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 4 / 5
Semi-Supervised Building Footprint Generation with Feature and Output Consistency Training
RESEARCH /
The authors combine feature and output consistency to use unlabeled imagery in footprint learning. It addresses the cost of strong supervision. Architectural mapping still needs geographically appropriate validation, especially where building materials, settlement patterns or image conditions differ from the training data.
Architectural question
How much local validation is needed when labeled imagery is scarce?
Learning to Extract Building Footprints from Off-Nadir Aerial Images
RESEARCH /
This study explicitly separates roof position from ground-level footprint in oblique imagery. It predicts the roof and an offset vector, then translates the roof mask. The distinction matters when aerial context becomes a site model: roof outlines should not silently become legal boundaries.
Architectural question
How can roof-to-ground displacement affect a site model?
Multi-task Learning for Concurrent Prediction of Thermal Comfort, Sensation, and Preference
RESEARCH /
This multi-task study distinguishes comfort, sensation and preference instead of treating them as interchangeable labels. It considers how shared learning can predict several occupant responses. Architectural review should keep the scales separate and inspect whether one successful output conceals weakness in another dimension of experience.
Architectural question
Why should comfort sensation and preference remain separate in a model?
Betty Lala, Hamada Rizk, Srikant Manas Kala, Aya Hagishima
Machine Learning-Based Automated Thermal Comfort Prediction: Integration of Low-Cost Thermal and Visual Cameras for Higher Accuracy
RESEARCH /
The authors investigate combining inexpensive thermal and visual cameras for comfort prediction. It is a sensing-and-modeling study that links occupant data to HVAC questions. Architectural adoption would require separate assessment of consent, privacy, representativeness and whether predictions lead to better comfort decisions.
Architectural question
What occupant data are justified for a comfort-control study?
MTBF-33: A multi-temporal building footprint dataset for 33 counties in the United States (1900-2015)
RESEARCH /
The dataset reconstructs historical building-footprint patterns across 33 US counties using information such as building age. It provides context for studying settlement change. Its architectural value depends on understanding reconstruction uncertainty and resisting the assumption that every historical boundary was directly surveyed.
Architectural question
How should reconstructed historical footprints be compared with contemporary maps?
Conditional Synthetic Data Generation for Personal Thermal Comfort Models
RESEARCH /
The authors investigate synthetic data for personal comfort models with imbalanced responses. It addresses the scarcity of warmer and cooler preferences relative to no-change feedback. A useful evaluation should preserve real-person test data and check whether synthetic samples improve rare-response recognition without masking genuine uncertainty.
Architectural question
Can synthetic samples improve rare comfort-preference predictions?
FloorGenT: Generative Vector Graphic Model of Floor Plans for Robotics
RESEARCH /
FloorGenT represents a plan as an ordered sequence of line segments and uses attention-based prediction. Its proposed tasks include new layouts and completion of partial observations. For architectural readers, the useful distinction is between plausible geometry and a verified building arrangement.
Systematic review of deep learning and machine learning for building energy
RESEARCH /
This review brings together machine-learning and deep-learning approaches for building energy systems. It helps orient readers among forecasting, analysis and management tasks. A design team should distinguish task-specific evidence from broad claims about model performance and check whether comparison datasets represent its own building conditions.
Architectural question
Which energy-model comparisons are transferable to a particular building?
Ardabili Sina, Leila Abdolalizadeh, Csaba Mako, Bernat Torok, Mosavi Amir
Personal thermal comfort models using digital twins: Preference prediction with BIM-extracted spatial-temporal proximity data from Build2Vec
RESEARCH /
This research adds spatial context from BIM to personal thermal-preference prediction, including proximity to openings and environmental sources. It addresses conditions that sparse sensors may miss. Architectural teams should examine how model geometry, occupant location and changing room arrangements influence the inferred comfort relationships.
Architectural question
Can BIM spatial context explain comfort differences within one room?
GridLearn: Multiagent Reinforcement Learning for Grid-Aware Building Energy Management
RESEARCH /
GridLearn links building energy models with power-flow models in a multiagent learning platform. It investigates control of resources behind the meter while considering grid goals. The architectural question is how collective objectives can coexist with individual building comfort and owner requirements.
Architectural question
How can building controls pursue grid goals without hiding local costs?
Aisling Pigott, Constance Crozier, Kyri Baker, Zoltan Nagy
Room Classification on Floor Plan Graphs using Graph Neural Networks
RESEARCH /
This study represents rooms as graph nodes and adjacency as edges, then predicts room categories. Its relevance is the use of relationships rather than image appearance alone. Architectural review should check unusual program combinations and distinguish a predicted category from an explicitly documented room function.
Structural Design Recommendations in the Early Design Phase using Machine Learning
RESEARCH /
This research explores machine-learning support for structural input during early architectural design. The aim is to bring feedback forward while options remain flexible. Such recommendations can inform a comparison, but they still need an engineer's review of loads, assumptions and project-specific conditions.
Architectural question
What structural feedback is useful before a concept is fixed?
Spyridon Ampanavos, Mehdi Nourbakhsh, Chin-Yi Cheng
A Multi-Task Deep Learning Framework for Building Footprint Segmentation
RESEARCH /
This framework couples footprint segmentation with image reconstruction and boundary tasks. It studies whether shared learning improves recognition amid varied urban arrangements. Architectural interpretation should check each auxiliary task's contribution and inspect cases where closely spaced or unconventional buildings still produce misleading maps.
