Designing for Usability in Modular Space Habitats: A Space Syntax Perspective

As mentioned above, NASA’s descriptions of habitat subsystems are focused primarily (and understandably) on the functional [3][5][4]. The consensus seems to be that early settlements will most likely consist of modular habitats linked together, via pressurized passages or tunnels, to form larger aggregations [13]. The modules being linked can be sorted into three sub-categories; hard-shell modules; inflatable or expandable habitats (essentially pressurized structures with their internal volume increasing after deployment); and lastly, structures made of in-situ resources (ISRU). These are termed classes-I, II and III respectively [19]. Prototype examples of class-I include the Habot module (habitat + robot), a pressured rigid vessel (with the ability to move) and a basic module diameter of 3.75m [6]. The Habot comes in multiple different types, with varying pressure port angles, which is of significance when considering layout possibilities.

Prototype examples of class-II include the “planetary surface habitat and airlock unit”, a 3.65m diameter inflatable fabric structure [13] and the MB10 habitat [28], a three-storey inflatable structure which should reach a diameter of 26m when fully deployed. In another paper [28], Sinn presents a hypothesized aggregation of 19 such units in a densely packed honeycomb arrangement. Finally, the NASA TransHab project describes a hybrid type module combining a rigid core and inflatable exterior. It provides a four-storey, 3.35m wide inflatable habitat [18]. For spherical modules, Vogler [29] suggests that 4.0m diameter should be the minimum size. It appears that most deployable units are initially conceived of as being between 3m-4m in diameter. Class-III examples can be built using a number of construction methods applied to the ISRU, including 3D printing and brick-making [20][31][30].

All of the above examples permit a reconfigurable aggregation of modules. We have based our simulations on the Habot module, as it seems the best documented in terms of the range of different module-types and as it offers a variety of connector-layouts. Our next challenge was to consider how large an extraterrestrial city might reasonably be. Outside of the rich world of science fiction, we have struggled to find any considered discussion of how expansive an aggregation of modules might ultimately become. Pabila [26] suggests that they may expand to fill entire craters on Mars: “The natural shape of the impact crater creates predetermined sized cities, or states that would spread to fill the crater.” Beyond the latter, some clue might be found in a presentation by Elon Musk in 2017 when he presented an artists’ visualization of what a Martian city might look like [25] (Figure 2).

Setting aside the physical dimensions of the modules in Musk’s presentation for the time being, a simple count of the number of modules at each stage of growth provides us with some insights. The settlement’s initial phase expands from “one ship” to 40 modules. Expansion stage two doubles the size to 80 modules, whilst the latter “city-state” is depicted as reaching 200 modules. Notably, not all modules are shown as being physically connected, although it is reasonable to assume the presence of subterranean tunnels, or established external pathways between proximal airlocks. Given the lack of any other data, we have decided to simulate habitat aggregations for up to 200 modules. Such an approach captures the early-stage architectural level challenges of settlement growth and configuration, and as such helps to identify key architectural constraints and risks that any such habitat is likely to face in its more vulnerable formative years.

To better understand the growth of such extraterrestrial settlements, we can turn to architectural and urban theories such as Space Syntax. Space Syntax is a framework of architectural theories, spatial representations, analytic methods and software tools, primarily developed at University College London from the 1970s onward by Bill Hillier and colleagues. It seeks to understand the relationships between spatial configurations found in the built environment and people’s use of, movement through and social interactions within them [15]. Space Syntax tools and methods are widely applied across different scales from simple to complex buildings, neighborhoods and small settlements, through to large cities and geographic regions.

Fundamentally, Space Syntax methods decompose any built environment into different types of spatial representations or “primitives,” typically either convex spaces (spaces containing only mutually visible location pairs [15]), axial lines (longest lines of sight/movement, [27]), or isovists (visibility fields [1][23][24]). These spatial units are then typically represented as graphs, with spaces becoming nodes and edges representing spatial relations such as adjacency, accessibility, and/or visibility. This approach allows the extraction of graph-based metrics, many of which have been shown to correlate with observed use or behavior, such as occupancy rates, pedestrian flow [16], getting lost [8], and architectural function. Whilst a comprehensive review of Space Syntax is beyond the scope of our paper, we use a selection of Space Syntax concepts: Hillier’s space types; isovist field analyses and intelligibility; and “beady-ring” settlements. Below we briefly explain each in turn, and discuss how they are relevant to the modeling and analysis of aggregations of space habitats.

