Skip to content
AGI Future Foundation PBC

Article 70 · SmartCiti.X · AGI Future Foundation

The City That Thinks in Traffic: How Roads, Signals, and Models Decide How a City Moves

Traffic flow theory, from phantom jams and the fundamental diagram to signals and simulation — and what traffic modelling would mean for SmartCiti.X

Aerial illustration of a city street grid at dusk, with streams of warm golden light tracing the flow of traffic along the roads

Illustration: traffic as streams of light across a city grid at dusk. Generated illustration, not a photograph.

You are driving on an open motorway. There is no accident ahead, no roadworks, no lane closure, no visible reason to slow down — and yet the traffic slows, stops, crawls, and then, for no visible reason, speeds up again. You have just driven through a phantom traffic jam: a jam with no cause except the traffic itself. It is one of the strangest facts about cities, because a jam that appears from nothing can be understood, modelled, and — sometimes — prevented. This article is about the science that explains it, the models that predict it, and the question of what a city that understood its own traffic could do with that knowledge.

The science starts with a simple observation. Traffic is not a collection of individual cars. It is a flow, like water in a pipe, and it obeys laws that look more like physics than psychology. A single driver decides when to brake. A thousand drivers, together, produce waves that travel backwards through the traffic at a steady speed, as regular as ripples on a pond. The driver who brakes a little too hard does not just slow the car behind them. They start a wave that can travel for kilometres and last for half an hour, long after the original car has driven away. Understanding that wave — where it comes from, how fast it moves, and what feeds it — is the heart of traffic flow theory.

The jam that built itself: the circle experiment

The cleanest proof that jams can appear from nothing came from an experiment so simple it sounds like a prank. In 2008, a team led by Yuki Sugiyama at Nagoya University put 22 cars on a circular track 230 metres around, told the drivers to cruise steadily at 30 kilometres per hour, and watched. There was no bottleneck, no merging lane, no obstacle, and no reason for anything to go wrong. The cars drove smoothly for a while. Then a ripple appeared: one driver braked slightly, the driver behind braked slightly harder, and within minutes a jam had formed — a cluster of stopped cars on a road with nothing wrong with it. The jam then travelled backwards around the circle, against the direction of the traffic, while the cars themselves moved forward through it. The paper, published in the New Journal of Physics as “Traffic jams without bottlenecks,” settled an argument that had run for decades. Jams do not need a cause. They need only enough cars, driving closely enough, with human reaction times.

The experiment matters for a practical reason: it shows that the cause of most congestion is not the road. It is the drivers, or more precisely the interaction between drivers. A road can be perfectly built, perfectly maintained, and perfectly signed, and still jam, because the jam lives in the flow, not in the asphalt. That is bad news for anybody who thinks congestion can be solved by pouring more concrete, and good news for anybody who thinks it can be solved by changing how the flow behaves.

Three numbers that describe any road

Every road, at any moment, can be described with three numbers, and the relationship between them is the foundation of the whole science. The first is flow: how many vehicles pass a point in an hour. The second is density: how many vehicles occupy a kilometre of road. The third is speed: how fast they are travelling on average. The three are linked by a single equation, flow equals density multiplied by speed, and that equation is not a theory. It is arithmetic. If 50 cars occupy a kilometre and they travel at 60 kilometres per hour, then 3,000 cars pass a point each hour. The equation holds whatever the drivers are doing.

The interesting part is what happens when density rises. At low density, the road is empty, cars travel at whatever speed their drivers choose, and flow rises as more cars join. At some point the road reaches its capacity: the maximum flow it can carry, and the speed at which that maximum occurs. Add more cars beyond that point and something counterintuitive happens. Flow falls. The road carries fewer cars per hour when it is fuller, because the cars are moving so slowly that the extra density cannot compensate for the lost speed. Push density to its maximum — bumper to bumper, nobody moving — and flow falls to zero. Plot flow against density and you get a curve that rises, peaks, and falls: the fundamental diagram of traffic flow, the single most important picture in the science.

