GENERATIVE URBAN MODELING: A DESIGN WORK FLOW FOR .
Submitted to SimBuild 2012Rakha & Reinhartpg.1GENERATIVE URBAN MODELING:A DESIGN WORK FLOW FOR WALKABILITY-OPTIMIZED CITIESTarek Rakha1 and Christoph Reinhart11Massachusetts Institute of Technology, Cambridge, MAINTRODUCTIONExpanding urban grids and massing is a process that isoftentimes unplanned in informal settlements. Localgovernment and planning authorities routinely face thischallenge with very limited, if any, budget. Hence, thereis a pressing need to develop urban design workflowsthat support a smarter approach towards street gridsubdivision and generation of urban massing thatconsider environmental performance. The purpose ofsuch workflows is to enable the evaluation of multipledesign iterations and optimize for certain performancecriteria, such as resource efficiency and resident’shealth and comfort. In this day and age, designcomputation has become ubiquitous throughout thedesign world, from small scale offices to multinationalfirms. Given the ever growing power of personalcomputers and the increasing use of cloud computing,workflows based on such technologies can thus helpdesign teams throughout the world to develop low-techurban solutions using high-tech design tools.Cities are growing exponentially across the globe.Studies report that the cumulative change in urbanexpansion for the period of 1970 to 2000 was 58,000km2, which is approximately in the order of 2% of theglobal urban land area in 2000 (Seto et al., 2011). TheUnited Nations’ latest figures demonstrate that by theyear 2100, the world population is projected to reach10.1 billion (United Nations, 2011). Accordingly, newneighborhoods are being built every day; pushing thedefinition and boundaries of cities, which significantlydecreases urban densities (Angel et al., 2010) andcontributes considerably to carbon emissions (Hutyra etal., 2011). This expansion tends to take place at theoutskirts, where the terrain morphology is often lessbenign to urban developments due to irregularities inthe landscape. This expansion process necessarilyinvolves the planning of road networks that will, withcertain reasoning, adapt to that terrain. Interestingly, aroad network, once in place, tends to be remarkablyresistant to change as exemplified by a visualcomparison of part of Egypt’s capital, Greater Cairo’sdowntown core (Zamalek, Tahrir and Garden City) in1933 and today (Figure 1).Although generative tools for urban form werepreviously investigated computationally (Beirão et al.,2011; Luca, 2007) and in terms of certainenvironmental performance criteria (Oliveira Panão etal., 2008; Keirstead, et al., 2011), site design and itsrelationship to terrain in the third dimension has thus farbeen disregarded. Given the likeliness that newdevelopments increasingly take place in non-flatterrains, this paper presents a new urban analysisworkflow that develops street and massing layouts fornew neighborhoods in such environments. A parametricurban massing tool was developed in theRhinoceros/Grasshopper environment that allows urbanenvironmental master planning to take place within athree-dimensional terrain elevation model. The tool canbe linked to a number of existing environmentalperformance analysis tools in Rhinoceros/Grasshopperthat include operational building energy use, access tosolar radiation and daylighting. In this particular study,the urban massing component has been linked with anew walkability calculator. Walkability was consciouslychosen as an initial sustainability performance indicator,since planning of urban density is a necessary step toABSTRACTThis paper presents an urban analysis work flow using aRhinoceros/Grasshopper massing tool. The tool utilizesterrain elevation models as part of the design process tosubdivide sites and generate urban form to be exploredparametrically. It can then be linked to variousperformance assessment methods. As a proof ofconcept, the study uses a walkability calculator for threeurban form alternatives, and applies genetic algorithmsto optimize generated designs through allocation ofland-use. Results show a great diversity that convergesto near optimal solutions. A discussion is drawn aboutthe effort and time spent to model such iterations versusit’s automation using this work flow, and conclusionsshow the potentials, limitations and directions for futureresearch work.