We began by redrawing 25 module configurations from Cohen [6], using his layouts as a basis for a broader analysis of modular systems (Figure 3). These include a variety of connection types and pressurized port arrangements. We analyzed each layout for the distribution of Hillier’s space types ($a$-, $b$-, $c$-, $d$-), our new critical space types ($b^{\ast }$, $c^{\ast }$, $d^{\ast }$, $e$, $e^{\ast }$), and the number of circulation rings. We also used the Isovists app [24][23][22] to calculate the integration and intelligibility values, establishing a baseline for spatial usability. A new vulnerability measure was also calculated.

To test our theory, we simulated four large settlements of 75, 100, 150, and 200 modules using an adapted version of Hillier and Hanson’s “beady ring” algorithm, based on habitat-connector compatibility rules. In each case, modules were placed, by the software, in a semi-random order but constrained to connect only to available ports on previously placed habitats, reflecting the opportunistic but rule-governed nature of accretive growth. These were exported as vector-based plans (.dxf) and analyzed in the Isovists app [24][23][22] to calculate integration and intelligibility, as well as the distribution of Hillier’s original plus new critical space types ($a$-, $b$-, $c$-, $d$-, $e$, $b^{\ast }$, $c^{\ast }$, $d^{\ast }$ and $e^{\ast }$). We then generated an additional 25 large settlements, each containing 100 habitats, in GraphML format, unsuitable for isovist-based field analysis, but ideal for calculating the distribution of space types and vulnerability scores (Section 3.3).

The vulnerability measure analysis was applied to all 29 city-scale configurations (as well as to the original 25 small settlements). Next, we developed a new metric for habitat vulnerability to quantify and compare the inherent resilience of settlement configurations. We define vulnerability at two levels: local vulnerability, which assesses the impact of the loss of access to each individual module (due to pressure loss, maintenance, contamination, repair, and so forth); and global (or systemic) vulnerability, an evaluation of the overall resilience of the whole system. Each metric can be considered in absolute terms, although we suggest normalizing global vulnerability by the total number of modules in the system.

To calculate the true vulnerability of a settlement configuration, we begin by identifying all habitats whose removal would split the settlement into two or more disconnected parts. These include every $b$-space, $b^{\ast }$-space, $c^{\ast }$-space, $d^{\ast }$-space, and any extended categories such as $e$- and $e^{\ast }$-spaces. $a$-spaces (single-connection dead ends) are excluded, as their isolation is trivial and expected; the same applies to ordinary $c$- and $d$-spaces which are considered only when they act as structural connectors (their starred variants).

We then subject every habitat – critical or otherwise – to a uniform random-failure test. For each habitat, $(h)$, containing $(N)$ modules we temporarily remove it from the settlement and record the size of the largest surviving, disconnected section and then normalize by the remaining settlement size $(N-1)$ or $(L)$ [11]. The space-type categories decide whether $(L(h)=1)$ (for non-splitting $a$-, $c$-, $d$-spaces) or (for $b$-, $b^{\ast }$-, $c^{\ast }$-, $d^{\ast }$-, $e$-, $e^{\ast }$-spaces), so their structural role is fully embedded in the calculation. Averaging those scores over all $(N)$ habitats yields the expected post-failure connectivity. In the two analytically tractable extremes – a perfect ring and an infinitely long line – the expected vulnerability is 0 and 0.25, respectively. Multiplying this by 4 therefore stretches the range to [0,1] and anchors the scale at “perfectly robust” (ring) and “maximally fragile” (path). The final vulnerability score is:

High global vulnerability values indicate that significant portions of a habitat-system are vulnerable to fragmentation and configurational collapse if key nodes are removed or blocked. In this way, the measure directly reflects a settlement’s configurational fragility. Conversely, low values would indicate a degree of configurational resilience. Similarly, high local vulnerability values identify critical modules within the whole system, and vice versa.

