---
title: Planning Site Formation
url: https://www.emergentmind.com/topics/planning-site-formation
type: topic
---

# Planning Site Formation

Planning site formation, as realized in recent computational planning literature, denotes the formation of spatial entities that planners act upon: land-use configuration tensors for unplanned urban areas, contiguous demand-oriented regions for adaptation planning, excavation workspaces and dig sequences, screened or ranked infrastructure sites, and machine-readable site boundaries reconstructed from planning documents [2008.09912][2511.10857][2308.11478][2104.05792][2605.14714][2606.02747]. Across these settings, the planning object is not merely a label or forecast but an explicit spatial construct with geometry, adjacency, feasibility, and operational implications. The resulting methods span adversarial generative models, Variational Graph Auto-Encoders, representative initialized spatially constrained self-organizing maps, linear programming, SCUC/SCED-based screening, coverage planning, and multimodal geospatial reconstruction.

## 1. Problem classes and formal planning objects

The literature instantiates planning site formation in several technically distinct ways. In automated urban planning, Wang et al. define a land-use configuration as a longitude-latitude-channel tensor and cast planning as generating that tensor for an unplanned area from surrounding contexts [2008.09912][2112.14699]. In disaster adaptation planning, Noorani et al. define a “demand-oriented planning site” as a spatial unit that is a contiguous collection of grid cells, generated dynamically from high-resolution socioeconomic and environmental data rather than from pre-existing administrative boundaries [2511.10857]. In autonomous excavation planning, the planning object is a sequence of excavator base poses whose local workspaces cover the entire dig region while preserving machine mobility [2308.11478]. In renewable generation site selection and planner-initiated data-center siting, the planning object is a screened or ranked subset of candidate sites subject to system-level feasibility criteria [2104.05792][2605.14714]. In document-grounded planning-record reconstruction, the object is a valid GeoJSON boundary recovered from indirect spatial evidence [2606.02747].

| Planning setting | Planning object | Formal representation |
|---|---|---|
| Automated urban planning | Land-use configuration | \(M \in \mathbb{R}^{n \times n \times m}\) or \(M \in \mathbb{R}^{n \times n \times C}\) |
| Adaptation regionalization | Demand-oriented planning site | \(S=\{S_1,\dots,S_K\}\) partition of \(N\) cells |
| Autonomous excavation | Coverage-valid work sequence | \(B_1,\dots,B_N\) over a 2.5D elevation map |
| Renewable screening | Relevant RES sites | \(\mathcal N^n_{\mathrm{SITE}}\) |
| Flexibility-aware siting | Pre-qualified bus-envelope pairs | \(\mathcal Q_f^{(1)}\) |
| Planning-record reconstruction | Site boundary | GeoJSON Feature, Polygon or MultiPolygon |

This variety shows that “site” is domain-specific. In one class of problems, the site is the area to be configured; in another, it is the unit to be delineated; in another, it is the location to be selected; and in another, it is the boundary to be reconstructed. A plausible implication is that the unifying technical issue is the translation of heterogeneous evidence into spatially explicit planning units that satisfy domain constraints.

## 2. Representational primitives and input modalities

A consistent theme is the use of structured spatial encodings. In LUCGAN, a central site \(R\) is a \(1\,\mathrm{km}\times 1\,\mathrm{km}\) square divided into an \(n\times n\) grid, with \(M_{i,j,k}\) equal to the number of POIs of type \(k\) whose \((\mathrm{lon},\mathrm{lat})\) falls into cell \((i,j)\); one reported implementation uses \(n=16\), while the later systematic study uses \(n=50\) by default and experiments with \(n=5,10,25,50,100\) [2008.09912][2112.14699]. The surrounding context is represented by eight adjacent context squares together with feature blocks for monthly housing-price change, POI-category ratios, bus metrics, and taxi metrics, assembled into an attributed spatial graph for VGAE-based embedding [2008.09912][2112.14699].

The hierarchical urban-planning formulation in IHPlanner replaces a direct grid-only view with a three-level representation: a Spatial Attributed Graph \(SAG=(V,E,F)\) for surrounding regions, latent zone labels \(Z\in\{1\dots M\}^{N\times N}\), and fine-grained grid assignments \(G\in\mathbb{R}^{N\times N\times C}\) [2212.00904]. Human instructions are embedded as \(H\in\mathbb{R}^O\) and concatenated with the graph-derived context embedding, so regulatory targets such as greening rate, FAR, and setback enter the generative process as explicit conditioning variables.