Submitted to SimBuild 2012Rakha & Reinhartpg.2Figure 1 A comparison between minimally changed street structures in downtown Cairo, Egypt.(Left) Author adapted map of Cairo in 1933 (Nicohosoff, A., 1933).(Right) An online contemporary map of the same area (Bing Maps, 2011).contain urban growth. It constitutes a key challengeto Engineerassignment to accommodate various programmaticCivilsustainable urban developments worldwide as explainedneeds (housing, commercial, green areas, etc.). Theabove. The paper describes details of the urban massingproposed tool utilizes this form generation processtool, walkability calculations and optimizationcomputationally through the following steps:procedures along with an example case study.1. Load terrain elevation map (Figure 1).URBAN FORM GENERATIONMETHODOLOGY2.Iteratively subdivide terrain following design logic.3.Manipulate the terrain for build-ablity (Terraform).The proposed workflow for the conception of urbanform is twofold: Firstly, an exploration of parametricmassing is performed using a generative street divisionand urban massing tool. In a second step, a walkabilitycalculation is applied to the resulting street grid toevaluate the potential walkability of the design.4.Set street widths offsets and building lots.5.Zone parametrically controlled building forms.Generative Urban Form WorkflowGeneration of urban form in its primary stages typically,but not necessarily, involves the subdivision of adevelopment plot area using a certain design rationale.From this subdivision, street networks are planned andland lots are assigned setbacks and massing heightlimitations. This is coupled with land-use zoningIn this paper, terrain subdivision logic is based onutilizing an orthogonal brute-force search for minimumslopes with control on minimum lot size in pixel values(Min Lot). The code determines whether the giventerrain is in the orthogonal horizontal or vertical sense,and slopes are calculated in the opposing sense bysubtracting the lowest elevation height from the highestone in each pixel row. This determines build up“blocks” that interface with the design of walkablestreets, which is a performance metric to be optimizedlater in the assessment process. Figure 1 showsFigure 1 Arbitrary elevation map converted from pixels to a terrain model. Subdivisions (Div) are parametric
Submitted to SimBuild 2012Rakha & Reinhartsubdivision slider-controlled iterations (Div) in theGrasshopper definition, limited by conditionalminimum lot sizes and the divided blocks’ orientation.pg.3generated neighborhood is presented in Figure 3. Thegenerated urban form that is adapted for the terraincondition can now be tested and optimized for variousperformance metrics. In this study, the evaluation ofhow “walkable” a neighborhood can be is undertaken,and the appraisal methodology is presented next.Figure 3 Example neighborhood and massing optionsWalkability AssessmentFigure 2 Terraform ProcessesBuilding lots are then terraformed through two options:flat areas that maintain an average elevation betweenthe four corners of each lot in the terrain, or a bilinearinterpolation of the elevation of those same corners(Figure 2). Street offsets are directly proportional to lotsize, and are slider-controlled as well. Building formsare parametric in depth and height, and follow threemassing options that emulate typical urban typologies(Figure 3). By defining 2D geometry in Rhino, the usercan link these geometrical “zones” to massing optionsto act as a land-use allocation tool. The tool is thereforeused to explore massing parametrically in the earlyurban design and planning stages. An exampleThe evaluation of neighborhood walkability and itsrelationship to human health and carbon emissions hasbeen the subject of numerous publications (Hoehner etal., 2011; Frank et al., 2010). Any chosen scheme toassess the walkability of generated neighborhoods willbe supported by the workflow’s current designrationale. Since the subdivisions are based on minimumslope, the produced streets will have the lowest slopesthat insure less effort in walking activities.In this paper, the validated (Carr et al. 2011) “StreetSmart” walk score algorithm was utilized to assess thewalkability of generated urban form. Street gridsgenerated from the tool are linked to a Grasshopperwalk score definition. It is assumed that each block willhost a multi-functional building with housing. Differentamenities are randomly placed on the grid, and thedefinition utilizes a shortest path script that is based onthe A* algorithm to compute distance to surroundingamenities. A score between 0 and 100 is then given toeach housing point based on the walking distances tothe following land-use categories:
