AI Architecture Research: Floor Plans, BIM & Building Performance
Read 116 author-sourced research records through specific architectural questions, dates and evaluation limits.
These records summarize author-reported research, including preprints and supporting datasets. They do not establish available product features or independently reproduced performance. Publication and revision dates are distinct.
116 records · Page 2 / 5
Indoor thermal comfort management: A Bayesian machine-learning approach to data denoising and dynamics prediction of HVAC systems
RESEARCH /
The authors use Bayesian machine-learning methods to denoise data and model indoor temperature dynamics. Their problem includes inertia, ventilation effects and uncertain observations. The architectural relevance is a transparent control model, where confidence and input quality should be reviewed alongside a predicted temperature curve.
Architectural question
How should temperature predictions communicate uncertainty to operators?
Javier Penuela, Sahar Moghimian Hoosh, Ilia Kamyshev, Aldo Bischi, Henni Ouerdane
Large Language Model-Driven Code Compliance Checking in Building Information Modeling
RESEARCH /
This study links language models and Revit to interpret code text and generate checking scripts. The authors describe residential and office examples. Architectural review must separate text interpretation, script correctness and regulatory acceptance; a successful demonstration cannot establish permission for every jurisdiction or model.
Architectural question
How can a BIM code-checking workflow expose interpretation and script errors?
Soumya Madireddy, Lu Gao, Zia Din, Kinam Kim, Ahmed Senouci, Zhe Han, Yunpeng Zhang
Can Time-Series Foundation Models Perform Building Energy Management Tasks?
RESEARCH /
This paper evaluates whether broadly trained time-series models can support building energy tasks beyond bespoke models. It questions generalizability across tasks and settings. Architectural and facilities readers should inspect which tasks transfer, what calibration remains necessary and whether apparent scalability comes with meaningful error in a particular building.
Architectural question
Which building-energy tasks genuinely transfer to a foundation model?
Ozan Baris Mulayim, Pengrui Quan, Liying Han, Xiaomin Ouyang, Dezhi Hong, Mario Bergés, Mani Srivastava
Hybrid Mesh-Gaussian Representation for Efficient Indoor Scene Reconstruction
RESEARCH /
This study assigns texture-rich flat areas to meshes while retaining Gaussian representations for intricate geometry. It explores an efficiency tradeoff in indoor reconstruction. For an architectural survey workflow, visual speed and surface detail must be assessed separately from dimensional reliability.
Architectural question
Which surveyed surfaces need mesh geometry rather than visual splats?
GAN-SLAM: Real-Time GAN Aided Floor Plan Creation Through SLAM
RESEARCH /
GAN-SLAM applies adversarial methods to clean and complete occupancy maps during robotic mapping. It addresses noise that can disrupt downstream floor-plan creation. An architectural survey should distinguish observed geometry from model-completed regions and preserve a traceable link to the original measurements.
Architectural question
Which parts of a reconstructed plan were actually observed?
Leon Davies, Baihua Li, Mohamad Saada, Simon Sølvsten, Qinggang Meng
BADGR: Bundle Adjustment Diffusion Conditioned by GRadients for Wide-Baseline Floor Plan Reconstruction
RESEARCH /
BADGR jointly refines camera poses and floor-plan layouts from widely separated panoramas using diffusion and optimization signals. It addresses view consistency in reconstruction. Architectural surveying should inspect how capture density, occlusion and initial estimates affect the recovered walls before adopting dimensions from the output.
Architectural question
How many viewpoints are needed for a credible recovered floor plan?
Yuguang Li, Ivaylo Boyadzhiev, Zixuan Liu, Linda Shapiro, Alex Colburn
FloNa: Floor Plan Guided Embodied Visual Navigation
RESEARCH /
FloNa studies how a navigation system can use a floor plan alongside observed images. It explicitly raises mismatches between drawings and real layouts. For architectural readers, this is a way to examine the limits of plan-based orientation when obstacles and spatial conditions change.