The first version of that picture came from Bruce Greenshields, who in 1935 published “A study of traffic capacity” in the proceedings of the Highway Research Board, based on photographs of traffic taken from the air. Greenshields proposed that speed falls linearly as density rises, and from that line the curved fundamental diagram follows directly. The linear assumption was a simplification, and later research refined it, but the shape survived: every road has a capacity, every road has a critical density where flow peaks, and every road, pushed past that density, carries less, not more.

Rivers of cars: the Lighthill-Whitham-Richards model

If traffic is a flow, then it can be modelled the way physicists model flows — and in 1955 and 1956, three researchers did exactly that. M. J. Lighthill and G. B. Whitham published their kinematic wave theory in 1955, and P. I. Richards published a closely related theory in 1956. Together, the Lighthill-Whitham-Richards model, universally known as LWR, treats traffic as a continuous fluid, governed by the conservation of vehicles: cars are neither created nor destroyed on a stretch of road, so any change in density must be matched by a change in flow. From that single conservation law, plus the fundamental diagram, the model predicts the most distinctive feature of real traffic: shockwaves.

A shockwave in traffic is a sharp boundary between two states — fast-moving light traffic on one side, slow-moving dense traffic on the other — and the LWR model says that boundary moves at a speed determined by the change in flow divided by the change in density across it. When a fast stream meets a slow queue, the boundary moves backwards through the traffic, exactly as the circle experiment showed. When the queue clears, a second boundary moves forward. The model’s power is that it predicts these movements from the fundamental diagram alone, without modelling any individual driver. Its limitation is the same as its power: it treats traffic as a fluid, and drivers are not water molecules. They hesitate, they overreact, they look at their phones, and they behave differently on a Monday morning than on a Sunday afternoon. The fluid approximation captures the waves. It misses the reasons the waves start.

The LWR model has been extended, refined, and argued over for seventy years, and it remains the backbone of macroscopic traffic modelling — the kind that treats traffic in bulk, the way a weather model treats air. Its endurance is a lesson in how science works: a simple model that captures the main phenomenon survives, not because it is complete, but because it is clear enough to be improved.

The four rules of a traffic jam: Nagel and Schreckenberg

The fluid model explains the wave. It does not explain the driver. For that, traffic science turned to a different kind of model, one that treats each car as an individual following simple rules. The most famous is the cellular automaton model of Kai Nagel and Michael Schreckenberg, published in 1992 in the Journal de Physique I. Their model divides a road into cells, each about the length of a car, and updates every car once per time step according to four rules.

Rule one, acceleration: if a car can go faster without hitting the car ahead, it speeds up by one unit. Rule two, slowing down: if the gap ahead is smaller than the car’s speed, it slows to fit the gap. Rule three, randomisation: with some probability, a driver slows down by one unit for no reason at all — a moment of inattention, a tap of the brake, a glance at a sign. Rule four, movement: each car moves forward by its new speed. That is the whole model. Four rules, no psychology, no weather, no roadworks.

And yet the model produces phantom jams. The randomisation rule is the key: a single random slowdown, in dense traffic, triggers a chain of braking behind it, and a jam forms and propagates backwards, exactly as in the circle experiment. Turn the randomisation off and the jams vanish. Turn it up and the road jams at lower densities. The model shows, with unusual clarity, that the jam is not caused by bad drivers or bad roads. It is caused by the combination of density and ordinary human variability. The practical conclusion is striking: the most effective way to prevent jams may not be to rebuild roads, but to reduce variability — through smoother driving, through cruise control, through coordinated speeds, or through vehicles that communicate with each other.