Architectural question
Which auxiliary tasks improve a footprint map's geometric usefulness?
Intelligent Building Control Systems for Thermal Comfort and Energy-Efficiency: A Systematic Review of Artificial Intelligence-Assisted Techniques
RESEARCH /
This review examines AI-assisted control techniques that balance HVAC energy use with thermal comfort. It maps a field of competing objectives rather than proving one universal solution. Architectural teams can use it to compare control assumptions and identify where occupant outcomes need direct evidence.
Architectural question
Which tradeoffs should an AI-assisted HVAC comparison make explicit?
Ghezlane Halhoul Merabet, Mohamed Essaaidi, Mohamed Ben Haddou, Basheer Qolomany, Junaid Qadir, Muhammad Anan, Ala Al-Fuqaha, Mohamed Riduan Abid, Driss Benhaddou
A Semantic Segmentation Network for Urban-Scale Building Footprint Extraction Using RGB Satellite Imagery
RESEARCH /
The authors propose an urban-scale segmentation network using RGB satellite imagery, focusing on scale and class imbalance. Its design relevance is preparing contextual geometry with limited sensor inputs. A trustworthy comparison should include diverse urban forms and report missing or defective footprints explicitly.
Architectural question
How reliable is a city-context map built from RGB imagery alone?
Aatif Jiwani, Shubhrakanti Ganguly, Chao Ding, Nan Zhou, David M. Chan
Knowledge driven Description Synthesis for Floor Plan Interpretation
RESEARCH /
The authors investigate turning floor-plan images into richer descriptions, comparing image-cue and transformer approaches. The architectural issue is whether a description retains useful room and route information rather than merely producing fluent prose. Read it as an interpretation study.
Architectural question
Can a floor-plan description preserve spatial relationships?
A Novel Adaptive Deep Network for Building Footprint Segmentation
RESEARCH /
The paper proposes a Pix2Pix-related network to address imprecise boundaries in building segmentation. Its architectural relevance lies in the edge conditions of a contextual map, where small errors can alter spacing, coverage or apparent street alignment. Predictions still need source-image comparison.
Architectural question
How do segmentation errors change apparent site spacing?
Walk2Map: Extracting Floor Plans from Indoor Walk Trajectories
RESEARCH /
Walk2Map studies extracting floor plans from indoor walking trajectories, addressing the effort of conventional survey capture. It offers a different observation route for spatial reconstruction. Architectural users should examine coverage gaps and the difference between paths people walked and boundaries the system infers around those paths.
Architectural question
Can walking trajectories reveal boundaries that were never directly observed?
Claudio Mura, Renato Pajarola, Konrad Schindler, Niloy Mitra
Integrating Floor Plans into Hedonic Models for Rent Price Appraisal
RESEARCH /
This research asks whether visual features of apartment plans add information to rent appraisal alongside structured property data. It offers a way to study layout attributes in a market model. A predicted rent should not be mistaken for a judgment of spatial quality.
Architectural question
Which plan features add information beyond a property listing?
Preparing Weather Data for Real-Time Building Energy Simulation
RESEARCH /
The proposed framework detects anomalies and fills gaps in measured weather inputs using neural-network methods. Its relevance precedes energy simulation: flawed weather records can undermine an otherwise careful model. The study invites comparison with simpler imputation baselines and explicit gap-length reporting.
Architectural question
How should missing weather data be handled before simulation?
Robust building footprint extraction from big multi-sensor data using deep competition network
RESEARCH /
The authors combine optical imagery with LiDAR using a deep competition network. The paper addresses footprint extraction where one sensor alone may leave ambiguity. Its relevance is evidence fusion for urban context; combining sources still requires coordinate, coverage and timing checks.
Architectural question
When does LiDAR resolve ambiguity in aerial footprints?
Balancing thermal comfort datasets: We GAN, but should we?
RESEARCH /
The paper questions synthetic balancing of comfort datasets, where dissatisfied responses may be uncommon. Its value is methodological: improving class proportions may not improve a model's usefulness. Building teams should preserve the real response distribution and evaluate rare discomfort cases separately.
Architectural question
Can synthetic balancing distort an occupant-comfort model?
Matias Quintana, Stefano Schiavon, Kwok Wai Tham, Clayton Miller
A Review of Deep Reinforcement Learning for Smart Building Energy Management
RESEARCH /
The review discusses deep reinforcement learning for smart-building energy management amid uncertain weather, generation and occupancy. It helps frame control problems before choosing a method. Architectural readers should compare observation limits, feasible actions and the comfort constraints that define a genuinely usable policy.
Architectural question
Which uncertain inputs should an energy-control review expose?
Boundary Regularized Building Footprint Extraction From Satellite Images Using Deep Neural Network
RESEARCH /
The study targets the difference between pixel-level recognition and a usable geometric representation of buildings. Boundary regularization is proposed to improve extraction from satellite imagery. For designers, the important test is whether the resulting map preserves corners, separations and contextual relationships.
Architectural question
Which geometric relationships are lost in pixel-only footprints?