In RepSC-SOM, the study area is discretized into \(N\) grid cells, each cell \(j\) carrying a feature vector \(x_j\in\mathbb{R}^d\) such as flood-risk, income, or population density, and a planning site is a partition whose components are spatially contiguous and internally homogeneous in those features [2511.10857]. In excavation planning, the site is first mapped from LIDAR point clouds into a georeferenced point cloud in the ECEF frame and then rasterized into a 2.5D elevation map at \(0.1\,\mathrm{m}\) resolution, with elevation and occupancy/traversability layers; final excavation geometry is encoded through user-defined cut and dump polygons with target elevation \(z_{\rm target}(x,y)\) [2308.11478].

The energy-siting papers adopt network-level abstractions. Radu et al. represent candidate renewable sites by capacity variables \(K_{nm}\), dispatch \(p_{nmt}\), and technical potential bounds \(\bar\kappa_{nm}\) in a two-stage CEP formulation [2104.05792]. The flexibility-aware data-center framework defines standardized load envelopes \(d^{(f)}_\tau\) over 24 hours for \(f\in\{\mathrm{firm},\mathrm{pause},\mathrm{shift}\}\), together with contingency-feasibility indicators \(\delta_{i,f,c,t}\) and pass rates \(\mathrm{PR}_{i,f}\) [2605.14714]. Plan2Map starts from raw planning PDFs and constructs a structured record containing site address, postcodes, British-Grid references, road names, place names, map scale, and page-level metadata before registration and segmentation [2606.02747].

## 3. Generative formation of urban land-use sites

The adversarial urban-planning line begins from the claim that land-use configuration can be treated as a deep generative learning problem conditioned on surrounding context [2008.09912][2112.14699]. In the 2020 LUCGAN formulation, the surrounding area is encoded as a 9-node graph \(\{C_1,\dots,C_8,R\}\) with adjacency \(A\in\{0,1\}^{9\times 9}\), node-attribute matrix \(X\in\mathbb{R}^{9\times K}\), and a VGAE objective
\[
L_{\mathrm{VGAE}}= KL[q(z|X,A)\|p(z)] + \lambda \cdot \|A-\hat A\|_F^2,
\]
with two GCN encoder layers, a decoder \(\hat A=\mathrm{sigmoid}(zz^\top)\), global average pooling over node embeddings, and reported settings \(d=64\) and \(\lambda=1\) [2008.09912]. The generator maps the context embedding \(z\) through a fully connected layer, reshape, and a cascade of ConvTranspose2d layers with BatchNorm and ReLU to an \(n\times n\times m\) nonnegative tensor \(\tilde M\), while the discriminator applies Conv2d layers with stride \(>1\), LeakyReLU activations, and sigmoid output [2008.09912]. The discriminator is trained against three classes of tensors: excellent plans, terrible plans, and generated plans. The quality score used for excellent-versus-terrible labeling is
\[
Q=\frac{2\cdot \mathrm{freq}\cdot \mathrm{div}}{\mathrm{freq}+\mathrm{div}},
\]
where \(\mathrm{freq}\) is total check-ins in \(R\) and \(\mathrm{div}\) is POI-type Shannon-diversity [2008.09912].

The later systematic treatment by Wang et al. keeps the tensorial formulation but introduces LUCGAN\(^+\), which performs conditioning augmentation on the context embedding,
\[
\mu=\mathrm{ReLU}(W_\mu z+b_\mu),\quad
\delta=\mathrm{ReLU}(W_\delta z+b_\delta),\quad
c=\mu+\delta\odot \epsilon^s,\quad \epsilon^s\sim\mathcal N(0,I),
\]
and concatenates this with additional noise to form the generator input \(\eta\in\mathbb{R}^{2H}\) [2112.14699]. On Beijing data comprising 2,990 residential communities, 328,668 POIs in \(C=20\) classes, taxi GPS traces, bus smart-card transactions, housing prices, and Weibo check-in logs, the paper reports that at \(n=50\), LUCGAN\(^+\) outperforms the best baseline by KL \(\downarrow 16.2\%\), JS \(\downarrow 0.25\%\), Hellinger \(\downarrow 28.4\%\), and Wasserstein \(\downarrow 48.6\%\), and improves over original LUCGAN by \(8.92\%\), \(0.23\%\), \(8.43\%\), and \(4.32\%\), respectively [2112.14699].

