AI and IoT merge to revolutionize urban mobility with real-time smart traffic optimization

By employing multimodal data fusion, merging traffic volumes, weather patterns, and vehicle telemetry, the system boosts the contextual accuracy of forecasts and decision-making. The deep learning-based fusion model achieved 91.3% accuracy in traffic predictions, outperforming feature-level techniques and providing robust generalization across different cities and congestion levels.

With the increasing population and climate change crisis, cities worldwide are struggling with rising congestion, emissions, and energy demands. A new study offers a technology-first roadmap for tackling urban mobility challenges through data fusion, real-time AI, and edge-to-cloud orchestration. The 2025 study, titled “Internet of Things and Artificial Intelligence for Secure and Sustainable Green Mobility: A Multimodal Data Fusion Approach to Enhance Efficiency and Security”, was published in Multimodal Technologies and Interaction. It proposes a modular, scalable, and secure framework that merges Internet of Things (IoT) data streams with AI-driven optimization to deliver real-time traffic improvements, significant environmental benefits, and robust cybersecurity performance.

At its core, the framework harnesses multimodal data fusion from electric vehicles, traffic monitors, and weather sensors, running predictive analytics and reinforcement learning to optimize routes dynamically. Deployed in a simulated urban environment with over 50,000 IoT nodes, the system demonstrated a 20% reduction in travel time, a 15% increase in energy efficiency, and a 10% cut in CO₂ emissions. It also integrates privacy-preserving security protocols such as blockchain logging, anomaly detection, and end-to-end encryption - essential for real-world adoption in smart cities.

How does the proposed framework integrate AI and IoT for sustainable urban mobility?

The system’s architecture centers on collecting real-time data from diverse sources including EV diagnostics, traffic sensors, environmental monitors, and smart charging stations. This data is synchronized and fed into deep learning models for traffic congestion forecasting and into Deep Q-Networks (DQNs) for adaptive route optimization. Long Short-Term Memory (LSTM) models handle predictive tasks, while reinforcement learning agents select optimal routes based on travel time, energy use, and congestion metrics.

This hybrid approach ensures not only smarter traffic flows but also operational resilience through an edge-cloud infrastructure. Critical AI inference tasks are run on edge devices like NVIDIA Jetson AGX Xavier nodes to minimize latency, while the cloud handles more intensive tasks like model training and blockchain verification. Real-time responsiveness is key: inference latency remains under 100 milliseconds, making it feasible for on-the-fly decision-making even in high-density urban traffic.

By employing multimodal data fusion, merging traffic volumes, weather patterns, and vehicle telemetry, the system boosts the contextual accuracy of forecasts and decision-making. The deep learning-based fusion model achieved 91.3% accuracy in traffic predictions, outperforming feature-level techniques and providing robust generalization across different cities and congestion levels.

What performance gains were observed in mobility, energy, and emissions?

Experimental validation, conducted using SUMO (Simulation of Urban Mobility) and synthetic datasets such as METR-LA and OpenWeatherMap, revealed that the AI-enhanced system significantly outperforms traditional routing algorithms. Average travel time dropped from 100 to 80 minutes, a 20% improvement, while energy consumption per kilometer was reduced by 15%, and CO₂ emissions by 10%.

The system’s adaptive learning capacity was key to these results. The reinforcement learning agent used an epsilon-greedy policy to balance exploration and exploitation during training and deployed its learned strategies via edge devices in real time. Comparative tests showed that adaptive RL policies yielded 63.2 average reward points per episode, far superior to static or baseline methods.

Battery longevity, energy cost savings, and fairness in traffic distribution also improved under the optimized framework. For example, battery lifespan saw a 39.1% gain, while traffic congestion was more evenly distributed across road segments, reducing bottlenecks and idling time.

Additionally, the explainability of the AI models was enhanced through SHAP analysis and attention visualization, providing interpretability in policy decisions and traffic forecasts - crucial for transparency in public sector deployments.

How does the framework address cybersecurity, privacy, and system scalability?

Security and compliance are embedded in the framework’s design. AI-powered intrusion detection systems classify and mitigate threats such as unauthorized access or malware through a hybrid model combining rule-based validation, reinforcement learning, and human-in-the-loop mechanisms. To reduce blockchain overhead, only critical security events are stored on-chain using Practical Byzantine Fault Tolerance (PBFT), while routine logs remain off-chain to maintain efficiency.

End-to-end data encryption via AES-256 and TLS 1.3, role-based access controls, and source-level anonymization of personally identifiable information ensure compliance with GDPR, CCPA, and PDPA. This focus on secure-by-design architecture makes the framework viable for deployment in sensitive applications such as emergency routing, smart grid load balancing, and public transit coordination.

The system also adapts well to external constraints. While privacy enforcement at the edge caused a minor optimization trade-off (6% performance reduction), it preserved compliance without compromising core efficiency. Conversely, the cloud-based configuration delivered higher optimization scores, albeit with increased latency, highlighting the system’s flexibility in balancing responsiveness and performance based on use-case requirements.

Real-world applications extend across sectors. In emergency response, the framework reduced route-planning time by 32%. In electric vehicle charging coordination, energy cost dropped by 18%. And in public transit, schedule adherence improved by 21%. These results, consolidated across various smart city domains, confirm the framework’s adaptability and high-value impact.