Braess’s paradox: when a new road makes everything slower

If the phantom jam shows that traffic can create problems from nothing, Dietrich Braess showed in 1968 that well-meaning engineers can create problems from something: a new road. Braess’s paradox, published in the German journal Unternehmensforschung, describes a network where adding a road — a shortcut, apparently — makes every driver slower. The mechanism is selfish routing. Each driver chooses the route that is fastest for them, given what everyone else is doing. Before the new road, the traffic splits between two routes and balances. After the new road opens, the shortcut looks attractive to everyone, everyone takes it, and the shortcut, plus the roads feeding it, become congested. The equilibrium after the addition is worse for every driver than the equilibrium before it. Nobody is behaving badly; everybody is behaving rationally, and the result is worse for all.

The paradox is not a curiosity. It has been observed in real networks, and it is the reason traffic engineers test changes in simulation before building them. A city that widens a road, opens a shortcut, or retimes a junction without modelling the network can make congestion worse while spending public money to do it. The paradox also has a mirror image: closing a road can make traffic better. When a road is removed, drivers redistribute, and the network can flow more freely than before. Each case depends on its own network, and no general rule replaces a model of the specific streets involved. The lesson is not that new roads are bad. It is that a network is not a set of roads. It is a system, and systems surprise the people who change one part without modelling the rest.

The conductor: how traffic signals orchestrate a city

If the road network is the orchestra, the traffic signal is the conductor — and most conductors in most cities are still working from a score written years ago. The simplest signals run on fixed time: green for forty seconds, red for thirty, repeating all day regardless of what the traffic is doing. Fixed-time signals are cheap, predictable, and often wrong, because traffic is not fixed. A junction that is empty at noon and gridlocked at six cannot be served well by a single timing plan.

The next step is actuated signals, which respond to the traffic they see. Inductive loops buried in the road surface detect waiting cars, and the signal extends the green while cars keep coming, or skips a phase when nobody is waiting. Actuation is a genuine improvement: the signal listens before it decides. But an actuated junction still thinks locally. It optimises its own corner, and a corner optimised in isolation can be a disaster for the street it sits on, releasing platoons of cars into the next junction just as it turns red.

The third step is coordination, and the fourth is adaptation. Coordinated signals along a corridor are timed so that a platoon travelling at the speed limit meets green after green — the “green wave” that drivers notice when it works and curse when it does not. Adaptive systems go further: they measure the traffic across the whole network, predict where queues will form, and adjust the timings continuously.

Two of the oldest and most widely deployed adaptive systems are SCOOT, the Split Cycle Offset Optimisation Technique developed by the Transport Research Laboratory in the United Kingdom, and SCATS, the Sydney Coordinated Adaptive Traffic System, developed in New South Wales. Both have been running in cities around the world for decades, and both work by the same principle: small, frequent adjustments, driven by measurement, rather than grand plans drawn up once and left to age. Signals also have to serve people who are not in cars. A pedestrian crossing, a bus lane, a cycle track, and a freight route all compete for the same seconds of green, and the signal plan is where a city’s priorities become concrete. There is no signal plan that serves everyone best. There is only a plan that reflects what the city values, written in seconds of green and red.

The rehearsal room: simulation

A city cannot experiment on itself. It cannot close a bridge for a month to see what happens, or retime every signal on a Tuesday to test a theory, without imposing the cost of the experiment on everyone who drives. Simulation is how a city rehearses. A traffic simulation is a model of a network — its roads, its junctions, its signals, its demand — run forward in time, so that a change can be tested in software before it is built in concrete.

Simulations come in three sizes. Macroscopic models, descended from LWR, treat traffic as a flow and are fast enough to model a whole city, but coarse enough to miss the details of a single junction. Microscopic models, descended from Nagel and Schreckenberg, simulate individual vehicles with car-following and lane-changing rules, and are detailed enough to show a queue forming at a specific turn, but slow enough that modelling a whole city can take hours. Mesoscopic models sit between the two, grouping vehicles into platoons. The choice is a trade-off between scale and detail, and a serious study uses more than one.

The most widely used open tool in the field is SUMO, the Simulation of Urban Mobility, developed at the German Aerospace Center (DLR) and released as open-source software under the Eclipse Foundation. SUMO is microscopic, it handles large networks, it models signals, public transport, pedestrians, and electric vehicles, and it is free — which is why it appears in research papers, city planning offices, and university courses around the world. A city that wants to test a new junction, a new bus lane, or a new signal plan can build the network in SUMO, feed it the demand measured from real counts, and watch the queues form before a single kerb is moved.