Figure 3 shows the results of analyzing Cohen’s 25 Habot configurations [6] and the location of $a$-, $b$-, $c$-, $d$-, $e$-, and the “star hub” space-type modules. Navigable space has been colored according to integration – a necessary computational stage towards calculating intelligibility. High-integration spaces are shown in red, with the spectrum shifting through green to blue for lower-integration spaces. We also present the resultant intelligibility values for each configuration (the higher the value, the less confusing and less maze-like the layout).

Figure 4 shows the number and proportion of $a$-, $b$-, $c$-, and $d$-space types in the original Habot configurations (Section 2.2.1) It is clear that a high proportion of some space types contributes to an inherent vulnerability of the layout. We also calculated the number of rings (also known as cycles in graph theory), and counted the number of independent rings, nested rings, and total rings in each layout. A ring in a layout contributes to the redundancy of that layout: if a module forming part of a ring becomes unusable (due to damage, contamination, repair, blow-out, depressurization, etc.), the rest of the modules can still be reached without passing through the non-functioning module. This continuity of space is important for the circulation of people and life support.

For independent rings, 56% of Cohen’s layouts contained at least one ring, with the number of rings ranging from 1 to 6. Only 2 layouts contained nested rings, with the highest number being 15 total nested rings. Finally, we calculated the resultant vulnerability score for each small configuration, as shown in Table 2.

Examples of the results of the “beady-ring” habitat growth simulation, and distribution of space types, can be seen in Figure 5. The distribution of the space types in the simulated cities differs from those in the smaller configurations (Figure 4). For the larger settlements, $b$-spaces, $b^{\ast }$-spaces, $c^{\ast }$-spaces, and $d^{\ast }$-spaces were the least commonly occurring space types, while $a$-spaces and $c$-spaces were significantly more prevalent, reflecting the accretive nature of large-scale growth. To test this further, we generated an additional 25 simulated city layouts (in GraphML format only) each consisting of N=100 habitats. We determined the number of different space types for each layout. For these additional, simulated settlements, the most frequently occurring space types were $c$-spaces (spaces on a single ring) and $a$-spaces (the ends of linear sequences). These formed, on average, 22% and 21% of all habitats, respectively. In the middle of the distribution were the $d$-spaces (spaces on multiple rings) and $b$-spaces (two-connector spaces as part of tree-like subsystems), each of which constituted 16% of the spaces. The least frequently occurring spaces were the star-hub types – $c^{\ast }$-spaces, $b^{\ast }$-spaces, $d^{\ast }$-spaces, and $e^{\ast }$-spaces – the configurationally critical space in the event of failure. They represented 12%, 8%, and 5% of all large-scale habitats, respectively. The new $e$-spaces and $e^{\ast }$-spaces typically constituted $1\%$ of habitats.

The intelligibility scores of the settlements (a measure of how confusing or maze-like a settlement is) are markedly lower than those of the 25 Cohen small layouts. For $N=75$, intelligibility is $I=0.336$; for $N=100$, $I=0.686$; for $N=150$, $I=0.302$; and for $N=200$, $I=0.467$. These four simulated “beady-ring” layouts have an average intelligibility of $I=0.448$. Compare this value to the 25 Cohen layouts, which have a mean intelligibility of $I=0.797$, which can be interpreted as “highly intelligible” and also notably higher than the intelligibility of the larger settlements (Figure 5). (As a general heuristic, may be interpreted as indicating a maze-like and confusing configuration.)