Submitted to SimBuild 2012Rakha & Reinhartpg.4"restaurants": [.75, .45, .25, .25, .225, .225, .225, .225,.2, .2],score. Given that there are an enormous number ofcombinations possible, an exhaustive search is notpractical. The optimization problem is suited for theutilization of evolutionary algorithms. This step isdescribed next."shopping": [.5, .45, .4, .35, .3],Optimization"coffee": [1.25, .75],The method used for land-use allocation optimization inthis research was a Genetic Algorithm (GA). It is ascheme that imitates evolutionary processes throughsimulating procedures of population, crossover andmutation of competing solutions. A GA is commencedwith randomly chosen locations for amenities (genes),creating parent solutions of zoning (chromosomes) froma controlled search space to create an initial population.Within each chromosome, housing egress has a walkscore (W) generated based on the location of genes.Those walk scores are tested for the followingconditions:amenity weights {"grocery": [3],"banks": [1],"parks": [1],"schools": [1],"books": [1],"entertainment": [1],}Assigned weights for amenities are the numbers placedafter each category. Multiple numbers denote the scoreother amenities of the same type get after the first count.A polynomial distance decay function is used. It gives afull score for amenities that are within quarter mile ofhousing egress. Walk scores beyond this decrease withdistance. At a distance of one mile, amenities receiveabout 12% of the score. After one mile, scores slowlydecrease with greater distance. Other penalties for lowstreet intersection densities and average block length arealso factored into the score (Walk Score, 2011). Thetotal sum of the weights listed above is 15. However,the walk scores are linearly expanded to range from 0 to100. Table 1 demonstrates the meaning of the computedwalk scores.Table 1 Definition of Walk DESCRIPTIONWalker's ParadiseDaily errands do not require a car.Very WalkableMost errands can be accomplished on foot.Somewhat WalkableSome amenities within walking distance.Car-DependentA few amenities within walking distance.Car-DependentAlmost all errands require a car.The walkability of an overall neighborhood andindividual locations within it depend on where theabove mentioned amenities are located. In order to findthe potential for walkability within a neighborhood theamenities should therefore be distributed so that themajority of housing units receive maximized walk-If W minWThen N 0-Else if minW W aWThen N N ((W-minW))/aW-Else if W maxWThen N N 1Where (minW) is the minimum W that would beconsidered acceptable, (aW) is the threshold of anacceptable walkscore, and (maxW) is the maximumsatisfactory walk score. In this study, minW 50, aW 69 and maxW 70 according to corresponding valuesin table 1. N is a placeholder of performance initiated asa zero value number. The population evolves towardsbetter chromosomes by applying the following fitnessfunction:-f(x) N/n(1)Where (n) is the number of housing egress points testedduring the population. The function evaluates theperformance of each chromosome, to be chosen asparents later to generate a new population. “Survival ofthe fittest” is applied through random selection that isweighted towards chromosomes of better performance.As a process of evolutionary search-and-find, twochromosomes are chosen for either operation ofcrossover or mutation. This populates new generationsto be tested and reselected, and through manygenerations, the chromosomes within the finalpopulations are near optimal.
Submitted to SimBuild 2012Rakha & ReinhartURBAN PERFORMANCE APPLICATIONAs an example application of the method, an arbitraryhilly site with an area about 1.45 km2 with maximumelevation difference of 360m was chosen as a virtualplatform for urban modeling. Three street divisionswere generated as shown in figure 4. The aim was tosimulate equal population densities (21600 people) indifferent urban form configurations. The “light” settingrefers to minimizing site subdivisions, giving higheremphasis on massing height and grouping functionality(27 buildings, with 800 people / building). The “dense”configuration suggests smaller lots with a compactmassing (150 buildings, with 144 people / building).The “moderate” is a contrast between both settings (82buildings with 144 people / building and 14 buildingswith 700 people / building). For the example, amenitieswere chosen to be of great challenge to the site area,and were as follows: 2 Grocery, 3 Restaurants, 3Shopping, 1 Bank, 1 School, 1 Books, 2 Entertainmentand 2 Coffee. The park areas were pre-selected for eachscheme. Figure 5 demonstrates an example walk scoreanalysis for arbitrarily placed amenities in the lightconfiguration. It shows that land-use zoning affectswalk scores considerably.pg.5Optimization was implemented through a tool inGrasshopper named Galapagos, an evolutionary solverthat utilized a GA to optimize the walkability of thethree explored urban massing options. The GA evolvedzoning for the cases through 50 iterations, controlled byproducing 50 populations/iteration. Figure 6 shows theland-use placement results on the generated grids of thenear-optimal solutions, and the resultant walk score forpre-generated housing egress.The results explored by the GA showed a great diversityin the imitation of each run. This eventually convergedto reveal near-optimal zoning in the differentconfigurations.Tested fitness reached minimumbounds between 0.1 and 0.4, which shows how aneighborhood could have poor walk scores if notcarefully planned. However, the maximum fitnessreached in the light setting was 0.842, and in themoderate 0.719 and 0.828 in the dense. This satisfied anoverall neighborhood evaluation, but if examinedcloser, may not be a pleasant setting for all individuallots. The full tabulated optimization results are shownin Figure 7. The optimization process ran approximatelyfor 15, 60, 240 minutes for the light, moderate anddense configurations respectively on a laptop equippedwith an Intel Core i7 2.8GHz CPU and 8 GB RAM.Although optimized solutions varied in the three casesin terms of land-use placement, they all shared acommon feature: the calculated centroid of the threesolutions was almost central to the arbitrary terrainmodel. While it may be intuitive to create diversity byspreading functionality across a development site, thisconsistent result shows that having a neighborhoodcenter that assembles varying zones improveswalkability significantly.Figure 5 Example generated Walk ScoresImportant amenities that give higher scores, such as“Grocery”, spread out in all sites to give equality acrossthe housing egress points. In all cases, some points onthe outskirts do not receive the minimum acceptableFigure 6 Optimized land-use allocations and consequent walk scores for generated housing egress
