How AI can shape the future of urban and transport planning
Where neural networks meet road networks
Cities are complex systems
Saskia Sassen, the influential urban theorist, wrote in The Global City, “The city is a space where complexity gets made, handled, and often made invisible.”
This complexity is a result of continual interactions between geography, policy, social psychology and economics, played out at scale. But the dynamism that gives cities their vibrancy also gives planners a headache. Understanding how so many shifting factors interact is virtually impossible with conventional tools - meaning yesterday’s forecasts and expectations rarely match today’s reality.
A false divide: urban vs transport planning
Traditionally, urban planning and transport planning have been treated as separate disciplines, each with its own data, models, and priorities. Urban planners focus on land use, zoning, and where people live or work, while transport planners concentrate on networks, capacity, and travel demand.
In reality, the two disciplines are inseparable. Land use decisions shape travel patterns, and transport infrastructure reshapes land use. A new rail line can spark waves of housing development, while rezoning an industrial area for apartments can create demand for new bus routes, cycleways, and road upgrades.
Why traditional models fall short
Traditionally, these influences are seen as simple cause-and-effect: more housing means more traffic. Conventional planning and modelling struggle to represent substitution effects. For transport, this refers to how changes in one transport mode changes demand on other modes – for example how a metro line will shift people from cars onto trains. In land use planning, substitution occurs where more local activity and connections reduce the overall need for longer-distance travel. Concepts like mode shift and containment (to use transport-speak) are often reduced to broad generalisations, or tied to major infrastructure projects with extensive modelling. This approach misses the subtle but important shifts in travel patterns that result from myriad small changes - such as new local services, walking and cycling links, or incremental changes in public transport can bring.
For example: The extension of Sydney’s Metro Northwest, while a transport project, has rapidly transformed surrounding suburbs like Rouse Hill and Kellyville into higher-density residential hubs, with new shopping centres and schools. What began as “just” a rail project has had profound land use impacts, while those land use changes in turn affect the transport system.
Furthermore, conventional planning relies on a set of well-trodden data including household travel surveys, location-based traffic counts, and static models that require many assumptions - some of which look shaky when confronted with reality. The result is that when a new project is delivered, the community it was meant to serve may already look different - the Melbourne Docklands Redevelopment is an example. Originally designed for offices and events, it ended up as heavily residential - due to a suite of economic, policy and social factors.
The promise of AI in planning
Helpfully, Artificial intelligence (AI) can make sense of this complexity. By using real-world ‘big data’ about where, when and how residents live, travel, work and play - all processed using sophisticated algorithms - AI offers planners some much-need analytical firepower. AI can take large areas and complex datasets and allow planners to adjust almost any factor within the model. The processing power provides a level of accuracy, clarity and granularity that scales from small projects to major infrastructure initiatives. It could also consider non-infrastructure policy effects – such as changes to pricing, or land use policy constraints. Traditional methods, by contrast, are constrained by the limits of static models – such as the scale of designs on a single page or screen, the limited number of modelling links or variables, the number of assumptions to be made or simply the amount of work it takes to update a model within the project timeframe.
Five ways AI can reshape cities
There are many ways in which AI can - and currently is - reshaping urban and transport planning. We have selected five areas that our team have played a lead role in advancing:
1. Forecasting transport demand and land use interaction with precision
Why it matters: Transport demand and land use are tightly coupled. A new metro line can spark waves of housing development; rezoning industrial land can trigger traffic where little existed before - all moderated by demographics, macro economics, employment and other transport networks. Linear forecasts simplify these complex interaction which can result in materially significant error. Cities risk building infrastructure in the wrong place, at the wrong scale, or at the wrong time—locking in costly mistakes for decades.
AI’s contribution: Machine learning algorithms digest a wide range of data from a wide range of sources: smartcards, mobile phone GPS traces, land-use records, demographic and even social media reactions - revealing patterns that older methods miss. They can simulate thousands of scenarios and refine forecasts as new information arrives.