Architectural question
How should navigation respond when the plan and occupied scene disagree?
SINERGYM -- A virtual testbed for building energy optimization with Reinforcement Learning
RESEARCH /
Sinergym presents an open simulation environment for energy-control research and data collection. It supports comparable experiments before real-building intervention. Readers should inspect the simulated building, available observations and control actions, and avoid equating a policy's virtual performance with verified occupied-building outcomes.
Architectural question
How should a virtual control test document its building and action space?
Alejandro Campoy-Nieves, Antonio Manjavacas, Javier Jiménez-Raboso, Miguel Molina-Solana, Juan Gómez-Romero
Pix2Poly: A Sequence Prediction Method for End-to-end Polygonal Building Footprint Extraction from Remote Sensing Imagery
RESEARCH /
Pix2Poly predicts vertex sequences and their connectivity to produce explicit building polygons. It treats footprint extraction as a vector representation problem. A site-context workflow should inspect ring closure, topology and adjacent boundaries, since a valid-looking polygon can still misrepresent the underlying building.
Architectural question
Which topology checks should follow direct polygon prediction?
Development of Low-Cost IoT Units for Thermal Comfort Measurement and AC Energy Consumption Prediction System
RESEARCH /
This study combines low-cost sensing, image recognition and machine learning for small and medium office buildings. It links measured indoor conditions with air-conditioner settings and energy prediction. Architectural adoption should verify sensor placement, index assumptions and the difference between a prediction and a measured energy reduction.
Architectural question
What must low-cost monitoring capture for a useful comfort-and-energy study?
BiTSA: Leveraging Time Series Foundation Model for Building Energy Analytics
RESEARCH /
BiTSA combines forecasting models and interactive visualization to help managers interpret building energy data. It addresses the gap between complex datasets and operational decisions. A useful review should ask how predictions, uncertainty and unusual events become visible rather than judging the interface solely by dashboard appearance.
Architectural question
Can an energy dashboard make uncertainty and unusual events actionable?
Xiachong Lin, Arian Prabowo, Imran Razzak, Hao Xue, Matthew Amos, Sam Behrens, Flora D. Salim
Combining Physics-based and Data-driven Modeling for Building Energy Systems
RESEARCH /
This work examines ways to combine physics-based and data-driven energy models, including residual learning and surrogate approaches. It offers a vocabulary for model selection. The architectural decision is which assumptions should remain physically explicit and which patterns can reasonably be learned from available operational data.
Architectural question
Which building-model assumptions should remain physically explicit?
Leandro Von Krannichfeldt, Kristina Orehounig, Olga Fink
ChatHouseDiffusion: Prompt-Guided Generation and Editing of Floor Plans
RESEARCH /
ChatHouseDiffusion combines natural-language interpretation, topology encoding and diffusion for floor-plan changes. Its appeal is iterative adaptation rather than a fixed generation pass. Architectural review should examine which relationships survive an edit, including adjacency, access and the dimensions absent from a convincing plan image.
Architectural question
Which spatial constraints survive a language-guided plan edit?
Sizhong Qin, Chengyu He, Qiaoyun Chen, Sen Yang, Wenjie Liao, Yi Gu, Xinzheng Lu
Thermal Comfort in Sight: Thermal Affordance and its Visual Assessment for Sustainable Streetscape Design
RESEARCH /
The authors introduce thermal affordance as a way to connect streetscape appearance and physical comfort potential. They propose a visual assessment method for urban design. Its reading value lies in identifying which visible cues matter, while acknowledging conditions that an image alone cannot measure.
Architectural question
What can a streetscape image reveal and miss about thermal comfort?
Sijie Yang, Adrian Chong, Pengyuan Liu, Filip Biljecki
PolyFootNet: Extracting Polygonal Building Footprints in Off-Nadir Remote Sensing Images
RESEARCH /
PolyFootNet targets polygonal footprints from oblique imagery, where viewing angle complicates direct extraction. It addresses a different geometric problem from overhead segmentation. Designers using aerial context should compare ground-level polygon positions against source views and identify ambiguity from occluded walls or overlapping roofs.
