Query Complexity Defines Limits of Unitary Time-Reversal and Control-NOT Implementation.

The increasing complexity of modern urban environments demands innovative approaches to managing and understanding mobility patterns. Yue Ding, Conor McCarthy, Kevin O’Shea, and Mingming Liu present a new platform that combines the power of large language models (LLMs) with cloud-based traffic simulation to address this challenge. Their research introduces a system capable of providing personalised route recommendations via a mobile application, while simultaneously offering comprehensive analysis for city planners and service operators. This work demonstrates a significant step forward in ‘smart mobility’ solutions, achieving high accuracy in responding to both complex operator queries and everyday user requests, and paving the way for more efficient and responsive urban transportation systems.

Recent advances in smart mobility and data analytics are driving the development of sustainable transportation solutions. Within this evolving landscape, Mobility as a Service (MaaS) integrates various transport options – public transit, shared mobility, and private vehicles – into a unified, user-centric platform. Shared e-mobility services, such as e-cars, e-bikes, and e-scooters, offer flexible alternatives to private vehicles, promoting sustainable urban travel. As clusters of these services – known as eHubs – become more common, there is a growing need for simulation tools to evaluate their impact on urban mobility. Existing simulation platforms often lack the open-source accessibility and scalability required for detailed study of e-mobility scenarios. While some tools address congestion and scheduling, they often fall short in areas like energy management and sustainability. To address these limitations, researchers have developed a new cloud-based platform that integrates dynamic traffic simulation with natural language interaction, powered by Large Language Models (LLMs). This platform supports the deployment and evaluation of eHub distributions, offering a scalable solution for shared mobility infrastructure. The platform builds upon previous research in energy modelling and optimisation algorithms, employing a modular and data-centric architecture. This design allows for easy adaptation to digital twin modelling and deployment in different regions, recreating realistic local mobility patterns without extensive reengineering.

Key components include modules for traffic simulation, docking station deployment, multi-modal transport, and energy-aware route planning, all accessible through an Android app using an agent-in-the-loop approach. A crucial innovation is the use of Retrieval-Augmented Generation (RAG) to enable natural language interaction with simulation outputs. This RAG-based approach, tailored for structured traffic and user travel data, allows both users and system operators to explore personalised insights beyond traditional charts and views. The system achieves high execution accuracy – 0.98 for users and 0.81 for system operators – in responding to queries, demonstrating the potential of LLMs to make complex simulation data more understandable and accessible. Alongside the platform development, researchers investigated the impact of network depth on the accuracy of Convolutional Neural Networks (CNNs). Their analysis revealed that accuracy generally increases with the addition of layers, up to a certain point. While a 2-layer CNN achieved 82.3% accuracy, this improved to 90.5% with 6 layers and 92.8% with 8 layers. However, a 10-layer model only reached 93.1% accuracy, suggesting diminishing returns and a potential risk of overfitting. This highlights the importance of carefully balancing network complexity with generalisation and computational efficiency. The research demonstrates a promising platform for improving urban transportation by leveraging cloud computing, real-time data processing, and RAG. Future work includes expanding the system to consider multiple objectives – such as cost, time, and emissions – supporting scenarios with multiple users, incorporating public transportation networks, and enhancing the RAG framework with external knowledge sources. Ultimately, this work aims to link the platform to real-world data and live traffic conditions, creating more dynamic and realistic simulations for sustainable urban mobility.