Step 1 of 5
Views
Visible layers
Rendering
Activity layers
Data: synthetic population, Marcel Bartumeu Raméntol
Appearance
Alignment target
Table state
No state hub: the segments are chosen in Pre-form.
Agents
Data: synthetic population, Marcel Bartumeu Raméntol
Mode of travel
Each person's colour is how they are travelling now, on this leg of their day.
Near the table
Close up, people on foot become figures, riders get a bicycle or a scooter, drivers a car; people on transit ride these.
* Illustrative times. Red Line and buses: MBTA weekday schedule (GTFS). Map data © OpenStreetMap contributors.
Numbers
Research
Scene controls
Enable Realistic to adjust solar month and textures.
Playback speed
Solar month changes lighting and vegetation only.
Population uses the same weekday dataset.
Functional architecture
Four stages in a loop: the model, the changes people make to it, the simulation and its feedback.
Discuss together. Revise the scenario.
Indicators
Simplified interpretation of the published platform.
Behind CityScope Volpe
CityScope: A Data-Driven Interactive Simulation Tool for Urban Design. Use Case VolpeLuis Alonso · Yan Ryan Zhang · Arnaud Grignard · Ariel Noyman · Yasushi Sakai · Markus ElKatsha · Ronan Doorley · Kent Larson
- ICCS 2018
- Pages 253–261
- DOI: 10.1007/978-3-319-96661-8_27
- City Science · MIT Media Lab
The MIT project page lists this chapter more than once, with different dates or wording. It is one publication, identified by its DOI.
Project contributors
Besides the publication's authors, the MIT project page also acknowledges:
- Dalma Foldesi
- Jung In Seo
Markus F. ElKatsha, also listed on that page, is the publication's author Markus ElKatsha.
About this diagram
“This diagram is an explanatory synthesis of the published platform: physical model, tangible interaction, computation and feedback.”
Development on the site
Choose what is built on each of the site's five segments, and see it on the table.
Development graphs on the site: Off
Site segments
A program for each segment
The Volpe site is divided into five segments. Each can hold a development program, drawn on the table as a graph of its towers and plinths.
A program that breaks a rule is refused, and the reason is said; nothing is shown as placed before it passes. The same graphs can come from the menus, from a file, or from the reference application.
Behind Pre-form
Preformative design modelThe research explanation and sources for Pre-form are not available in this app yet. They will appear here once they are provided; nothing is written in their place.
The day on the table
Show the synthetic population moving through its day, and read what it is doing now.
Agents
On the table ·
Or tap an agent on the table to open their profile.
Close to the table, people on foot and on bicycles become figures, those driving become cars, and those on transit ride the buses and trains drawn on the MBTA's weekday schedule (GTFS). The freight train, the EZRide shuttle and the boats on the Broad Canal run on illustrative times. Map data © OpenStreetMap contributors.
Flows
Trips between the H3 cells of resolution 8 on the table (about 0.9 km across), in the hour of the time shown, all modes together, as the population's data counts them. Trips within one cell are not counted.
Occupancy
People present in places (not travelling), per H3 cell of resolution 9 on the table (about 0.35 km across), in the 10-minute slot of the time shown. A cell's height is its share of the busiest slot of the day.
Inside a simulated decision.
An agent observes its context, considers what to do, and acts; what it does changes the context it observes next.
The action updates the context.
H-ABM is a modular conceptual framework for bringing cognition supported by large language models into agent-based models. This cycle is an editorial summary of it, not its full technical diagram; memory and emotions are modules a model may add to cognition.
Conceptual research framework. Active modules depend on the running implementation.
Behind H-ABM
Humanized Agent-based Models: a FrameworkVersion 3 · 11 September 2025 · Preprint
From data to urban patterns
On this table, each agent of a synthetic population follows a recorded day of activities and trips. Occupancy and flows add up those same records: people present in places, and trips between places.
The layers show these records as they are. They do not show the reasoning behind them.
Research scope
Version 3 of the paper proposes the framework. It does not show that the agents exported to this app use its modules, and it does not offer a predictive validation of this population.
A routine or a walking animation does not establish that an agent has memory, emotions or individual reasoning. Which modules are active depends on the implementation that runs.