Submitted to SimBuild 2012Rakha & Reinhartwalk score. However, in such cases, if entrances tobuildings change, it will achieve a better score that maybe acceptable. Optimization shows performancedirections, yet it should be used with flexibility.pg.6Figure 8 shows population percentage plotted againstwalk scores. It demonstrates that 65-70% of the peopleliving in all scenarios receive a walk score higher than70; making them living in a neighborhood that is “verywalkable”. The remaining population lives in situationsthat are mostly “car dependent”. The visualization ofensuing massing options is shown in figure 9. Massingmodels were generated based on the optimized walkscores. The light design scenario adapted LeCorbusier’s approach to urbanism: “towers in the park”,with the heart as bigger towers to accommodate allamenities. The dense configuration was generated as acompact neighborhood with central “down-town” areathat is proportionally larger, and the intermediate was aset as a gradient between both. The variation inperformance between the three configurations is slight,but favoring the “light” scenario. Reasons for that arediscussed next.DISCUSSIONThe utilization of automation procedures to generateform gives unlimited degrees of freedom to designexploration. When applied to urban design, inquiriesinto performance become more delicate. Theinvestigation of urban form is taken from amorphological approach to a performative one; aquestion the designer must ask is: what are the urbanqualities we seek through the act of design?Figure 7 Iterations against fitness function inwalkability optimizationThe employment of the current minimum slope designrationale combined with the utilization of a numericevaluation of walkability, such as walk score, makes thequantitative optimization of the problem successful.However, disregarding terrain when calculating thewalk scores is a weakness, and the development ofnumeric penalties for reaching amenities that are higherin elevation, and where the shortest route may be “hilly”should be taken into consideration. In addition, thescoring system is street dependent, meaning thatwalking distances from the housing unit to the street areignored. This makes the “light” configuration performbetter, although in reality a distance from the building to50%45%Population ight7%019%48%26%WalkscoreFigure 8: Population percentages against Walk score
Submitted to SimBuild 2012Rakha & Reinhartthe street should be taken into account and wouldinfluence walk score dramatically.pg.7Cultural adaptation of the work flow should beconsidered. The current choice of amenities reflectsaverage North American interests. However, relativelyimportant destination points, such as location of water,should replace certain amenities when the value of suchlocations is considered vital.CONCLUSIONThe analysis of the previous research results showed thepotential and limitations of this workflow. The toolsuccessfully explored urban form in hilly situationsusing Grasshopper, which is an accessible, user friendlyplatform for parametric investigations. This makesinvestigations into massing particular to non-flat terrainscenarios achievable and flexible.This work flow highly complements current paralleldevelopments in urban modeling environments. Thepresented application utilizes performance placeholdersfor the ability of the tool to question urban metrics. Forfurther development, it is suggested to investigate theutilization of optimization schemes to be urban formfinders. A number of competing fitness attributes couldbe studied, such as neighborhood operational energyuse, urban daylight availability, fluid dynamics of windand consequent ventilation, or walkability andbikeability schemes, to name a few. Therefore, theexploration of virtual, parametric urban space throughthe design of weighted fitness functions controlled bydesigners will prove vital. The fact that differentperformance metrics are competing is a driver for urbanform that explores unlimited possibilities onlyconceivable due to building performance simulation.Figure 9 Light, moderate and dense massing options asgenerated by the work flowThe utilization of this tool diminishes effort and timespent to model hundreds of street divisions that areadapted to complicated terrains. The focus shifts togaining insight into urban morphology and its effect onperformance through iterative explorations andoptimization procedures. The learning curve is steep,and the work flow outcomes are of great value to urbandesigners and planners. However, the tool needs furtherdevelopment to include effective capabilities such