The vulnerability scores of the four simulated cities are: for N=75 vulnerability is 0.180; for N=100 vulnerability is 0.082; for N=150 vulnerability is 0.106; for N=200 vulnerability is 0.075. On average, the 25 Cohen layouts have a mean vulnerability score of 0.291 which is interpreted as being moderate (excluding the perfect rings, which by definition have a V of 0.000, the mean vulnerability is 0.419 (more vulnerable) and a standard deviation of 0.291. For the four large, simulated beady-ring layouts (Figure 5), the mean vulnerability score is 0.111. The average vulnerability score across all 25 additional GraphML habitats (N = 100) is 0.342 and also less vulnerable than the 25 Cohen layouts. If we plot all 54 vulnerability scores (25 Cohen layouts + 25 GraphML simulations + 4 large settlements (Figure 5)) against their settlement size (i.e. total number of habitats) simple linear fit gives a coefficient of determination: $R^2=0.294)$. Thus, only about $29\%$ of the variance in vulnerability is explained by the size of the settlement alone; the remaining $71\%$ depends on the mix of habitat/space types rather than on the total habitat count.

The adapted “beady-ring” algorithm proved successful in generating large-scale modular settlement configurations with realistic accretive characteristics. By applying modified habitat-connector compatibility rules, the simulation was able to replicate the organic growth dynamics typical of extraterrestrial modular habitats and allow meaningful spatial analysis. These simulations provided a valuable testbed for evaluating spatial properties – particularly integration, intelligibility and vulnerability – at a scale beyond anything previously modeled.

One important and potentially surprising outcome of our analysis is the stabilization of Hillier’s space-type distributions as settlement size increases. While the smaller 25 Cohen configurations showed high variability in the proportions of $a$-, $b$-, $c$-, and $d$-spaces – along with their more critical $b^{\ast }$-, $c^{\ast }$-, $d^{\ast }$-, $e$-, and $e^{\ast }$-types – the larger simulated settlements (comprising four fully analyzed layouts and 25 additional graph-based simulations) demonstrated remarkably consistent proportions. This suggests that as modular settlements grow, they begin to exhibit predictable and self-organizing spatial typologies. From a design perspective, this is significant: if spatial properties tend to stabilize through accretive growth, it may be possible to anticipate and guide future configurations more effectively.

As space-type distributions stabilize, so too do associated vulnerability measures. Because vulnerability is defined by the impact of losing structurally critical nodes (specifically $b^{\ast }$-, $c^{\ast }$-, $d^{\ast }$-, $e$-, and $e^{\ast }$-spaces), their normalized value (as a proportion of total modules) tends to decrease as the number of modules increases. This supports the hypothesis that larger settlements are inherently more resilient to localized failures – a trend confirmed across our dataset. However, this apparent configurational robustness comes with a trade-off, which is usability.

Intelligibility – defined as the correlation between local visibility (isovist connectivity/area) and global configurational depth (isovist integration) – tends to decline as modular habitats increase in scale. This trend, observed for the 25 Cohen configurations and the four simulated cities, highlights a key usability challenge: as spatial systems grow, they become harder to understand and navigate. This is consistent with Hillier’s theories of urban complexity, which suggest that unstructured spatial growth leads to increased fragmentation and cognitive disorientation. In this context, while large settlements may be robust, they are also more functionally opaque – potentially making them difficult or stressful for inhabitants to use.

Preliminary findings suggest a relationship between intelligibility and vulnerability, although our dataset is not yet large enough for definitive statistical analysis. Nevertheless, the correlation is conceptually persuasive: as intelligibility declines, spatial legibility is lost, but at the same time, the systemic redundancy that supports resilience increases (larger configurations are broadly less vulnerable but the correlation coefficient is $R^2=0.294)$. Therefore, if growth proceeds without spatial structuring, the risk of disorientation rises yet the risk of disconnection, and sub-system isolation, drops. This points to an important direction for future research – specifically, the possibility of defining spatial usability thresholds based on combined intelligibility–vulnerability scores.

Cohen [6] argued that a straight-run configuration of interconnected modules would be among the most efficient for early off-world habitats. While this may hold from a construction or logistical standpoint, our findings challenge this view from a spatial resilience perspective. In our analysis, straight-run configurations consistently ranked among the most fragile. Their linear configuration offers minimal redundancy and is highly susceptible to complete fragmentation if a single module is lost. From the standpoint of usability and resilience, such configurations appear among the least favorable for long-term habitation.