Paper and authors
Humanized Agent-based Models: a Framework. Version 3, 11 September 2025, a preprint on TechRxiv. The full list of authors is at the source.
A vote on the table
Preview an example of one person's vote, split three ways, on the site.
Demo preview: Off
Illustrative demo — not live voting.One vote.
More ways to have a say.
People can support options directly, and also through people they trust or shared priorities. Their contributions propagate through a network.
Where your vote ends upStill 1 vote
Your preference and your trust can count together.
A teaching example, not the method's full mathematical formulation. Housing's equal split belongs to this example; it is not a rule of the method.
Behind the vote
A New Town Scale Consensus: Propagational Voting in Urban Decision MakingYasushi Sakai · MIT doctoral thesis · 2025
How it works · pp. 51–62
Each person starts with one divisible vote. Shares can be allocated to proposals, trusted people or groups, and policy categories. Delegated shares follow the recipients’ recorded allocations through the network, ending at concrete proposals.
The model uses an absorbing Markov chain to calculate proposal support. It can also analyze the influence of people and intermediaries. A support distribution does not by itself determine which policy is adopted.
In this example, the trusted person directs their received share to Option B. The housing category divides its received share equally between Options A and B. Only the original person’s vote is traced.
Pilot evidence · pp. 97–133
The thesis explores the mechanism through a Boston urban-planning survey and participatory workshops. These studies examine how people express preferences, delegate, and influence proposal support. They provide exploratory evidence at small civic scales.
Limits & open questions · pp. 141–147
Scale, voter identity, privacy, intermediary design, and concentrated influence remain open research questions. The mechanism does not guarantee agreement or prevent power concentration.
Closed delegation loops with no route to a proposal require explicit handling; the thesis discusses this exception on p. 68.
Related publication
Propagational Proxy VotingYasushi Sakai · Parfait Atchade-Adelomou · Ryan Jiang · Luis Alonso · Kent Larson · Ken Suzuki
- arXiv preprint · April 2025
- Social and Information Networks
- DOI: 10.48550/arXiv.2504.13641
The idea: when groups combine
One vote, split into fractions, can go to proposals, to other people, and to “sets containing proposals and participants”: groups and categories that bring them together.
This is where it becomes most interesting. A group passes on what it receives, so trust in people and priorities between proposals combine across layers, as a party shares its votes among its members. In the paper's words, “intermediary nodes, such as categories or groups, play a significant role in collective decision outcomes.”
Open questions
How a group should pass on what it receives is still open, and the calculation grows costly at large scale.
Try it on the table
Pick a request the app supports, and see the answer on the table.
From a question to the table
A spoken question about a place becomes a visualization in space, in three steps.
Where do people spend time?
A spoken question leads to the relevant information, and the answer arrives in space together with its explanation.
An agent's route is one concrete use: it appears only when it is asked for. Illustrative example · no live data.
Behind the conversation
Conversational Spatial Analytics in Augmented RealityCity Science · MIT Media Lab · Project contact: Matti Gruener
The research concept
The project explores voice-based access to spatial data in augmented reality. An agent retrieves relevant information in near real time and presents a visualization anchored to the surrounding space.
Supported interactions
Available in this app now: specific commands that show or hide layers, select an agent and draw their route, and open the research guides. They are typed in the conversation until the voice assistant is connected.
The project's overview describes specific queries and filtered views, and outlines future support for open-ended questions. This app does not answer open-ended questions.
Open questions
Does this improve learning, insight, speed, or access to complex data? Can it support meaningful policy change? These remain research questions.
“An interface that brings CityScope’s urban models, activity and research into one spatial experience, designed for exploration across screens and supported XR devices.”
Credits
- Mónica GómezInterface design & UX
- Felipe RiveraCross-platform stabilization
- Thomas Sánchez3D models
Explore through the interface
- Reveal the cityExplore point clouds, urban models and visual layers.
- See urban lifeExplore people, movement and activity over time.
- Connect to researchDiscover the projects behind the visualizations.
The interface uses what this app already has: layers, agents and the research windows. Where a project's window says a part is illustrative or not built yet, that still holds here.
Agent profile
Now ·
Route trail: Off