Street Networks & Grids

Start with the network itself, because it is the most durable thing a city has. A street network is the connected system of routes - streets, lanes, paths - through which everything and everyone moves. It is the skeleton: blocks are the leftover spaces between streets, plots subdivide the blocks, buildings fill the plots, and the pattern of daily life follows the streets. Networks outlive buildings by centuries, so the network you lay down is the longest-lasting decision in the whole project.

Networks come in three broad kinds. The grid is planned and regular - a repeating pattern of streets crossing at intervals, from the Roman castrum to Manhattan to the ordered blocks of Jaipur's old walled city or Chandigarh's sectors. Its virtues are legibility (you can find your way), capacity, easy subdivision, and adaptability - a generic grid accepts many different uses over time. Its vice is that a grid imposed without care can be monotonous, hostile to terrain, and indifferent to how people actually gather. The organic network grew with no single author, accreting over centuries through countless small decisions: the winding lanes of a medieval town, the dense fabric of an old Indian city, the intricate paths of an informal settlement. Its genius is fine-grained adaptation to slope, sun, use and life - a complexity tuned to real living that no planner drew. Its failures are legibility, capacity for large infrastructure, and sometimes equity.

Most real cities are hybrid: a structuring frame of larger routes with finer, more organic fabric woven between - Barcelona's Eixample grid meeting the tangled Gothic Quarter, or an Indian city where a planned arterial grid holds together dense organic mohallas. This matters for computational urbanism because each type is a different problem for an algorithm. A grid is trivial to generate and easy to measure. An organic network is what procedural methods (L-systems, agent growth) try to *imitate* - and what the intricate, humane fabric of a real grown city usually beats. The honest framing: computation can lay a skeleton fast, but whether that skeleton can carry a living city is a question the graph alone cannot answer.

A network is not just a picture; it has *properties* that decide how well it works, and two matter most: connectivity and hierarchy.

Connectivity is how many ways there are to get from one place to another. A network of many small blocks and frequent junctions gives short, direct, walkable routes and countless choices - if one street is blocked or dull, you take another; movement spreads; the street-corner economy of shops and encounters thrives. A network of long blocks, cul-de-sacs and few junctions forces every trip onto a handful of arterials, lengthens walking distances until people drive, concentrates traffic, and starves the in-between streets of life. This is one of the most robust findings in urban design: fine-grained, highly connected networks tend to support walking, mixing and street life; coarse, poorly connected ones tend to produce car dependence and dead streets. Block size is the lever - roughly, smaller blocks mean more connectivity.

Hierarchy is the ordered range of street types, from quiet lanes up through local streets and avenues to arterials, each sized to its role. A good hierarchy lets a network carry heavy through-movement on a few routes while keeping most streets calm enough to live on. The classic failure is a *rigid* hierarchy - the modernist tree of cul-de-sacs feeding collectors feeding one arterial - which maximises separation and minimises connectivity, producing exactly the car-bound, lifeless pattern above. The subtler truth, argued by Jane Jacobs and by the tradition of network analysis, is that the healthiest cities usually have a *connected* hierarchy: a legible frame of bigger routes, but with the fine network between them staying richly interlinked rather than broken into isolated pods.

This is where computation earns its place honestly. Network analysis - drawn from network theory and the urban tradition of space syntax - can measure these properties precisely on a real city: connectivity, block length, route directness, and measures like *integration* and *betweenness* that predict where through-movement concentrates and, remarkably, where street life and retail tend to cluster. That is genuine, useful, evidence-generating knowledge. But keep the discipline: these measures describe *potential* for movement, not whether a street is loved, safe, or alive. The graph is real; it is not the whole street.

This is the module where computation is genuinely at its strongest, so name what it does well. On the generative side, algorithms can grow street networks: procedural methods (L-systems, agent-based growth, tensor-field methods) produce plausible networks that respond to terrain, a river, a density gradient or a set of rules, and can populate a large site with a coherent network in seconds. Parametric models let you tune the network - block size, grid angle, spacing, how density rises toward a transit stop - and watch the whole skeleton re-form instantly, so you can explore a family of layouts rather than labour over one. Generative search can propose thousands of network variants and test each against goals. For handling scale, exploring options, and grounding a first layout in terrain and flow, this is a real and time-saving power.

On the analytical side - the more mature and trustworthy use - computation measures an *existing* or proposed network rigorously. Space-syntax and network-theory tools compute connectivity, integration, betweenness, choice, block size and route directness across an entire city, revealing where movement will concentrate, which quarters are cut off, where a new link would knit an isolated area back in, and where the network structure predicts street life and retail. Walkability and access analysis measures how many people can reach a park, a school or a station on foot within a given time. This is powerful, defensible, evidence-generating work that can genuinely improve a plan and expose injustice - a neighbourhood the network has stranded - that a drawing might hide.

Hold two honesties, though. First, a procedurally generated network can *look* like organic fabric while lacking what makes real organic fabric humane - it mimics the geometry, not the centuries of adaptation to real life. Second, every generative and analytical result is only as good as its rules and its data, and rewards exactly what it was told to measure. A network optimized purely for traffic throughput or for a walkability *score* can hit its target and still produce streets no one wants to be on. Use computation to lay skeletons fast and to analyse them honestly - then judge the result as a place, not a graph.

Now the discipline the whole course insists on. The street network is a graph - nodes and links you can measure - but the street is a place, and almost everything that makes a street worth walking is invisible to the graph. The analysis can tell you a lane is highly integrated and carries through-movement; it cannot tell you that the movement is there because of the tea stall on the corner, the shade of one old tree, the shopkeeper who watches the children, the doorway conversations, the fact that women feel safe there after dark - or that they do not. Connectivity is a precondition for street life, not street life itself. Optimize a network hard for a single measurable - traffic flow, a directness index, a walkability score - and you can hit the number while quietly destroying the corner where life actually happened, producing a network that scores beautifully and is dead to walk.

And the network is never neutral. A new road that scores well on through-movement may cut a poor settlement in two, or wall off a community behind an arterial it cannot cross - the classic tool of exclusion, dressed as connectivity. What the model counts as a "street" is itself a choice: the informal lanes, the shortcuts through a settlement, the paths that are not in the official dataset simply do not exist to the algorithm, so a naive generative masterplan can lay a grid straight over living fabric it cannot see. In India especially, where an enormous share of movement happens on fine organic lanes and informal paths that no clean dataset captures, the network you *measure* can silently erase the network people actually *use*.

So the competent stance is the course's stance. Use network analysis powerfully - it is one of the most genuinely useful things computation does in urbanism, and it can reveal severed neighbourhoods and injustice a drawing would hide. Generate skeletons fast to explore options. But hold three truths: the graph is not the street; a network optimized for a metric can still be lifeless; and the choice of which network to build - which areas to connect, which to sever, whose movement counts - is a political and democratic decision, belonging to the planning authority, the affected communities and the statutory process, never to the algorithm.