Architectural question
How should an oblique-image footprint expose uncertain ground geometry?
DStruct2Design: Data and Benchmarks for Data Structure Driven Generative Floor Plan Design
RESEARCH /
This benchmark emphasizes structured data and numerical requirements in generated plans. It distinguishes attractive raster outputs from layouts that satisfy room sizes and other specified constraints. The architectural question is how evaluation can expose unmet requirements before visual plausibility earns a plan premature acceptance.
Architectural question
How should a plan benchmark measure compliance with numerical requirements?
Zhi Hao Luo, Luis Lara, Ge Ya Luo, Florian Golemo, Christopher Beckham, Christopher Pal
RoBus: A Multimodal Dataset for Controllable Road Networks and Building Layouts Generation
RESEARCH /
RoBus provides multimodal data for controllable city-layout generation, including roads and buildings. It focuses attention on urban characteristics as conditioning information. Architectural users can study what a generated urban pattern actually encodes and where land use, accessibility or local planning rules still require independent treatment.
Architectural question
Which urban characteristics can meaningfully control generated layouts?
Tao Li, Ruihang Li, Huangnan Zheng, Shanding Ye, Shijian Li, Zhijie Pan
Scaling Data-Driven Building Energy Modelling using Large Language Models
RESEARCH /
The authors investigate generated code for processing building-management data and developing data-driven models. The focus is scaling the modeling workflow. Architectural and facilities teams should verify units, time alignment and model assumptions in the generated code before treating faster preparation as improved predictive reliability.
Architectural question
What checks should precede automated building-data modeling?
From Text to Blueprint: Leveraging Text-to-Image Tools for Floor Plan Creation
RESEARCH /
This paper examines text-to-image tools for residential plan creation using prompts, sketches and references. It is useful as an exploration of image-based ideation. Architectural teams should separate the ability to suggest a layout from reliable dimensions, consistent symbols and a model that can support documentation.
Architectural question
When is a plan image useful without being technically complete?
Generating Daylight-driven Architectural Design via Diffusion Models
RESEARCH /
The proposed workflow connects massing generation, window-layout decisions and language/image models. It explores daylight as part of generative architectural ideation. Review should keep daylight calculation separate from visual output and check whether a convincing image still represents the evaluated massing and opening configuration.
Architectural question
Does a generated image retain the massing used in daylight evaluation?
The authors describe a workflow from sketches to conceptual plans and three-dimensional models, with text-guided rendering. Its value is the relationship between successive representations. A design review should record which information survives each transition and which geometry has been invented or simplified by the models.
Architectural question
Which design decisions survive the transition from sketch to generated model?
Occupancy Prediction for Building Energy Systems with Latent Force Models
RESEARCH /
The authors combine Gaussian processes with mechanistic modeling to predict occupancy for energy control. It links learned patterns with a physical-control framework. Architectural and facilities teams should inspect scheduling assumptions, unusual events and the consequences of predicting too few or too many occupants.
Architectural question
How do occupancy forecast errors change building-control decisions?
Adapting HouseDiffusion for conditional Floor Plan generation on Modified Swiss Dwellings dataset
RESEARCH /
This technical report adapts HouseDiffusion to a dataset that supplies structural walls as an input constraint. It studies conditioning generation on wall lines rather than treating the whole layout as freely movable. Architectural use should verify which elements truly remain fixed through generation and editing.
Architectural question
How can a generative plan respect an existing structural skeleton?
CityTFT: Temporal Fusion Transformer for Urban Building Energy Modeling
RESEARCH /
CityTFT applies a temporal fusion transformer to urban heating and cooling demand. It addresses the computational burden of examining multiple climate conditions. Architectural interpretation should separate trigger prediction from load magnitude and check how far future climate inputs differ from the training distribution.
Architectural question
How should urban energy predictions be tested under changed climate inputs?