Simulation has its own discipline, and it is easy to get wrong. A model is only as good as its inputs: the demand figures, the turning counts, the signal timings, and the driver behaviour parameters. Feed it a demand figure from a quiet Tuesday and it will bless a junction that fails every Friday. The sound practice is validation: run the model on a period where the real outcome is known, and check whether it reproduces it, including the places where it fails. A simulation that is never validated is a video game with a city map. A simulation that is validated, with its errors published, is a tool a city can trust.

The eyes of the network: data, and what it costs

A model without data is a guess, and the last twenty years have transformed what a city can know about its own traffic. The oldest sensors are inductive loops, cut into the road surface, counting cars and measuring speed and occupancy, one lane at a time. Cameras add classification — car, bus, lorry — and, with automated number plate recognition, journey times between points. Bluetooth and Wi-Fi detectors track the passage of devices. GPS probes from phones and vehicles report speeds across the whole network, in real time, at a resolution no fixed sensor can match. Floating car data, as the probe feeds are called, turned every equipped vehicle into a sensor.

The power of that picture is real, and so is its cost. The same data that shows a city where the traffic is slow can show where a particular person was, and when. A licence plate read is a location record. A phone’s GPS trace is a diary. The difference between a traffic system and a surveillance system is not the sensor. It is the governance: what is collected, how long it is kept, who can see it, and what it may be used for. A city that builds a traffic model on probe data must answer those questions in public, before the sensors go in, because the answers are harder to give after the data exists. Aggregation is the usual defence — counts and speeds, not identities — and it is a good defence when it is enforced by design rather than promised by policy. The rule is simple to state and hard to follow: collect the least data that answers the question, keep it for the shortest time that serves the purpose, and delete the rest.

Induced demand: the road that fills itself

The most counterintuitive finding in traffic science is also the most consequential for any city thinking of building its way out of congestion. When a road is widened, traffic grows to fill it. The mechanism is not mysterious. A faster road attracts drivers from other routes, other times of day, and other modes of transport. People who avoided the trip make it. People who took the train drive. People who travelled off-peak travel at peak. The road fills, speeds fall back, and the congestion returns, now serving more vehicles. The phenomenon is called induced demand, and it is one of the best-documented effects in transport economics.

The strongest evidence comes from Gilles Duranton and Matthew Turner, whose 2011 paper in the American Economic Review, “The Fundamental Law of Road Congestion,” studied US cities and found that vehicle kilometres travelled rise in proportion to the road capacity provided — more lanes, more driving, with congestion roughly unchanged. The finding has been replicated and debated, and the magnitudes are argued over, but the direction is not. A city that widens a road to relieve congestion is, in the usual case, buying a larger congested road.

The implication for SmartCiti is direct, and it is the reason the traffic modelling work matters before any road is drawn. A city that models first can see induced demand in simulation, before spending money on concrete. It can test the alternative — signal coordination, bus priority, demand management, land-use changes — and compare the outcomes on equal terms. The model does not make the decision. It makes the decision visible, with numbers, before the money is committed.

What the models cannot do

Traffic models are powerful, and they have limits that must be stated alongside their powers, because a model that is trusted beyond its limits is worse than no model at all. The first limit is behaviour. Models assume drivers behave as they behaved in the past, and drivers change. A new phone app, a new delivery pattern, a change in working from home, a fuel price shock — each one rewrites the demand the model was calibrated on, and the model does not know it has happened until the data tells it. The second limit is the network itself. A model of the roads is not a model of the city. Land use, housing, jobs, schools, and shops determine where people need to go, and a traffic model that takes demand as given cannot say whether the demand itself is the problem. The third limit is uncertainty. A model produces numbers with the appearance of precision — 3,412 vehicles per hour, an average delay of 47 seconds — and the precision is misleading. The inputs are estimates, the parameters are fitted, and the outputs carry error bars that are rarely printed alongside them.

