Abstract
Sustainable municipal resource stewardship requires a harmonized artificial intelligence (AI) architecture that aligns technical performance with human-centered behavioral dynamics. Traditional smart city enforcement frameworks frequently trigger psychological reactance and perceptual failure in noisy, crowded urban spaces. This study presents HARMONY, a socio-technical edge-computing and agent-based framework that integrates localized computer vision (YOLOv8) with Mesa 3.0 agent modeling to promote non-coercive civic compliance in waste disposal. We formally calibrate behavioral constants, including the Politeness Modifier (Lp = 0.98), Gratitude Loop increment (δ = 0.025), and irreducible human defiance floor (ε = 0.12), grounded in self-determination theory, politeness theory, and empirical habit formation models. We also mathematically resolve sensory asymmetry in Hierarchical Orchestration Logic (HOL) by enforcing strictly normalized, bi-directional modality allocation (wv + wa = 1.0) conditioned on acoustic noise (N) and optical reliability (Qv). Across 100-run Monte Carlo simulations, HARMONY achieves a terminal Correct Disposal Rate (CDR) of 87.9% ± 3.3% (95% CI: [87.3%, 88.6%]), outperforming the coercive baseline (57.0% ± 4.9%). A comprehensive six-arm factorial ablation study isolates the distinct contributions of interactional politeness (+30.5% compliance gain), HOL sensory resilience (+1.6% to +21.4% under severe noise), and gratitude habituation (+0.5% to +30.3% sustained retention). Furthermore, population heterogeneity evaluations across diverse civic cohorts confirm robust performance from civic champions to highly resistant populations. Localized edge prototyping demonstrates 18.4 ms inference latency with 94.2% mAP, operating under an Edge-Only Ephemeral Vector Processing (E-EVP) paradigm that guarantees privacy-by-design. These findings demonstrate that cooperative, polite AI stewardship offers a scalable, legally robust, and ethically aligned alternative to punitive urban surveillance.