focuses on the Project-Based Learning Cluster proposed in previous research. Its learning activities, required unit spaces, spatial characteristics, and spatial relationships were structured as AI input conditions. ChatGPT was then used to progressively refine the Activity Journey, spatial relationships, and zoning. The results were reviewed against the existing design conditions, and Midjourney was used to conceptually visualize key space-planning characteristics. [Figure 1]

A. Space Planning for Student-Centered University Teaching and Learning Spaces

Student-centered education moves beyond uniform, lecture-centered learning by emphasizing diverse activities such as collaborative learning, project-based learning, discussion, and practice based on learner participation and interaction. Accordingly, university teaching and learning spaces need to consider not only the functions of individual spaces but also diversity, flexibility, and connectivity in supporting learning activities.

Previous research analyzed teaching and learning methods and activities to identify the required unit spaces and spatial characteristics for student-centered university teaching and learning spaces. Seven spatial characteristics were identified: Adaptability, Openness, Flexibility, Commonality, Connectivity, Multifunctionality, and Diversity. Based on the hierarchy and interrelationships among the required unit spaces, five cluster types were proposed: Learning Support, Learning Community, Project-Based Learning, Technology-Based Learning, and Convergence Learning [1].

Among these, the Project-Based Learning Cluster consists primarily of a Project-Based Classroom, Laboratory/Practice Room, and Small-Group Seminar Space, supporting various activities such as presentation and sharing, small-group discussion, collaboration, and hands-on practice and making. Based on the characteristics and interrelationships of these spaces, spatial relationships involving Direct, Flexible, and Visual Connections, along with Single-Loaded Corridor, Double-Loaded Corridor, and Hall-Type prototypes, were proposed as shown in Table 1 and Figure 2 [1]. [Table 1] [Figure 2]

B. Space Planning Using Generative AI

Generative AI has recently been used as a design-support tool for idea exploration and planning because of its ability to generate and restructure various forms of information, including text and images [2], [3]. In particular, text-based generative AI can structure design conditions, analyze their relationships, and explore planning alternatives, while image-generative AI can visually represent design concepts and spatial characteristics [2], [4].

In spatial design, generative AI should be approached as an interactive tool that enables researchers or designers to explore and review relationships and alternatives based on given conditions, rather than as a means of determining final design solutions. Therefore, the design conditions provided to AI and the criteria used to evaluate its outputs are critical [5].

From this perspective, this study uses ChatGPT as a space-planning support tool to structure the design conditions of the Project-Based Learning Cluster derived from previous research and progressively refine the Activity Journey, spatial relationships, and zoning. Midjourney is then used as a conceptual visualization tool to explore the key characteristics of the resulting space planning. Rather than evaluating the generative capabilities of AI itself, this study examines how existing space-planning knowledge can be developed into more specific spatial plans through generative AI.

A. Step 1: Structuring Design Conditions for the Project-Based Learning Cluster

The Project-Based Learning Cluster proposed in previous research was selected for the application of generative AI. The cluster consists primarily of a Project-Based Classroom, Laboratory/Practice Room, and Small-Group Seminar Space, supporting learning activities such as presentation and sharing, small-group discussion, collaboration, and hands-on practice and making [1].

For space planning using generative AI, the required unit spaces, spatial characteristics, learning activities, spatial relationships, connection types, and prototype types identified in previous research were structured as AI input conditions. These conditions were used not to generate new space types, but to refine learning activity flows, spatial relationships, and zoning based on the existing prototype [Table 2].

B. Step 2: Development of Space Planning Using ChatGPT

The structured design conditions were entered into ChatGPT, and the existing Project-Based Learning Cluster was progressively examined through Activity Journey, Spatial Relationship Refinement, and Zoning Development. No new space types or connection types beyond those identified in previous research were added, ensuring that the space planning remained within the existing design conditions.

Experiment 1) Activity Journey

Based on the learning activities identified in previous research, changes in learner activities and spatial use throughout the project-based learning process were analyzed. The AI analysis revealed a recurring flow of learning activities among the Project-Based Classroom, Laboratory/Practice Room, and Small-Group Seminar Space [Table 3]. The results indicate that project-based learning does not occur within a single space but involves repeated transitions among discussion and collaboration, hands-on practice and making, and review and sharing. This Activity Journey was subsequently used as a basis for refining spatial relationships and zoning.

The analysis identified an Activity Journey in which discussion and collaboration, practice and making, and review and sharing repeatedly transition across the Project-Based Classroom, Laboratory/Practice Room, and Small-Group Seminar Space. This indicates that project-based learning involves the continuous use of all three spaces rather than being confined to a single space.

Experiment 2) Spatial Relationship Refinement

The Activity Journey from Experiment 1 was compared with the existing spatial relationships to examine connection types among the three spaces. The relationships were refined within the existing Direct, Flexible, and Visual Connection types without adding new ones [Table 4].

The Direct Connection between the Project-Based Classroom and Laboratory/Practice Room was supported by their high connectivity and adjacency and repeated movement in the Activity Journey. In contrast, Flexible Connection between the Project-Based Classroom and Small-Group Seminar Space and Visual Connection between the Laboratory/Practice Room and Small-Group Seminar Space were considered possible ways to support activity transitions and continuity. As these were not explicitly specified in the existing design conditions, they were further examined in the zoning stage.