The fourth limit is the one that matters most for a city. A traffic model optimises what it is asked to optimise. Ask it to minimise vehicle delay and it will produce a city of fast cars and long pedestrian waits. Ask it to minimise total person delay and it will favour the bus carrying forty people over the car carrying one. Ask it to minimise emissions and it will favour smooth flow over high speed. The model is neutral. The question is not. Every traffic model carries, hidden in its objective function, a statement about what the city values, and the most important part of any modelling exercise is to drag that statement into the open and argue about it, in public, before the model is run.

The SmartCiti research programme: what is proposed, and what is not

The traffic modelling work described in this article connects to SmartCiti.X, Series 10 in the W3 list, as proposed research. The proposal is to build the modelling capability before building anything physical: a validated simulation of a street network, calibrated on measured demand, used to test signal plans, junction designs, bus priority, and parking policy before any of them are built. The research would sit alongside Parking Smart-X, Series 12, because parking and traffic are the same problem seen from two ends: a car that cannot find a space becomes a car circling the block, and a circling car is traffic.

What is not proposed is equally important. No city has been named. No site, address, price, seller, or union has been identified. No traffic system has been built, no signal has been retimed, and no road has been drawn. The programme’s suggested figures for SmartCiti.X — a $25,000,000 suggested pre-money figure and a $10,000,000 first raise tranched as $5,000,000 first and $5,000,000 on milestones — are planning figures in internal documents, not prices, not offers, and not amounts anyone has committed. The first raise, if it ever happens, would fund robotics training, union trades AR/VR, robotics manufacturing, and site acquisition and fit-out in Louisiana and Indiana — and no site, city, address, price, seller, or union is named in those documents either.

The programme’s own record states the gap plainly. The traffic-modelling document has no designated repository, which means there is no codebase, no dataset, and no simulation to point at today. That is a gap, and this article records it rather than filling it with a description of work that has not happened. When the work begins, its first deliverable should be a repository with a name, a validated model, and a public record of what the model gets right and what it gets wrong. Until then, the traffic modelling for SmartCiti.X is a plan to do science, not science that has been done.

Parking: the last hundred metres

Parking deserves its own section, because it is where traffic modelling meets daily life most directly. A driver searching for a parking space is not parked. They are driving, slowly, in the most congested part of the network, often circling the same blocks. The classic study, by Donald Shoup, measured cruising for parking in a Los Angeles commercial district and found that drivers spent substantial time circling, adding measurable traffic to streets that were already full. Shoup’s book, The High Cost of Free Parking, published in 2005, made the case that the price of parking — usually zero, or close to it — is the cause: when parking is free, demand exceeds supply, and the excess demand becomes circling traffic. Price the parking to keep a space or two free on every block, Shoup argued, and the circling disappears, because a driver who can see a space will drive to it directly.

The modelling consequence is that parking cannot be treated as an afterthought to traffic. A network model that ignores parking will under-predict congestion in commercial districts, because it misses the circling vehicles. A parking model that ignores traffic will over-predict the benefit of a new garage, because it misses the vehicles the garage attracts. The two must be modelled together, and Parking Smart-X, as a proposed series, is the place in the programme where that joint modelling would live. The series is proposed, not formed, and its work is research, not construction.

The connected future: vehicles that talk

The most consequential change in traffic science is not a new model. It is a new data source: vehicles that communicate. Vehicle-to-vehicle and vehicle-to-infrastructure communication — V2V and V2I, together called V2X — would let cars share their speed, position, and intentions with each other and with the signals, in real time. A car that knows the car ahead is braking can brake sooner and more gently, and a signal that knows a platoon is coming can hold the green. The Nagel-Schreckenberg model predicts the benefit directly: reduce the randomisation — the variability in driver behaviour — and the jams form later, at higher densities. Connected vehicles are, in the model’s terms, a way of turning the randomisation down.

