Page 726 - AI for Good Innovate for Impact
P. 726
AI for Good Innovate for Impact
Government housing schemes in India aim to provide shelter for economically weaker
sections, but often the layouts are poorly optimized. These homes become difficult to navigate,
overcrowded, and underutilized. Traditional space planning is manual, static, and lacks
customization based on user needs.
Objective and Aim:
This use case proposes a smart AI system that suggests optimal furniture and storage layouts
within constrained room dimensions using reinforcement learning, deep learning, and rule-
based filtering. The goal is to improve the livability of small homes without increasing cost or
requiring redesign.
Technological Approach:
2D/3D Layout Encoding:
• CAD plans are parsed into 2D grid encodings.
• 3D voxel maps simulate volumetric constraints.
• Graph-based room representations encode furniture relationships.
Computer Vision + CNNs:
• Analyze spatial inefficiencies and extract layout patterns from real floor plans.
Reinforcement Learning:
• An agent (PPO/DQN) optimizes layout decisions by simulating placement actions and
receiving reward feedback based on space usage, accessibility, and aesthetics.
Constraint Filtering:
• OR-Tools filters infeasible layouts (e.g., blocked pathways, overlapping items) to ensure
usability.
Expected Impact:
• Enhanced user comfort and usability in compact homes
• Efficient use of taxpayer-funded government housing
• Scalable optimization adaptable to various living arrangements (hostels, dorms, rental
units)
Use Case Status:
Currently in prototyping phase. Synthetic data from FloorPlanCAD and SYNBUILD-3D used
for training and evaluation. Real-world layout data integration in progress.
Partners
Academic: VIT University (Model development, simulation, evaluation)
Mentor: Dr. Geetha S – Technical oversight and guidance
Planned Collaborations: NGOs in Tamil Nadu & housing development boards for pilot
deployments and feedback validation
690