IHPlanner extends this line by making site formation hierarchical and human-instructed rather than directly tensor-conditional [2212.00904]. The model first generates latent zone labels \(Z\), then computes functionality projections
\[
\alpha=\mathrm{Softmax}(\mathrm{AVG}(F)\cdot W_a),\qquad
T_i=\alpha_i\cdot z,
\]
and finally models peer dependencies among the \(M\) projections using multi-head self-attention,
\[
A_h=\mathrm{Softmax}\!\left(\frac{Q_hK_h^\top}{\sqrt{d_k}}\right)V_h.
\]
A planning layer reshapes the resulting representation into \(G\in\mathbb{R}^{N\times N\times C}\), and training minimizes a total objective combining stage-1 GAN loss, KL regularization, and stage-3 reconstruction loss [2212.00904]. On the Beijing case with five green-rate levels \(\{\mathrm{Green0}\dots \mathrm{Green4}\}\), IHPlanner attains the lowest AVG\_KL, AVG\_JS, AVG\_HD, and AVG\_Cos across all levels and remains robust over \(N\in\{5,10,25,50,100\}\) [2212.00904].

These urban models share three commitments: explicit quantification of the planning object, learned context encoders that preserve spatial dependence, and evaluation against both distributional criteria and planner-defined quality proxies. At the same time, the papers explicitly state limits: the framework is “purely statistical,” does not explicitly enforce domain-specific zoning rules such as setback or FAR, and still recommends final approval by a human planner [2112.14699][2008.09912].

## 4. Spatially constrained regionalization and infrastructure siting

In adaptation planning, site formation becomes the generation of dynamic planning units rather than land-use tensors. Noorani et al. define a demand-oriented planning site as a partition \(S=\{S_1,\dots,S_K\}\) of \(N\) grid cells such that each \(S_k\) is spatially contiguous and internally homogeneous in features relevant to the planner’s disaster-adaptation objectives [2511.10857]. RepSC-SOM extends classical SOM with representative seed initialization, adaptive geographic filtering of Best Matching Unit assignments, and post-SOM region-growing refinement. The SOM update follows Kohonen:
\[
w_i(t+1)=w_i(t)+\eta(t)\cdot h_{ci}(t)\cdot[x_j-w_i(t)],
\]
and geographic filtering restricts BMU search to nodes whose geographic distance is within threshold \(R\), where \(R\) is chosen from the semivariogram “range” of the selected features [2511.10857]. After convergence, regions with the same BMU label are merged and refined; the paper gives an illustrative similarity score
\[
S(a,b)=1-\frac{\|\mu_a-\mu_b\|}{\max_{\alpha,\beta}\|\mu_\alpha-\mu_\beta\|}.
\]
Evaluation includes within-region variance, a silhouette coefficient adapted to contiguous clusters, a fragmentation index, and comparison against baseline units such as census tracts [2511.10857].

Radu et al. treat site formation as model reduction through renewable generation site selection in capacity expansion planning [2104.05792]. Their two-stage method begins with SITE, a screening LP that keeps only those RES sites whose omission would significantly change system cost or design. The objective is
\[
\min_{K,p,p^e}
\;\omega\Bigl[\sum_{n}\sum_{m}(\zeta^m+\theta_f^m)K_{nm}\Bigr]
+\sum_{t}\Bigl[\sum_{n,m}\theta_v^m p_{nmt}+\sum_n\theta^e p^e_{nt}\Bigr],
\]
subject to feed-in target, dispatch-limit, and technical-potential constraints; any site with \(K_{nm}\ge 1\,\mathrm{MW}\) in the SITE solution is deemed relevant and retained for the reduced CEP problem [2104.05792]. On a 33-country ENTSO-E case with 1,740 candidate sites, SITE consistently retrieves over \(90\%\) of the optimal RES sites, yields objective difference \(\le 0.52\%\) between reduced and full LP, and reduces peak memory by \(39\)–\(41\%\) and total solver time by \(31\)–\(47\%\) [2104.05792].

