Five directions, one question each.
Each direction is framed by a question, the data it draws on, and the methods it uses. Outputs are released through the platforms rather than listed here.
Urban representation & multimodal learning
How can cities be represented from imagery, maps, mobility, and text in ways that support valid comparison across places?
Representations are the layer every downstream measure depends on. The direction asks what a model has actually learned about a city, and whether that transfers.
- Questions
- What urban features do multimodal models encode, and how do their representations differ between cities?
- Data
- Street-level and satellite imagery, basemaps, mobility traces, and urban text corpora.
- Methods
- Multimodal learning, spatial embeddings, and interpretability analysis of learned features.
Human-centered sensing & perception
How do people perceive and experience urban environments, and how can perception be measured at scale without losing what makes it subjective?
Perception links the physical environment to behaviour. The direction treats perception as a measurement problem with validity, bias, and heterogeneity to account for.
- Questions
- Which environmental features shape perceived safety, comfort, and vitality, and for whom?
- Data
- Surveys and ratings, crowdsourced annotations, environmental sensors, and imagery.
- Methods
- Perception models, measurement validation, and mixed-method designs.
On the network
Theme hubsMobility, accessibility & inequality
Who can reach what, at what cost, and how do urban form and infrastructure shape those differences?
Access is where urban form becomes opportunity or exclusion. The direction measures it across modes and groups, and tests which interventions change it.
- Questions
- How do street environments and networks affect route choice, mode choice, and access to services?
- Data
- Transit and road networks, mobility and trip data, census and amenity records.
- Methods
- Accessibility measures, spatial econometrics, and causal research designs.
Climate resilience & environmental justice
How are heat, air pollution, flooding, and green space distributed across a city, and which communities bear the exposure?
Environmental risk is unevenly distributed and often compounds social disadvantage. The direction maps exposure, links it to outcomes, and evaluates responses.
- Questions
- Where do exposure, vulnerability, and adaptive capacity overlap, and what reduces the gap?
- Data
- Remote sensing, environmental monitoring, land cover, and health and demographic records.
- Methods
- Exposure modelling, spatial statistics, and equity-focused evaluation.
Urban agents, simulation & decision support
When can computational agents and simulations support planning decisions, and how should their claims be tested before anyone relies on them?
Agents and simulations promise cheaper evidence. The direction treats that promise as a hypothesis, with evaluation protocols that test structure, not only averages.
- Questions
- Do simulated populations preserve the social and spatial structure of real ones, and where do they fail?
- Data
- Scenario inputs, administrative records, survey experiments, and planning documents.
- Methods
- Agent-based simulation, evaluation of language and multimodal models, and decision-support design.
Released through the platforms.
Data and indicators are published on UrbanProbe. Models, agents, and benchmarks are published on XUrbanAI. Collaboration runs through the City Computing Collective.