Urban Planning AI

AI is rapidly reshaping how cities are studied, governed, and planned. But “not all urban uses of AI are planning in the institutional sense” (Peng, Feng, & Wu, 2026). To see why, it helps to start with two related ideas: Urban AI and Urban Planning AI.

“Urban AI is the broader domain in which computational systems are used to observe, model, predict, generate, or optimize aspects of cities and urban systems” (Peng, 2026b). It works as an analytical framework within urban science, “organized around three core functions: identifying patterns, analyzing causes, and optimizing urban systems” (Peng, Feng, & Wu, 2026).

Common methods include machine learning, deep learning, and computer vision. They are applied to areas such as mobility prediction, land-use change, environmental monitoring, and infrastructure management (Peng, Feng, & Wu, 2026).

Urban Planning AI is a specialized subfield of Urban AI (Peng, Feng, & Wu, 2026). It “refers more specifically to the use of AI in institutional planning processes affecting land use, housing, mobility, zoning, infrastructure, public investment, and access to urban resources” (Peng, 2026b).

In our 2024 study, we define Urban Planning AI as “the process of plan-making that is conducted, either partially or entirely, by an AI agent during the urban planning process” (Peng et al., 2024). Once AI enters this space, “it is no longer only helping us understand cities; it is participating in processes that shape public decisions” (Peng, Feng, & Wu, 2026).

Urban AIUrban Planning AI
Role of AIUnderstands and operates within urban systemsTakes part in institutional plan-making and policy formation
What it doesIdentifies patterns, analyzes causes, optimizes urban systemsSupports, augments, automates, or autonomizes parts of plan-making
What it is judged byTechnical performanceTechnical performance, plus legitimacy and accountability

Source: Peng (2026b); Peng, Feng, and Wu (2026).

“Planning differs from ordinary information processing because it exercises or informs public authority” (Peng, 2026b). Decisions about land use, housing, transportation, infrastructure, and public investment “distribute opportunities and burdens across people and places.” They “require contextual interpretation, value judgments, political negotiation, and democratic authorization” (Peng, 2026b).

AI can perform bounded tasks effectively. But current AI systems “cannot independently reconcile competing claims to justice, determine what communities should value, or accept responsibility for public consequences” (Peng, 2026b).

“When we use AI to analyze a city, we are doing science; when we use AI to allocate resources or zone neighborhoods, we are exercising public power.”

In that second case, the question is not only whether a model is accurate. It is also “whose goals it is serving, what constraints bind it, who can challenge it, and who remains responsible for the final decision” (Peng, Feng, & Wu, 2026). When AI informs or exercises public authority, “legitimacy, authorization, transparency, equity, contestability, and accountability also become relevant standards” (Peng, 2026b).

  • Urban AI uses AI to understand cities. Urban Planning AI brings AI into the institutional process of making plans.
  • Planning informs or exercises public authority, so technical performance alone is not enough.
  • That is why our research treats Urban Planning AI as a governance question, not just a technical one.

Our 2024 study in the Journal of Planning Education and Research introduces a four-phase pathway describing how AI’s role in planning could evolve, from supporting planners to taking on more substantial parts of the planning process (Peng et al., 2024). The typology “is based on different levels of participation from planners and the autonomy levels of AI agents in the plan-making process” (Peng et al., 2024).

AI autonomy: Low

Planner: in the loop

“AI has low autonomy and planners play a dominant role in the plan-making loop.” AI collects, processes, and visualizes urban data, and helps identify planning issues and patterns. Planners lead the process, engage stakeholders, propose solutions, and make decisions.

AI autonomy: Low to intermediate

Planner: in the loop to on the loop

AI “assumes a more prominent role, transitioning from low to intermediate autonomy levels.” It simulates and evaluates the planning options that planners propose. Planners still set goals, refine options, and make the final plan.

AI autonomy: High

Planner: on the loop

“AI is more autonomous in generating plans under the supervision of planners.” It can “semi-automate urban planning processes and suggest the optimal option.” Planners define the problem, set goals, supervise the AI, and monitor its outputs.