Experiment 3) Zoning Development of the Double-Loaded Corridor Prototype

Among the three existing prototypes, the Double-Loaded Corridor was selected as the representative case. Its basic spatial structure and three required unit spaces were maintained, while the Activity Journey and refined spatial relationships were applied to clarify spatial roles, movement paths, and boundaries. [Table 5]

Repeated movement between the Project-Based Classroom and Laboratory/Practice Room highlighted the importance of their direct connection. The Project-Based Classroom serves as the central space for discussion, collaboration, presentation, and sharing; the Laboratory/Practice Room supports practice and making; and the Small-Group Seminar Space supports activity transitions and continuity between them.

Based on the existing criteria, Direct Connection between the Project-Based Classroom and Laboratory/Practice Room was prioritized, while Flexible and Visual Connections involving the Small-Group Seminar Space were considered as possible options. Specific physical elements were not proposed; instead, zoning was refined by adjusting boundaries and connections among the existing spaces without adding new shared spaces.

In the generative AI-assisted space-planning process, the required unit spaces and basic spatial structure of the existing Project-Based Learning Cluster were maintained, while the Activity Journey clarified recurring movements and spatial use across learning activities. Based on these findings, the importance of a Direct Connection between the Project-Based Classroom and Laboratory/Practice Room became clear, while the Small-Group Seminar Space was defined as supporting activity transitions and continuity between the two spaces [Figure 3].

Therefore, the role of generative AI in this study is not to automatically generate new spatial configurations or zoning, but to refine how the existing prototype functions by integrating the temporal flow of learning activities and spatial use relationships with the space-planning criteria derived from previous research.

C. Step 3: Conceptual Visualization of Space Planning Using Midjourney

Midjourney was used to visually explore the space-planning conditions refined through ChatGPT. Rather than generating new spatial structures or zoning, it was used as a conceptual visualization tool to represent spatial roles, relationships, and learning activities.

The existing Double-Loaded Corridor prototype was used as a reference image, and the relationships among the Project-Based Classroom, Laboratory/Practice Room, and Small-Group Seminar Space, along with learning activities such as discussion, collaboration, practice and making, and presentation and sharing, were incorporated into the prompts. Using the same reference image and generation settings, the prompts were progressively refined from spatial and activity relationships (G1) to specific functions and activities of each space (G2) and finally to stronger preservation of the existing spatial layout and adjacency (G3) [Table 6].

Of the four images generated at each stage, two representative images were selected to show how the prompt conditions were reflected and changed across stages. In G1, an interconnected project-based learning environment was represented, but functional differentiation among spaces was limited. In G2, spatial functions and learning activities became clearer, although the existing prototype structure tended to be reinterpreted. In G3, greater emphasis was placed on preserving the layout, adjacency, and central circulation. This strengthened the representation of spatial organization and relationships, but limitations remained in accurately reproducing the existing prototype structure [Figure 4].

The stage-by-stage comparison showed that more specific prompts improved the representation of spatial functions, learning activities, and relationships among spaces. However, despite repeated emphasis on the existing prototype layout and adjacency, these elements tended to be reinterpreted rather than accurately preserved. This indicates that Midjourney is useful for conceptual visualization of space-planning conditions but has limitations in reproducing precise spatial structures [Table 8].

This study examined the potential and limitations of generative AI as a space-planning support tool by progressively refining and visualizing design conditions for project-based university learning spaces derived from previous research. The required unit spaces, spatial characteristics, learning activities, and spatial relationships of the Project-Based Learning Cluster were structured as design conditions, followed by space-planning refinement using ChatGPT and conceptual visualization using Midjourney.

ChatGPT generated an Activity Journey reflecting the temporal flow of learning activities and used it to refine spatial connections and zoning. A Direct Connection between the Project-Based Classroom and Laboratory/Practice Room was clearly supported, while Flexible and Visual Connections remained possible options requiring additional design conditions. This suggests that generative AI can support the progressive refinement of space-planning conditions by relating existing design conditions to learning activities.

With Midjourney, more specific prompts improved the representation of spatial functions, learning activities, and spatial relationships. However, accurately preserving the layout and adjacency of the existing prototype remained difficult. Thus, image-generative AI is more appropriate as a conceptual visualization tool for exploring the spatial meaning and possibilities of planning conditions than as a tool for determining precise layouts or zoning.

The significance of this study lies in applying generative AI to a sequential space-planning process of structuring design conditions, analyzing activity flows, refining spatial relationships, developing zoning, and conceptual visualization, rather than using it simply for form generation. The findings suggest that generative AI can support designers in developing evidence-based space-planning knowledge rather than replacing their judgment. As an exploratory study limited to the Double-Loaded Corridor prototype of the Project-Based Learning Cluster, future research should extend the approach to other learning-space types and validate the process through objective evaluation criteria and expert review.

* This research was supported by BK21 Four Service Design driven Social Innovation Educational Research Team in Dongseo University.