[Paper Review] Autonomous AI Agents for Real-Time Affordable Housing Site Selection: Multi-Objective Reinforcement Learning Under Regulatory Constraints

Olaf Yunus Laitinen Imanov, Duygu Erisken+3 more|arXiv (Cornell University)|Feb 3, 2026

Housing Market and Economics0 citations

TL;DR

AURA is a hierarchical multi-agent reinforcement learning framework that performs real-time, regulation-aware affordable housing site selection by jointly optimizing multiple objectives and ensuring hard regulatory compliance, achieving high Pareto performance and rapid decision-making.

ABSTRACT

Affordable housing shortages affect billions, while land scarcity and regulations make site selection slow. We present AURA (Autonomous Urban Resource Allocator), a hierarchical multi-agent reinforcement learning system for real-time affordable housing site selection under hard regulatory constraints (QCT, DDA, LIHTC). We model the task as a constrained multi-objective Markov decision process optimizing accessibility, environmental impact, construction cost, and social equity while enforcing feasibility. AURA uses a regulatory-aware state encoding 127 federal and local constraints, Pareto-constrained policy gradients with feasibility guarantees, and reward decomposition separating immediate costs from long-term social outcomes. On datasets from 8 U.S. metros (47,392 candidate parcels), AURA attains 94.3% regulatory compliance and improves Pareto hypervolume by 37.2% over strong baselines. In a New York City 2026 case study, it reduces selection time from 18 months to 72 hours and identifies 23% more viable sites; chosen sites have 31% better transit access and 19% lower environmental impact than expert picks.

Motivation & Objective

• Address the global affordable housing crisis by enabling rapid, data-driven site selection at metropolitan scale.
• Embed hard regulatory constraints within the optimization loop to ensure feasibility across federal and local programs (QCT, DDA, LIHTC, LIHTC allocations, zoning).
• Balance four objectives—accessibility, environmental sustainability, construction cost, and social equity—within a constrained MDP framework.
• Develop a scalable, multi-agent architecture that coordinates geospatial analysis, compliance verification, and Pareto-front optimization.

Proposed method

• Formulate site selection as a Constrained Multi-Objective MDP (CMO-MDP) with 127 hard constraints.
• Introduce AURA, a four-agent hierarchy: Geospatial Analysis Agent (GAA), Regulatory Compliance Agent (RCA), Multi-Objective Optimization Agent (MOOA), and Coordination Agent (CA).
• Develop regulatory-aware state representations combining continuous geospatial features and discrete constraints.
• Propose Pareto-Constrained PPO (PC-PPO) to learn a Pareto front while enforcing feasibility via a constraint-penalized objective and an explicit feasibility term.
• Implement multi-fidelity rewards to separate immediate costs from long-term social and environmental impacts, using temporal abstraction up to 10 years.
• Use a graph neural network for spatial encoding, a neural satisfiability solver for constraints, and Pareto-front maintaining policy populations with diverse preference vectors.

Experimental results

Research questions

• RQ1 How can affordable housing site selection be cast as a constrained multi-objective optimization problem under hard regulatory requirements?
• RQ2 Can an autonomous, multi-agent system respect 127 regulatory constraints while discovering Pareto-optimal trade-offs among accessibility, environmental impact, cost, and social equity?
• RQ3 Does a Pareto-constrained PPO approach enable feasible, high-quality Pareto fronts faster than traditional MORL or constraint-agnostic methods?
• RQ4 What is the impact of regulatory-aware state representations and multi-agent coordination on solution quality and compliance rates in large metropolitan datasets?

Key findings

• AURA achieves 0.715 hypervolume with 94.3% regulatory compliance across eight U.S. cities.
• AURA improves Pareto hypervolume by 37.2% over the best baseline and outperforms baselines on multiple objectives.
• Regulatory compliance is significantly enhanced by integrating the RCA, with compliance rate dropping dramatically when RCA is removed (78.4% vs 94.3%).
• AURA enables much faster site selection (72 hours vs 18 months) and identifies more viable locations under constraints (e.g., NYC case).
• Deployment results show 31% better transit accessibility and 19% lower environmental impact for sites selected by AURA compared with human expert selections.
• Across cities, AURA consistently dominates baselines on the Pareto front while maintaining strong equity characteristics (lower Gini-like inequity).