The technology is real, the standards exist, and the deployment is slow, because it requires coordination across manufacturers, road operators, and regulators, and because the benefits arrive only when enough vehicles are equipped. A single connected car in a stream of unconnected cars gains little. A stream in which half the cars communicate gains a great deal. That threshold — the penetration rate at which the network effect appears — is itself a modelling question, and it is one of the questions a SmartCiti simulation would be built to answer.

The governance questions from the data section return here, sharper. A vehicle that broadcasts its position is a vehicle that can be tracked, and a network that knows every vehicle’s location knows a great deal about its citizens. The design choices — what is broadcast, to whom, for how long, and under whose control — are not technical details to be settled after deployment. They are the deployment.

How to read a traffic claim

The reader who has followed this far now has a checklist for any traffic claim, whether it comes from a city, a developer, or a programme like this one. First, ask for the fundamental diagram: what is the road’s capacity, what is its critical density, and is the road above or below it? Second, ask whether the change was tested in simulation, and whether the simulation was validated against real data, with its errors published. Third, ask about induced demand: if the proposal adds capacity, what is the evidence that the new capacity will not simply fill? Fourth, ask who benefits and who waits: does the proposal minimise vehicle delay or person delay, and who made that choice? Fifth, ask about the data: what is collected, how long is it kept, and who governs it? Sixth, ask what is filed and what is proposed, and insist on the two being kept apart.

Traffic is the most visible thing a city does, and the least understood. Every driver experiences it, every driver has a theory about it, and most of those theories are wrong in the same way: they blame a cause — the lights, the lorries, the roadworks, the other drivers — when the cause is the flow itself. The science replaces blame with mechanism: jams form from nothing, new roads can make things slower, widening a road fills it, and the most powerful tool a city has is not concrete but understanding. A city that understands its traffic can choose. A city that does not can only react. The difference between the two is the difference between a city that thinks and a city that merely moves.

Glossary

  • Flow: The number of vehicles passing a point in an hour.
  • Density: The number of vehicles occupying a kilometre of road.
  • Fundamental diagram: The curve relating flow to density on a road, rising to a peak at capacity and then falling.
  • Critical density: The density at which a road carries its maximum flow; beyond it, flow falls.
  • Shockwave: A sharp boundary between two traffic states that moves through the flow, usually backwards.
  • Phantom jam: A traffic jam that forms without any bottleneck, from the interaction of drivers alone.
  • LWR model: The Lighthill-Whitham-Richards model (1955–56), which treats traffic as a continuous fluid governed by conservation of vehicles.
  • Cellular automaton: A model, such as Nagel and Schreckenberg’s (1992), in which each vehicle follows simple rules on a grid of cells.
  • Braess’s paradox: The finding (1968) that adding a road to a network can make every driver slower.
  • Induced demand: The effect by which added road capacity generates additional traffic, leaving congestion roughly unchanged.
  • SCOOT / SCATS: Adaptive traffic signal systems — SCOOT from the UK’s Transport Research Laboratory, SCATS from Sydney — that adjust signal timings continuously from measured traffic.
  • SUMO: Simulation of Urban Mobility, an open-source microscopic traffic simulator developed at the German Aerospace Center (DLR).
  • V2X: Vehicle-to-everything communication, covering vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) messaging.
  • Floating car data: Speed and position data reported by equipped vehicles and phones, used as moving traffic sensors.

Frequently asked questions

What is a phantom traffic jam? A jam that forms with no bottleneck — no accident, no roadworks, no lane closure — from the interaction of drivers alone. It was demonstrated by Sugiyama and colleagues in 2008, with 22 cars on a 230-metre circular track, and it is predicted by the Nagel-Schreckenberg model.

Why does adding a lane sometimes not help? Two reasons. Past the critical density, a fuller road carries less flow, not more. And induced demand means new capacity attracts new traffic, so congestion returns at a higher volume. Duranton and Turner’s 2011 study of US cities is the key evidence.

