Future Directions in Human Mobility Science

TL;DR

This Perspective argues that advancing human mobility science requires linking cognitive mechanisms, social transport systems, and computational AI methods to address public health, climate, and inequality challenges. It reviews how spatial cognition, navigation strategies, and context shape mobility and how multilayer network and multimodal models can capture modern transport systems. It discusses AI-driven trajectory and flow generation, explainability, and the hybridization of mechanistic and data-driven models to improve transferability and policy relevance. It also highlights data-bias, privacy, and ethical considerations and calls for federated learning, synthetic data, and multi-source fusion to enable reliable, responsible mobility insights with tangible urban benefits.

Abstract

We provide a brief review of human mobility science and present three key areas where we expect to see substantial advancements. We start from the mind and discuss the need to better understand how spatial cognition shapes mobility patterns. We then move to societies and argue the importance of better understanding new forms of transportation. We conclude by discussing how algorithms shape mobility behaviour and provide useful tools for modellers. Finally, we discuss how progress in these research directions may help us address some of the challenges our society faces today.

Figures (4)

• Figure 1: Human Mobility data sources.a) The yearly number of publications on Human Mobility as obtained by querying Google Scholar noruzi2005google with $Q =$ ("human mobility" OR "mobility network" OR "mobility flow" OR "flow of mobility" OR "mobility dynamics" OR "mobility patterns"). b) The number of Human Mobility publications for Minds, Societies and Algorithms. Results are obtained by querying Google Scholar for Minds (circles): $Q$ AND "spatial cognition" OR "psychology" OR "navigation" OR "cognitive mechanism" OR "spatial memory"); Societies (squares): $Q$ AND ("transportation" OR "public transport" OR "multimodal" OR "sustainable"); Algorithms (triangles): $Q$ AND ("machine learning" OR "artificial intelligence" OR "AI" OR "computational"). c) Individual mobility paths. From left to right: mobility from CDR data, GPS data collected via smartphones apps, and travel paths estimated from smart-cards. For CDR data, mobility is re-constructed as trips between cell-towers (black dots) whenever individuals issue/receive an sms or call (the background shapes describe the corresponding Voronoi tessellation). GPS data positions are recorded with high spatial resolution, typically at fixed time intervals or whenever the accelerometer of the mobile device registers a change. Travel card data capture portions of a trips that are traveled using public transportation. d) Collective mobility data is obtained through aggregating individual-level data. Example show the aggregation of CDR mobility from multiple individuals.
• Figure 2: Summary of determinant factors and cognitive structures in spatial cognition a) Individual and contextual factors relating to the availability of external information, social interactions and collaboration, and individual characteristics, b) Spatial memory consists of interacting representations based on features, scenes, and the association between them within map-like models. Head direction cells keep track of the location of an unseen target; c) Different strategies shape navigation, working in conjunction, and with other choice architecture, and drawing on different memory representations. These strategies are inherently subjective, potentially results in objectively sub-optimal route choices.
• Figure 3: The complex spatio-temporal nature of multimodal transport systems.a) Human mobility occurs over a large range of spatial and temporal scales. Urban mobility deals mostly with day-to-day displacements, through different transport modalities. b-c) Example of multimodal transportation network in the city of Paris. In panel b the bottom layer corresponds to the walking layer (street network); the mid layer to the Velib network (micromobility bikes), where links are cycle lanes, and nodes are bike docks; the top layer is the metro network. Layers have widely different spatial densities (reported in the figure). While the walking network is static, such that travelling from a given origin to a given destination is always possible (left), travelling along the micromobility layer is subject to the availability of bikes (center) and trips along the metro network are subject to the public transport schedule (right). c) Boxplots displaying the availability of transport options in the Paris transport network as a function of the time of the day: (top) number of vehicle departures/hour across Paris metro stations; (bottom) fraction of Velib stations with at least one bike. For each hour of the day: the horizontal line corresponds to the median value, the box contains 50% of the data, whiskers contain 90% of the data. Street network data from OpenStreetMap OpenStreetMap; Metro data from RATP ratp; Vélib' data from data.gouv.fr datagouvfr
• Figure 4: Human Mobility, AI, and the Urban Environment. (a) A schematic example of how explainable AI tools can improve a deep learning solution to a common problem in human mobility, such as flow generation. The explanation provided by the tool may indicate the importance of variables that characterize the flow's locations, for example, through a Shap-like shapely2007 explanation plot where each point represents a flow, the position on the x-axis is the Shap value, and the color indicates the feature value. Tailored explanations for human mobility are needed, and future efforts in mobility science will require defining explanations that are specifically designed for human mobility. (b) An example of how three different navigation services (NS1, NS2, NS3) may suggest different routes to vehicles with the same origin-destination pair, each path having a different impact on the urban environment in terms of externalities like CO2 emissions. Understanding the collective impact of these services and designing next-generation navigation services that can mitigate their impact while meeting user needs will be a future challenge in human mobility science.