---
title: 'DisCo-Layout: 3D Indoor Layout Synthesis'
url: https://www.emergentmind.com/topics/disco-layout
type: topic
---

# DisCo-Layout: 3D Indoor Layout Synthesis

Searching arXiv for the specified papers and closely related layout-generation work.
First, retrieving the DisCo-Layout paper by arXiv id.
DisCo-Layout is a framework for 3D indoor layout synthesis that disentangles semantic refinement from physical refinement and coordinates them within a multi-agent system [2510.02178]. In its formulation, the input consists of a natural language prompt $T$, a set of retrieved 3D assets $O=\{o_1,\dots,o_n\}$, and a room with boundary walls $B_{\text{room}}=\{b_1,b_2,b_3,b_4\}$; the output is a set of poses $P=\{p_1,\dots,p_n\}$ with $p_i=(x_i,y_i,\theta_i)$ such that object placements satisfy both semantic coherence and physical plausibility [2510.02178]. The framework’s central claim is that semantic richness and robust flexible refinement should not be handled by a single coupled optimizer: instead, abstract object relationships are revised by a Semantic Refinement Tool (SRT), while coordinate-level feasibility is enforced by a Physical Refinement Tool (PRT), with a Planner–Designer–Evaluator loop orchestrating both [2510.02178]. A related but distinct usage of the same decoupling intuition appears in layout generation for 2D graphic scenes, where LDGM decouples diffusion across category, position, and size in a discrete diffusion model; that work does not reference DisCo-Layout explicitly, but it embodies the same high-level principle of separating attribute-specific corruption from coordinated generation [2303.05049].

## 1. Formal problem setting

DisCo-Layout addresses 3D indoor layout synthesis under open-ended prompts and heterogeneous asset sets [2510.02178]. The task is defined as producing object poses $P=\{p_1,\dots,p_n\}$ with $p_i=(x_i,y_i,\theta_i)$ given a prompt $T$, retrieved assets $O=\{o_1,\dots,o_n\}$, and room boundaries $B_{\text{room}}=\{b_1,b_2,b_3,b_4\}$ [2510.02178]. Two classes of requirements are imposed. Semantic coherence means that placements reflect functional and spatial relationships implied by the prompt, including patterns such as “chairs around table” and “TV faces sofa” [2510.02178]. Physical plausibility means that objects lie within $B_{\text{room}}$, avoid collisions, and satisfy explicit spatial constraints such as “against wall” [2510.02178].

The paper motivates this formulation by contrasting it with fixed-dataset approaches that rely on in-domain assets and fixed placement schemes, which impedes generalization to unseen assets and layouts as well as open-ended prompts and constraints [2510.02178]. It further distinguishes semantic richness from robust flexible refinement. Semantic richness refers to high-level relational logic such as adjacency, grouping, facing, or functional pairings; robust flexible refinement refers to iterative, targeted corrections for collisions, out-of-bounds placements, and wall alignment without degrading semantic structure [2510.02178]. The stated argument is that single-step LLM/VLM methods and coupled refiners either lack consistency or entangle objectives, causing interference [2510.02178].

A plausible implication is that the framework is designed less as a monolithic generator than as a decomposition of layout synthesis into semantically interpretable and physically verifiable subproblems.

## 2. System architecture and multi-agent coordination

The framework is organized around three agents and two refinement tools: a Planner, a Designer, an Evaluator, the Semantic Refinement Tool (SRT), and the Physical Refinement Tool (PRT) [2510.02178]. The Planner derives high-level placement rules and semantic asset groups with an ordered placement schedule, producing constraints $C$ and ordered groups $G$ via
$$(G,C)=f_{\text{Planner}}(T,O).$$
The Designer then proposes an initial layout for each group using VLM reasoning over the current scene image and the relevant constraints:
$$\tilde{P}_k=f_{\text{Designer}}\big(T,\{(o_i,c_i)\}_{o_i\in G_k},I_{k-1}\big),$$
where $I_{k-1}$ is the rendered top-down image of previously placed groups [2510.02178]. The Evaluator assesses semantic coherence and physical plausibility using targeted VQA-style questions:
$$(s_{\text{sem}},s_{\text{phy}})=f_{\text{Evaluator}}\big(P'_k,I'_k,\{c_i\}_{o_i\in G_k}\big),$$
with $P'_k=P_{k-1}\cup \tilde{P}_k$ [2510.02178].