The 2026 flexibility-aware framework moves from screening candidate generators to planner-initiated siting of large flexible loads [2605.14714]. Stage 1 defines the N–1 pass rate
\[
\mathrm{PR}_{i,f}
=
\frac{1}{|\mathcal T_{\rm yr}|\,|\mathcal C|}
\sum_{t\in\mathcal T_{\rm yr}}
\sum_{c\in\mathcal C}
\delta_{i,f,c,t},
\]
with default threshold \(\tau_{\rm PR}=0.95\). Stage 2 evaluates each qualified \((i,f)\) pair by year-hourly SCUC/SCED. Stage 3 ranks alternatives by entropy-weighted multi-criteria scoring and TOPSIS closeness coefficient \(\mathrm{CC}_r\in[0,1]\) [2605.14714]. The three planner-issued envelopes are firm, pause, and shift, defined by Eq. (14) with a within-day ramp-rate limit \(\Gamma=0.2P\). On the synthetic 2,000-bus Texas system, operational flexibility expands the pre-qualified set from 193 firm buses to 226 pause buses and 210 shift buses at \(1\,\mathrm{GW}\), and from 57 firm buses to 68 pause buses and 69 shift buses at \(2\,\mathrm{GW}\), while median all-hour mean LMP remains \(24.32\,\$/\mathrm{MWh}\) for the \(2\,\mathrm{GW}\) cases [2605.14714].

Taken together, these methods replace inherited planning units or exhaustive candidate sets with generated, screened, or ranked sites whose admissibility is explicitly tied to homogeneity, reliability, or system-cost fidelity.

## 5. Geometric site formation, excavation, and boundary reconstruction

In autonomous excavation, planning site formation has a literal geometric meaning: the system must form a valid sequence of workspaces and bucket motions that transform present topography into the target geometry [2308.11478]. The site is first mapped by LIDAR and ICP-based SLAM with GPS-RTK loop closures into a georeferenced point cloud and then rasterized into a 2.5D elevation map with elevation and occupancy/traversability layers. The user specifies target cut and dump polygons in Google Earth Pro with target elevations \(z_{\rm target}\), and the planner identifies where \(f_{\rm elev}(x,y)>z_{\rm target}(x,y)\) [2308.11478]. The global planner performs Boustrophedon decomposition along orientation \(\theta\), constructs a quotient graph, solves a Minimum-Branching-Vertices Spanning Tree Problem, and uses dynamic programming with recurrence
\[
D_{i,c_k}=\min_{c_n,c_j,\ell}\Bigl[D_{i-1,c_n}+d_{i,\ell}(c_n,c_j)+d_o(c_j,c_k)\Bigr]
\]
to choose coverage subroutines and entry/exit corners. Local excavation planning partitions the workspace into five radial-angular zones and selects dump zones by minimizing
\[
C_D(Z)=\frac{1}{|Z|}\sum_{x_i\in Z}\mathrm{SDF}_{\rm dump}(x_i)
+\alpha\|x_{\rm dig}-x_{\rm dump}\|^2,\qquad \alpha=4.0.
\]
The digging planner then optimizes the attack point \((r,\theta)\) under reach constraints using Bayesian optimization with a Gaussian-process surrogate and Expected Improvement, requiring about 30 evaluations rather than 150 naive grid samples [2308.11478]. The reported system excavates a \(15.6\,\mathrm{m}\times 11.5\,\mathrm{m}\times 1\,\mathrm{m}\) pit in \(4\,\mathrm{h}\,25\,\mathrm{min}\), achieves \(T_{\rm cycle}=32.08\pm 5.47\,\mathrm{s}\), and improves final grade absolute error from about \(9.7\,\mathrm{cm}\) before refinement to about \(7.2\,\mathrm{cm}\) after refinement [2308.11478].

Plan2Map addresses a different geometric problem: reconstructing valid geospatial site boundaries from planning documents that contain only indirect spatial evidence [2606.02747]. The benchmark contains 208 UK planning-record cases, and GeoPlanAgent decomposes the task into evidence extraction, localisation, map registration, boundary segmentation, projection, and verification. The Reader produces a structured record from raw planning PDFs; the Locate sub-agent uses signals such as site address, postcodes, grid references, house-number road pairs, and visible map labels to infer candidate coordinates; and the Worker performs auto-rotation with a 4-class ResNet50 classifier, sliding-window tile matching with MINIMA-LoFTR plus RANSAC, boundary segmentation with SAM 3 plus LoRA adapters, affine projection to WGS84, vectorisation, and GeoJSON assembly [2606.02747]. The segmentation model uses
\[
L=L_{\rm semantic}+L_{\rm instance},
\]
with focal, Dice, surface, classification, and presence terms, LoRA rank 16, \(\alpha=32\), dropout \(0.05\), and AdamW with learning rate \(2\times 10^{-4}\) [2606.02747]. On the full benchmark, GeoPlanAgent + Critic reports mean IoU \(0.740\), median IoU \(0.906\), IoU \(\ge 0.8\) on \(67.8\%\) of cases, median error \(4.6\,\mathrm{m}\), Acc@0.1D \(78.8\%\), cost \(\$0.045\) per document, and time \(155\,\mathrm{s}\) per document [2606.02747].

