Related Experiment Video
Updated: Aug 14, 2026
05:47
Simulation of a Scaled Assembly Process with Collaboration of a Robotic Arm and Monitoring through a Vision System for Quality Control
Published on: August 29, 2025
Vision-Based Digital Twin and AI Agent Framework for Low-Cost, Explainable Indoor Building Inspection and Safety
Zijian Jing1, Liyi Zhu2, Tianyi Chen1
1Key Laboratory of Urban and Architectural Heritage Conservation, School of Architecture, Southeast University, Nanjing 210096, China.
Sensors (Basel, Switzerland)
|August 13, 2026
Summary
This study introduces an AI-powered digital twin using smartphone video for indoor safety assessments. The novel framework offers scalable, evidence-based evaluations for aging homes, improving upon traditional manual inspections.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Structural Health Monitoring
Background:
- Aging residential buildings pose safety risks due to outdated design standards.
- Manual safety inspections are subjective, lack audit trails, and are not scalable.
- There is a critical need for objective, evidence-based, and scalable indoor safety assessment methods.
Purpose of the Study:
- To present a vision-based digital twin and AI agent framework for indoor safety assessment.
- To convert smartphone video into an explainable, evidence-constrained safety evaluation.
- To develop a scalable and auditable alternative to manual safety inspections.
Main Methods:
- Utilized MASt3R-SLAM for metric-scale 3D point cloud reconstruction from monocular video, calibrated with AprilTag fiducials.
- Employed SpatialLM for semantic entity and spatial relationship extraction from 3D geometry.
- Developed a LangGraph-based AI agent for dual-pathway assessment (whole-dwelling scan and iterative queries) with tool calls.
Main Results:
- Achieved risk recall rates of 77.8-100% and precision rates of 45.0-70.0% against manual inspection in a pilot validation.
- Demonstrated an average judgment closure rate of 71.7%.
- Showcased spatial granularity enhancement of up to 2.2× in complex residential environments.
Conclusions:
- The AI agent framework provides risk coverage comparable to manual inspections.
- The system offers enhanced spatial granularity in complex areas and quantitative precision in defined spaces.
- This approach represents a significant advancement in scalable and evidence-based indoor safety assessment for aging buildings.