AI autonomy: Highest

Planner: out of the loop

AI would have “strong capabilities of self-learning, decision-making, and plan assessment.” This phase remains prospective, and “achieving it in the short term is unlikely.” Planners would still evaluate results, adjust inputs, and maintain the AI system.

Source: Peng et al. (2024).

The study describes this shift through the “plan-making loop.” When planners are in the loop, they “have a dominant role in the plan-making process, with AI playing a supplementary role.” When planners are on the loop, AI takes on more responsibility “while planners assist AI in making planning decisions.” When planners are out of the loop, AI makes plans on its own, but “this does not diminish the significance of the planner’s role.” Planners still evaluate results, adjust goals, and maintain the AI system (Peng et al., 2024).

A table titled The role of AI agent and planner detailing five levels of urban planning AI, comparing the roles of the AI agent and the human planner, alongside flowcharts illustrating the plan-making loops for each level.
Table showing the roles of the AI agent and the planner in each phase: traditional planning, AI-assisted, AI-augmented, AI-automated, and AI-autonomized planning with diagrams of the plan-making loop between planners, AI agents, and planning decisions.

“The pathway is diagnostic, not a claim that every application will or should advance toward greater autonomy” (Peng, 2026b). Contemporary planning applications remain concentrated primarily in assisted and augmented uses, while automated and autonomized planning remain more prospective (Peng, 2026b). The framework helps “distinguish technical capability from delegated authority” (Peng, 2026b).

As AI’s role expands, a further question emerges: how can its analytical and generative capabilities be used without weakening human judgment, democratic legitimacy, or public accountability? The pathway paper raised this question but “could not yet answer” it (Peng, Feng, & Wu, 2026). Our research on Symbiotic Planning Theory addresses it.

Our 2026 paper introduces Symbiotic Planning Theory (SPT), which responds directly to a question the pathway framework left open: once AI begins moving from planning support toward plan-making, what conditions keep that role democratically accountable?

AI must surface the value questions built into a task, turning vague goals like “we want this to be fair” into explicit, enforceable constraints.

Community priorities cannot stay advisory. They must become hard constraints the model is required to satisfy before deployment.

AI’s authorization to operate expires unless renewed through evidence-based human review, preventing algorithmic authority from becoming permanent by default.

These conditions are put into practice through the CORE framework — Collaboration, Options, Refinement, Execution — a cycle that returns to Collaboration each time, so governance is renewed rather than assumed. Figure reproduced from the paper; click to enlarge.

Planners, communities, and agencies define the problem together and set the terms for AI’s involvement before any model runs.

AI generates alternative solutions, not a single final answer, making trade-offs visible rather than hidden inside an algorithm.

Proposals are reviewed, challenged, and revised. Humans hold final authority to accept, modify, or reject any AI-generated option.

Implementation is treated as the start of ongoing, monitored governance, not the end of the planning process.

Share of citywide scooter rides, by area

Equity zone

[p]%

Near campus

[p]%

This framework was illustrated through a case study of shared micromobility deployment in Gainesville, Florida, where a vendor met a formal equity rule (deploying scooters in lower-income “equity zones”) while still producing deeply unequal outcomes: those zones generated only about 3% of citywide rides, while roughly 84% of rides began near campus. The research reframed this as a governance problem, not simply a compliance failure, and used AI-driven analysis alongside structured community engagement to redesign the deployment approach around binding equity constraints rather than procedural compliance alone.

AI can expand what planners are able to analyze, simulate, and test. But without governance, that same capability can shift meaningful decisions into systems that communities, and even decision-makers themselves, cannot easily review or challenge.

Our research in this area works to keep that expanding capability paired with clear mechanisms for human authorization, transparency, and accountability.

Peng, Z.-R., Lu, K.-F., Liu, Y., & Zhai, W. (2024). Journal of Planning Education and Research, 44(4), 873–885.

doi.org/10.1177/0739456X231180568

Peng, Z.-R., Lu, K.-F., Liu, Y., Hou, Q., & Zhang, Q. (2026). Journal of the American Planning Association.

doi.org/10.1080/01944363.2026.2640038

Peng, Z.-R., Feng, X., & Wu, Z. S. (2026). Frontiers of Urban and Rural Planning, 4, 13.

doi.org/10.1007/s44243-026-00086-5

Scroll to Top