Multimodal transportation. Multimodal transportation combines public transit, private vehicles, shared mobility, and pedestrian travel, forming the core infrastructure of urban transport networks. Key challenges that multimodal transportation faces include difficulties in integrating data across transportation modes, insufficient infrastructure compatibility, and the low attractiveness of public transportation.

Logistics and supply chains. Logistics and supply chains integrate logistics resources and supply chain management technologies to achieve full-chain coordination from raw material procurement, production, and warehousing to last-mile delivery. The main challenges it faces are multiobjective, cross-scale optimization of logistics networks, requiring balanced consideration of supply–demand equilibrium, the network structure, and route allocation. In addition, the connection with multimodal transportation is also a key factor as the transportation time and cost depend on traffic congestion and stochasticity.

Intelligent control infrastructure. Intelligent control infrastructure is a multi-level, multimodule integrated framework, focusing on multidimensional collaborative optimization and dynamic regulation. The intelligent control infrastructure emphasizes signal control optimization, traffic flow dynamics regulation, urban parking management, and guidance of complex traffic networks to achieve the purpose of alleviating traffic congestion, constructing a global perception–decision–control closed loop.

Verification and interpretability of complex urban systems. Due to the inherent nonlinearity, emergence, and dynamic nature of urban mega-mobility systems, MMIF encounters substantial challenges in terms of validation and interpretability. MMIF modeling and simulation often rely on extensive assumptions and parameter settings, making it difficult to guarantee reproducibility and reliability in real-world systems [79]. Meanwhile, with the widespread application of data-driven methods such as AI and machine learning, the “black-box” nature of models is becoming increasingly prominent, limiting the transparency and credibility of decision-making [80].

Granularity and mesoscopic modeling. Existing MMIF models often struggle with determining the appropriate granularity for different subsystems. While microscopic models focus on individual agents and macroscopic models capture aggregated trends, mesoscopic modeling remains underdeveloped. A major challenge is designing mesoscopic models that can effectively bridge the gap between macro- and microscales while maintaining computational feasibility. One approach is to incorporate agent-based modeling with statistical aggregation techniques, enabling a more seamless transition between scales [81]. Additionally, data-driven mesoscopic models can leverage clustering techniques and dimensionality reduction methods to extract representative patterns from large-scale agent simulations [82]. To avoid potential ambiguity, we briefly clarify the 3 analytical tiers relevant to mega-mobility systems. The macrolevel captures system-wide dynamics, such as city-scale traffic flow patterns, while the microlevel focuses on individual-agent behaviors, such as vehicle driving maneuvers. Between them, a mesolevel scale can be recognized, typically addressing corridor or district-scale interactions. As an illustrative case, in transportation, congestion analysis at the mesolevel can be modeled using the cell transmission model, which explicitly captures the propagation of queues and bottlenecks across corridors. This lies between the fundamental diagram at the macrolevel, which aggregates flow–density–speed relationships, and car-following models at the microlevel, which reproduce stop-and-go waves and driver interactions. Similar bridging logic can be extended to other infrastructures. Introducing the mesolevel offers conceptual completeness and indicates how the MMIF framework can be extended to incorporate intermediate interactions when required, thereby strengthening its adaptability to diverse urban contexts.

Inconsistency in macro–micro model coupling. The differences in temporal, spatial, and granularity scales between macro- and micromodels often lead to coupling inconsistencies. While macromodels emphasize overall trends, micromodels capture individual behaviors, resulting in potential scale-mapping distortions and accuracy loss during coupling. For instance, in travel demand forecasting for transportation systems, macromodels focus on regional flow predictions, while micromodels simulate individual travel chains. The differences in spatial and temporal resolution between the 2 may lead to simulation biases [83]. To ensure simulation reliability, strict consistency requirements must be enforced between macro–micro simulation tools, primarily across 2 critical dimensions. Firstly, network information consistency necessitates preserving higher-fidelity information within low-resolution macroscopic networks. The core challenge lies in modifying conventional macroscopic models and algorithms to maintain topological completeness and attribute accuracy during scale transitions. Secondly, mathematical model consistency requires maintaining derivational equivalence between frameworks. Through rigorous analytical derivation, this ensures that the macrolevel impedance function models remain mathematically reconcilable with microlevel models, thereby guaranteeing theoretical coherence across modeling hierarchies [53].

Deployable systems for real-time use. While many state-of-the-art algorithms demonstrate superior performance in controlled simulation environments, their integration into real-time operational systems is often hindered by platform incompatibilities, code generalizability issues, and the lack of modular software architectures [84]. For instance, deep-reinforcement-learning-based algorithms require constant state updates and tight coupling with data streams in the decision-making phase, yet many existing transportation management platforms still lack the real-time responsiveness needed to support such control strategies.