The pipeline proceeds group by group. First, the Planner computes constraints per asset and groups them by placement priority. Second, the Designer places assets in group $G_k$ into the scene and renders $I'_k$. Third, the Evaluator performs structured VQA on semantic constraints and physical issues, producing booleans $(s_{\text{sem}},s_{\text{phy}})$. Fourth, if $s_{\text{sem}}=\text{True}$, SRT is applied; if $s_{\text{phy}}=\text{True}$, PRT is applied; the layout is then updated to $P_k$. Fifth, the process iterates until all groups have been placed and refined [2510.02178].

The paper also specifies the coordination loop in pseudocode. Starting from $(G,C)\leftarrow f_{\text{Planner}}(T,O)$, the system initializes $P_0=\varnothing$ and an empty room render $I_0$. For each group $k=1,\dots,K$, it computes $\tilde{P}_k$, forms $P'_k$, renders $I'_k$, evaluates $(s_{\text{sem}},s_{\text{phy}})$, conditionally applies SRT and PRT, renders the updated scene, and continues until the final layout $P_K$ is returned [2510.02178]. The stopping criterion is explicit: per group, refinement continues until both booleans are False, after which the system proceeds to the next group [2510.02178].

This staged design separates planning, initial proposal, diagnosis, semantic repair, and physical repair into discrete operators. The paper’s interpretation is that such modularity avoids the objective interference of coupled optimization [2510.02178].

## 3. Semantic refinement and constraint representation

Semantic refinement is handled by the Semantic Refinement Tool, which operates when the Evaluator detects semantic issues, i.e., when $s_{\text{sem}}=\text{True}$ [2510.02178]. Constraints are specified per object as
$$c_i=\{c_{\text{wall}},c_{\text{rel}},c_{\text{rot}}\},$$
where relationship types satisfy
$$c_{\text{rel}}\in\{\text{near},\text{side of},\text{in front of},\text{aligned with},\text{opposite},\text{around}\},$$
and the facing constraint $c_{\text{rot}}$ identifies the target object an asset should face [2510.02178].

When semantic failures are detected, the failed constraints $C_{\text{sem}}^*$ are converted to explicit textual feedback such as “The chair is not facing the table,” and the VLM proposes minimal position and rotation changes that satisfy the failed constraints while preserving the overall layout [2510.02178]. The tool is defined as
$$
P''_j=
\begin{cases}
P'_j, & \text{if } s_{\text{sem}}=\text{False},\\
f_{\text{SRT}}(P'_j,C_{\text{sem}}^*), & \text{if } s_{\text{sem}}=\text{True}.
\end{cases}
$$
Its interface takes the current layout $P'_j$, failed semantic constraints $C_{\text{sem}}^*$, the current scene image $I'_j$, and asset sizes, and produces JSON patches only for the objects that need changes [2510.02178].

The described SRT procedure is explicit. It extracts all failed constraints and formulates feedback text for each one; prompts the VLM with the current layout encoded as JSON with object name, size, position, and rotation; enforces a restricted output schema containing only changed objects; parses the resulting JSON; and minimally updates $(x_i,y_i,\theta_i)$ relative to prior values before returning the revised layout $P''_j$ [2510.02178].

In the planner’s JSON schema, each object may include `against_wall`, `relative_position`, `relative_object`, and `rotation` fields [2510.02178]. The Evaluator’s structured VQA schema likewise returns per-question JSON with keys `question`, `objects`, `reason`, and `answer` [2510.02178]. This suggests that semantic refinement is not merely an implicit latent correction mechanism, but an explicitly inspectable constraint-repair process.

## 4. Physical refinement and grid-matching optimization

Physical refinement is performed by the Physical Refinement Tool when $s_{\text{phy}}=\text{True}$ [2510.02178]. The tool is designed to correct coordinate-level errors including collisions, out-of-bounds placements, and wall-alignment violations using a grid-matching algorithm [2510.02178]. The update rule is
$$
P_j=
\begin{cases}
P''_j, & \text{if } s_{\text{phy}}=\text{False},\\
f_{\text{PRT}}(P''_j), & \text{if } s_{\text{phy}}=\text{True}.
\end{cases}
$$

The floor is discretized into a uniform grid $M=\{m_1,\dots,m_k\}$ [2510.02178]. Repositioning is constrained by three validity criteria. First, placements must be collision-free, with collision measured by
$$
\text{IoU}(R_i,R_j)=\frac{\text{Area}(R_i\cap R_j)}{\text{Area}(R_i\cup R_j)},
$$
where $R_i$ is object $i$’s oriented footprint rectangle on the floor [2510.02178]. Second, each footprint must be fully contained in $B_{\text{room}}$ [2510.02178]. Third, if $c_{\text{wall}}=\text{True}$, the object’s back face must be flush with the nearest wall and its rotation must be updated to face inward [2510.02178].