These two systems show complementary meanings of geometric site formation. Excavation planning generates executable geometry-changing actions from terrain state and machine constraints. Boundary reconstruction generates machine-readable spatial extents from documentary evidence and map imagery. In both cases, the output must be geometrically valid and operationally usable rather than merely semantically plausible.

## 6. Evaluation regimes, human oversight, and recurring limitations

Evaluation in planning site formation is strongly task-specific, but the reported metrics cluster around four concerns: plausibility, internal coherence, system feasibility, and geometric accuracy. For urban land-use generation, Wang et al. use KL divergence, JS divergence, Hellinger distance, Wasserstein distance, and a separate quality-scoring model trained on real configurations [2112.14699]. The earlier LUCGAN report also reserves \(20\%\) of hand-labeled excellent/terrible maps to train a random-forest scorer, obtaining average generated-map scores of about \(0.82\) for LUCGAN, \(0.71\) for VAE-only, \(0.62\) for AVG, and \(0.90\) for MAX, while noting that MAX is context-agnostic [2008.09912]. IHPlanner uses AVG\_KL, AVG\_JS, AVG\_HD, and AVG\_Cos across green-rate levels [2212.00904]. RepSC-SOM evaluates within-region variance, an adapted silhouette coefficient, fragmentation, and hazard-aligned homogeneity versus baseline units [2511.10857]. Renewable site screening uses spatial reduction \(\gamma_r\), screening accuracy \(\alpha_r\), TSCE, capacity differences, problem-size reduction, peak memory, and solver time [2104.05792]. Flexibility-aware siting emphasizes pass rate, mean LMP, P95–P5 price dispersion, binding hours, congestion rent, and TOPSIS closeness coefficients [2605.14714]. Excavation planning reports coverage fraction, planning-time success, cycle time, soil-movement rate, and grade error [2308.11478]. Plan2Map uses IoU, centroid-distance error, Acc@0.1D, runtime, and cost [2606.02747].

Human oversight remains explicit across the literature. LUCGAN is described as an assistant rather than a full replacement, and final approval by a human planner is still recommended [2008.09912]. Noorani et al. place the planner inside the loop for feature selection, spatial constraint override, desired region count, and interactive exploration [2511.10857]. The flexibility-aware siting framework is planner-initiated and produces ranked, pre-certified catalogues rather than automatic commitments [2605.14714]. GeoPlanAgent includes an optional Critic that can approve, switch candidates, or request relocalisation [2606.02747].

The limitations are similarly recurrent. The urban generative models are Beijing-centric and do not explicitly encode zoning rules such as setback and FAR, although a later LUCGAN implementation notes that fine-tuning on as few as 200 labeled excellent/terrible samples may support transfer to a new region when the POI taxonomy is comparable [2112.14699][2008.09912]. RepSC-SOM requires good-quality fine-grained data, imposes computational overhead for very large \(N\), and may require legal or political buy-in if dynamic units replace official zones [2511.10857]. SITE depends on manual choices of \(\delta\tau\) and \(\xi^n_\tau\), and its screening LP only approximates storage and transmission effects [2104.05792]. The data-center framework assumes standardized flexibility envelopes and DC-OPF-based reliability screening [2605.14714]. Plan2Map shows that direct VLM-to-GeoJSON prediction remains unreliable, with remaining errors concentrated in localisation and map registration [2606.02747]. Excavation planning, although fully autonomous in the reported system, still depends on accurate mapping, target geometry specification, and navigation/execution integration [2308.11478].

A plausible implication of these convergent findings is that planning site formation is becoming a multi-stage discipline in which representation design, spatial constraint handling, and post hoc verification are at least as important as the core learning or optimization module. The cited work does not present a single universal framework, but it does define a coherent computational agenda: explicit spatial objects, heterogeneous context ingestion, domain-specific feasibility filters, and measurable outputs that can be inspected, revised, and deployed.

Source: https://www.emergentmind.com/topics/planning-site-formation