Example AI system: Trellis AI combines land-use and transport network data, demographic trends and forecast housing and employment scenarios. This aggregation of real world, observed data produces accurate and adaptive forecasts that can quickly update as the city evolves. Crucially, the interaction shows at a street by street level where trips will be contained to the local area as well as the direct demand for walking and cycling trips on specific streets.
2. Optimising networks and operations
Why it matters: Infrastructure is costly and slow to build. Cities must extract more from what they already have.
AI’s contribution: Algorithms can optimise bus frequencies, retime signals, and reallocate kerb space to freight or cycling. Small adjustments deliver large benefits.
Example AI system: The C-MAPS (Cognitive Mobility and Place System) has already been used in Sydney to identify where cycleways will shift the most trips from cars, helping prioritise investment decisions and increase value for money - ultimately helping to cut congestion and emissions.
3. Building adaptive, responsive plans
Why it matters: A ten-year strategy is often obsolete within the first few years. As Lewis Mumford, an influential urban thinker wrote, “the city is a product of time,” and it constantly reshapes itself through new jobs, redevelopments, and unexpected shocks.
AI’s contribution: Continuous learning models refresh forecasts automatically, updating as conditions change.
Vivendi’s tools: Trellis AI provides that feedback loop, turning static strategies into adaptive systems—transport plans that evolve daily, like the cities they serve.
4. Creating safer, more walkable cities
Why it matters: As cities densify, more people walk, cycle, and cross busy roads. Traditional safety planning waits for crashes to occur before intervening.
AI’s contribution: Predictive risk models anticipate where accidents are most likely—even before they happen—so preventative action can be taken. AI can also test whether road safety investment is appropriate - whether the location is a genuine safety risk hotspot, or may simply have been the site of a freak crash. AI models can accurately predict how interventions will lessen risk – and which options – including multiple interventions - are likely to do so most effectively. This allows planners to compare say, traffic calming plus narrower and more visible crossings to simple speed reduction at a particular location – while also considering surrounding demographics and land use.
Vivendi’s tools:
PedSAT identifies pedestrian crash risk hotspots by analysing geometry, traffic, and behavioural factors.
PAWS (Place and Walking System) measures the accessibility of neighbourhoods, showing how new schools, parks, or redevelopments affect walking potential.
5. The need to consider - and demonstrate - strong governance
Why it matters: Technology cannot replace judgement. Data gaps, algorithmic bias, and privacy risks demand transparency. AI models need to be validated and even peer reviewed to ensure that they are robust and reliable, and fit for use beyond their training data area. Users - government and the public - must be confident that the data is as good or better than alternatives and will accurately reflect the future state of the built environment.
AI’s contribution: When embedded within strong governance, AI becomes not a black box but a transparent decision-support tool. It can be more granular and more accurate than traditional modelling, while providing understanding of traditionally less well modelled modes like walking.
The imperative: Clear communication and public engagement are essential for both trust in the process and product. The best AI will synthesise and supplement, as opposed to substitute for, community feedback. Even subjective data like sentiment can, if consistently recorded, be used by AI to enrich models.
As Jane Jacobs warned, “designing a dream city is easy; rebuilding a living one takes imagination.” AI can help—but trust and human oversight, and methods of synthesising human input (like community feedback) remain central.
From static to dynamic planning - with human input
Rather than treating planning as a static exercise, AI enables a dynamic, adaptive approach—learning from new data and responding in real time - to test all the follow-on effects of a change. This allows planners to understand many "what if" scenarios in a rapid, iterative and accurate way. This is substantially better than the current, resource- and cost-intensive sequential process of land use uplift, transport planning and community consultation we currently rely on.
In short, we at Vivendi believe that AI has the potential to untangle Saskia Sassen’s ‘invisible complexity’ when done well. It can avoid costly mistakes, help planners find optimal solutions for complex problems and build better communities – but it has to be underpinned by robust methods, rigorous quality assurance and strong governance.