ascontrol of massing orientation.In an ever-growing world, and as more populationsmigrate to cities, the significance of this work flow,which supports the generation of sustainable urbanform, is indisputable. It currently subdivides terrainmodels based on minimum slopes, and parametricallycontrols the number of divisions, street widths, massingtypes and its properties. This initiates the means toevade haphazard and unaware urban forms, and pavesthe way to discovering possibilities of performance thatis optimal for the design of sustainable cities.REFERENCESAngel, S., Parent, J., Civco, D. L., Blei, A. M., 2010.The persistent decline of urban densities: Globaland historical evidence of sprawl. Lincoln InstituteWorking Paper. Cambridge, MA.Bing Maps., 2012. Retrieved March 11, 2012, from anonline interactive map: www.bing.com/maps/
Submitted to SimBuild 2012Rakha & ReinhartBeirão, J. N., Nourian, P., Mashhoodi, B., 2011.Parametric urban design: an interactive sketchingsystem for shaping neighborhoods. RespectingFragile Places, 29th eCAADe sity of Ljubljana, Faculty of Architecture,Slovenia, 21-24 September 2011, pp.225–234.Carr, L.J., Dunsiger, S.I., Marcus B.H., 2011.Validation of Walk Score for estimating access towalkable amenities. British Journal of SportsMedicine 45 pp.1144–1148.Frank, L.D., Greenwald, M.J., Winkelman, S.,Chapman, J., Kavage S., 2010. Carbonlessfootprints: Promoting health and eventive Medicine (supplement) 50 pp.99–105.Hutyra, L.R., Yoon, B., H-C, J., A, M., 2011. Carbonconsequences of land cover change and expansionof lands: A case study in the Seattle metropolianregion. Landscape and Urban Planning 103 (1)pp.83–93.Hoehner, C.M., Handy, S.L., Yan, Y., Blair, S.N.,Berrigan, D., 2011. Association betweenneighborhood walkability, cardiorespiratory fitnessand body-mass index. Social Science & Medicine73 (12) pp.1707–1716.pg.8Keirstead, J., Shah, N., 2011. Calculating minimumenergy urban layouts with mathematicalprogramming and Monte Carlo analysis techniques.Computers, Environment and Urban Systems 35pp.368–377.Luca, C., 2007. Generative platform for urban andregional design. Automation in Construction 16pp.70–77.Nicohosoff, A., 1933. Map of Cairo. Arts Graphiques,Alexandria, Egypt.Oliveira Panão, M.J.N., Gonçalves H.J.P., FerrãoP.M.C., 2008. Optimization of the urban buildingefficiency potential for mid-latitude climates usinga genetic algorithm approach. Renewable Energy33 pp.887–896.Seto, K.C., Fragkias, M., Güneralp, B., Reilly, M.K.,2011. A meta-analysis of global urban landexpansion. PLoS ONE 6 (8), e23777.Walk Score, 2011. Walk score methodology whitepaper. Retrieved March 11, 2012, from:www.walkscore.comUnited Nations, 2011. World Urbanization Prospects:The 2010 Revision. United Nations, New York.
expansion for the period of 1970 to 2000 was 58,000 km 2, which is approximately in the order of 2% of the global urban land area in 2000 (Seto et al., 2011). The United Nations’ latest figures demonstrate that by the year 2100, the world population is projected to re
Urban Design is only is 85; there is no application fee. Further information and application form see the UDG website www.udg.org.uk or phone 020 7250 0892 Urban Degsi n groUp Urban U Degsi n groUp UrBan DesiGn145 Winter 2018 Urban Design Group Journal ISSN 1750 712X nortH aMeriCa URBAN DESIGN GROUP URBAN DESIGN
Combining information theoretic kernels with generative embeddings . images, sequences) use generative models in a standard Bayesian framework. To exploit the state-of-the-art performance of discriminative learning, while also taking advantage of generative models of the data, generative
1 Generative vs Discriminative Generally, there are two wide classes of Machine Learning models: Generative Models and Discriminative Models. Discriminative models aim to come up with a \good separator". Generative Models aim to estimate densities to the training data. Generative Models ass
Generative Design in Revit -Save Default Settings Generative Design in Revit -Drop-down Inputs Generative Design in Revit -Constant and Variable Inputs Generative Design in Revit -Manage Study Type Folders Dynamo for Revit 2.10 Multiple Routes for Path of Travel Spatial Grids for Documenting Layouts Autodesk Revit 2022
parametric modeling with emphasis on two different groups of parametric software and presents the possibilities of generative algorithms in modeling architectural form and development of cities and urban design. Keywords—geometry, parametric design, generative algorithms urbanism, architectural design. I. INTRODUCTION
veloped for both generative and discriminative models. A straightforward, generative semi-supervised method is the expectation maximization (EM) algorithm. The EM ap-proach for naive Bayes text classification models is discussed by Nigam et al. in [17]. Generative semi-supervised meth-od
probabilistic generative models, which includes autoencoders[10] and powerful variants[13, 1, 14]. The second class, which is the focus of this paper, is called Generative Adversarial Networks (GANs)[5]. These networks combine a generative n
ple generative models based on different feature detectors and different numbers of parts) into a single classifier. Section 8 discusses the main results and observa-tions of this work. 2 Generative Models In this section we briefly review a class of generative models which will be used in conjunction with