Robustness and runtime safety. Urban environments are inherently noisy, with unexpected incidents such as sensor failures, cyberattacks, or abnormal demand surges. However, most current models do not offer built-in fault tolerance, online error correction, or graceful degradation mechanisms [85]. Without rigorous mechanisms to ensure system resilience, such as redundancy-aware control, online anomaly detection, or adversarial robustness training, intelligent decision-making systems may fail silently or generate erratic outputs under stress, undermining stakeholder trust.

Cross-system data standardization. An urban mega-mobility system often involves multisource data from transportation, energy, communication, and other domains. Although multisource data fusion methods are increasingly sophisticated, their real-world implementation is often fragmented across agencies and vendors with incompatible data schemas, access controls, or update frequencies. Differences in format, timestamps, and spatial resolution hinder effective data fusion. Data fragmentation and inconsistency limit the accuracy and generalization ability of coupled models. This results in information silos that prevent the formation of a unified digital infrastructure. Efforts such as Mobility Data Specifications are still at early stages and face resistance from legacy systems and institutional inertia [86]. Moving forward, promoting cross-system data standardization and unified interface design will be essential to enhance interoperability.

Generative AI reliability and sustainability. Despite their powerful generative and causal reasoning capabilities, LLM-based generative agents are prone to illusion risk—wherein models generate plausible but incorrect or misleading outputs—especially under data sparsity or distribution shifts common in urban mobility contexts [87]. This poses important challenges for reliability in safety-critical applications such as emergency traffic management or energy grid control. Moreover, the substantial computational and energy demands of training and deploying large-scale generative models present a sustainability bottleneck, conflicting with the green and low-carbon goals of smart cities [88]. Future efforts should focus on developing energy-efficient architectures, robust uncertainty quantification methods, and hybrid models that integrate symbolic reasoning to mitigate these risks while maintaining performance.

Enhancing adaptability and generalization capabilities. Future MMIF models will need stronger adaptability and generalization capabilities to cope with the dynamic changes in urban transportation systems. This includes incorporating adaptive modeling, transfer learning, and online optimization mechanisms, enabling models to quickly generalize across different scenarios while reducing reconstruction costs [92]. Additionally, by implementing dynamic feedback mechanisms, models will be able to self-calibrate and optimize in real-time environments [84].

AI-driven complex system optimization. AI will sustain its critical function in advancing smart urban ecosystems [93]. By leveraging multi-agent distributed reinforcement learning and automated machine learning techniques, MMIF models can achieve autonomous scheduling and resource optimization in large-scale urban environments [94]. Additionally, causal discovery algorithms are becoming increasingly important for identifying hidden mechanisms in urban mega-mobility systems with numerous interacting elements, especially in domains where controlled experiments are impractical. Integrating causal inference with machine learning could enable AI to automatically extract interpretable causal structures from high-dimensional data, offering physically meaningful explanations for otherwise black-box models. This convergence of AI-driven optimization and causal reasoning has the potential to transform our ability to model, predict, and manage complex, interconnected systems in a scientifically rigorous and explainable manner.

Explainable decision and emergence. In the context of a mega-mobility system with deep coupling of various networks, a fundamental frontier challenge lies in the quantitative characterization of emergent phenomena and causal architectures. Current theoretical explorations include causal emergence criteria based on effective information, partial information decomposition frameworks, and synergetic information clusters in the mega-mobility system [95]. Future breakthroughs may yield universal emergence laws analogous to thermodynamics’ entropy principle, capable of predicting conditions under which autonomous macroscopic regularities emerge. Concurrently, causal discovery algorithms will play a pivotal role in extracting hidden mechanisms from the observational data of urban mobility systems, which involve highly stochastic and socially influenced human behaviors. These systems share key challenges with sociobiological domains, where controlled experimentation is typically infeasible. The synergistic integration of causal inference with machine learning is anticipated to automatically distill interpretable macroscopic causal networks from intricate datasets in the mega-mobility context. This convergence promises to endow black-box models with physics-grounded interpretations, bridging data-driven patterns with first-principles understanding through causal emergence theory [96].

Interdisciplinary collaboration and policy guidance. Research on urban mobility systems involving multiple networks will increasingly rely on interdisciplinary collaboration and policy guidance. Moving forward, promoting the deep integration of urban science, data science, social science, and policy research will be essential for tackling cross-domain challenges [97]. Additionally, government-led standardization and policy guidance will be necessary to facilitate data sharing, regulate algorithms, and promote technological deployment, thereby enhancing the feasibility and impact of models in real-world management.