Can closing a road improve traffic? Sometimes, yes — that is the mirror image of Braess’s paradox. When a road is removed, drivers redistribute, and the new equilibrium can flow better than the old one. Whether it does depends on the specific network, which is why the change must be modelled first.

What is the fundamental diagram? The curve relating traffic flow to density on a road. Flow rises with density up to a peak — the road’s capacity, at its critical density — and then falls as density rises further, reaching zero when the road is bumper to bumper.

What is SUMO? The Simulation of Urban Mobility, an open-source microscopic traffic simulator developed at the German Aerospace Center (DLR). It models individual vehicles, signals, public transport, and pedestrians, and is widely used to test changes before they are built.

Does SmartCiti have a traffic system? No. Traffic modelling for SmartCiti.X is proposed research only. No city has been named, no site has been identified, no system has been built, and the programme’s traffic-modelling document has no designated repository.

Do traffic models track individual drivers? They do not need to, and they should not. The models in this article work on counts, speeds, and densities. Probe data from phones and vehicles can identify individuals, which is why collection, retention, and governance must be decided in public, before the sensors go in.

What is the single most useful thing a city can do about congestion? Measure it, model it, and test changes in simulation before building them. The science does not offer a single fix. It offers a way to compare fixes with evidence instead of argument.

Further reading on this site

Related content elsewhere

Sources

  • Sugiyama, Y., et al. “Traffic jams without bottlenecks — experimental evidence for the physical mechanism of the formation of a jam.” New Journal of Physics 10 (2008) 033001. IOPscience
  • Greenshields, B. D. “A study of traffic capacity.” Highway Research Board Proceedings 14 (1935) 448–477.
  • Lighthill, M. J., and Whitham, G. B. “On kinematic waves II: A theory of traffic flow on long crowded roads.” Proceedings of the Royal Society of London A 229 (1955) 317–345.
  • Richards, P. I. “Shock waves on the highway.” Operations Research 4 (1956) 42–51.
  • Nagel, K., and Schreckenberg, M. “A cellular automaton model for freeway traffic.” Journal de Physique I 2 (1992) 2221–2229.
  • Braess, D. “Über ein Paradoxon aus der Verkehrsplanung.” Unternehmensforschung 12 (1968) 258–268.
  • Duranton, G., and Turner, M. A. “The Fundamental Law of Road Congestion: Evidence from US Cities.” American Economic Review 101 (2011) 2616–2652. American Economic Association
  • Shoup, D. The High Cost of Free Parking. American Planning Association, 2005.
  • SUMO — Simulation of Urban Mobility (Eclipse Foundation, developed at the German Aerospace Center, DLR). SUMO documentation
  • Traffic flow — the fundamental diagram, shockwaves, and the LWR and cellular automaton models. Wikipedia: Traffic flow
  • Braess’s paradox — statement, mechanism, and examples. Wikipedia: Braess’s paradox
  • Nagel–Schreckenberg model — the four rules and the emergence of phantom jams. Wikipedia: Nagel–Schreckenberg model
  • Traffic wave — stop-and-go waves and their propagation. Wikipedia: Traffic wave
  • Induced demand — mechanism and evidence. Wikipedia: Induced demand
  • W3 LLC internal list, April 2026 (33 named sub-series, Series 1 to 33; SmartCiti.X is Series 10, Parking Smart-X is Series 12). Local file.
  • AGI Future Foundation, internal document complete/32 (suggested valuations and raises). Local file. All figures cited are suggested planning figures.
  • AGI Future Foundation, internal documents complete/16 (Venture Studio Operating Model) and complete/17 (Stage Gates). Local files.
  • California Secretary of State, business filings: AGI Future Foundation, Entity No. B20260196664, Articles filed 2026-04-24, approved 2026-04-30; AGI Fellowship, Entity No. B20260229217, Articles filed 2026-05-13, approved 2026-05-18. California Secretary of State, bizfile

← The Venture Studio, Stage by Stage

← All articles