Cognitive Digital Twins for Smart City Mobility: An Adaptive Multi-Agent System Approach

Abstract

Digital Twins have become a central tool for Smart City planning and management, enabling simulation and analysis of complex urban systems such as mobility and energy infrastructures. However, most existing implementations primarily focus on physical and operational layers, relying on static rules or aggregate models that overlook the adaptive and goal-driven behavior of urban actors. This limitation reduces their effectiveness for realistic policy evaluation, stress testing, and what-if analysis in the presence of heterogeneous agents and non-linear feedback loops. In this work, we introduce the concept of the Cognitive Digital Twin (CDT), an extension of the Digital Twin paradigm that explicitly incorporates agent-level cognition, learning, and adaptation. Urban entities are modeled as autonomous agents with goals, memory, and online decision-making capabilities, allowing system-level dynamics to emerge endogenously from decentralized interactions. The proposed CDT adopts an adaptive multi-agent architecture based on lightweight online learning, specifically contextual bandits, prioritizing interpretability and reproducibility over opaque approaches. We formalize the CDT framework, describe its layered adaptive multi-agent architecture, and present a reproducible Python-based simulation platform. The approach is evaluated through a simulation-based case study on urban mobility and energy dynamics, including baseline, policy, and stress scenarios, supported by systematic parameter sweeps and multiple stochastic repetitions. We also provide an optional SUMO-compatible grounding layer that can connect CDT-generated demand and policy scenarios to calibrated microscopic traffic networks. Experimental results demonstrate that adaptive agent behavior generates meaningful and policy-relevant emergent patterns, highlighting trade-offs between travel efficiency, energy consumption, and user utility under different governance strategies. This work positions Cognitive Digital Twins as a promising prototype-level research framework for behavior-aware Smart City mobility simulation, bridging the gap between physical simulation and adaptive urban modeling while acknowledging that full city-scale empirical validation remains future work.