For an invalid object $i$ with current center $c_i=(x_i,y_i)$, the grid-matching objective is
$$
m^*=\arg\min_{m\in \mathcal{M}_{\text{valid}(i)}} \|m-c_i\|_2,
$$
subject to collision-free placement, room containment, and wall alignment if required [2510.02178]. The solver therefore selects the nearest feasible grid cell to minimize displacement while preserving semantic relationships [2510.02178].

The paper specifies a three-stage PRT pseudocode. In the wall-alignment stage, if an object is against a wall, the nearest wall is found, rotation is set by `wall2rotation`, the object’s back center is computed, candidate wall grids are sorted by distance, and the object is either pulled to the nearest feasible wall grid or deleted if none exists [2510.02178]. In the out-of-bounds correction stage, out-of-bounds objects are detected, general grid candidates are sorted by distance, invalid candidates are filtered, and the nearest valid position is chosen or the object is deleted if no candidate exists [2510.02178]. In the collision-resolution stage, collision pairs are found, the smaller object in each pair is selected, candidate grids are sorted by distance, collision-inducing candidates are filtered, and the nearest feasible position is assigned or the smaller object is deleted if no feasible cell exists [2510.02178].

The paper notes that footprint handling uses object sizes and rotations restricted to $0/90/180/270$ degrees to compute oriented bounding boxes, and that collision checks use occupancy maps with zero-overlap enforced by IoU [2510.02178]. Complexity analysis is not provided [2510.02178]. A notable consequence of the algorithm is that physical feasibility is guaranteed only up to the resolution of the chosen grid and the fidelity of the asset bounding boxes.

## 5. Models, interfaces, and implementation details

GPT-4o is used both for prompt generation in dataset construction and for agent reasoning in the Planner, Designer, Evaluator, and SRT through structured VQA and constrained JSON outputs [2510.02178]. The Designer operates in a 2D coordinate system whose origin is at the bottom-left corner $(0,0)$, with positions expressed as positive integers in centimeters and rotations restricted to $\{0,90,180,270\}$ degrees, where $0^\circ$ faces $+Y$, $90^\circ$ faces $+X$, $180^\circ$ faces $-Y$, and $270^\circ$ faces $-X$ [2510.02178]. Visual inputs are top-down rendered images with brown floors and white walls, and the Evaluator consumes these images together with JSON metadata [2510.02178].

The Planner outputs semantic asset groups such as `group1`, `group2`, and orders them by importance, typically placing larger or focal assets first [2510.02178]. Its relationship vocabulary is exactly `[near, side of, in front of, aligned with, opposite, around]` [2510.02178]. The Designer takes the previously placed objects (`arranged_objects`) and the top-down render $I_{k-1}$ as inputs and is instructed to output JSON only [2510.02178]. The Evaluator performs semantic assessment via targeted yes/no questions derived from $c_{\text{rel}}$ and $c_{\text{rot}}$, and a separate binary physical assessment that determines whether a physical refinement tool is required [2510.02178].

The asset catalog is retrieved from Objaverse and curated for semantic fidelity and geometric quality following the Holodeck protocol [2510.02178]. The paper does not report grid resolution, architectural hyperparameters, training or fine-tuning details, or runtime and memory usage; PRT is explicitly heuristic rather than learned [2510.02178]. Code is stated to be publicly available, and the appendix fully specifies prompts for the Planner, Designer, Evaluator, and SRT, along with the grid-matching pseudocode for PRT [2510.02178].

These interface decisions make the framework unusually transparent relative to end-to-end neural layout generators. A plausible implication is that reproducibility depends less on hidden weights in a trained physical module and more on prompt fidelity, rendering setup, and exact implementation of the grid-matching routines.

## 6. Evaluation, baselines, and empirical behavior

The evaluation uses a curated test set of 45 indoor scenes spanning 9 categories: five common categories—bathroom, bedroom, dining room, kitchen, living room—and four uncommon categories—buffet restaurant, classroom, children’s room, home gym [2510.02178]. Prompts are synthesized by GPT-4o and range from simple labels to multi-clause specifications; assets are retrieved via Objaverse and manually screened [2510.02178]. Baselines are LayoutGPT, Holodeck, and LayoutVLM [2510.02178].

Physical plausibility is measured with Collision Rate (CR), the proportion of object pairs with footprint IoU greater than zero,
$$
\text{CR}=\frac{2}{n(n-1)}\sum_{1\le i<j\le n}\mathbf{1}\big[\text{IoU}(R_i,R_j)>0\big],
$$
and Out-of-Bounds Rate (OOB), the average fraction of each footprint area that lies outside the room boundaries,
$$
\text{OOB}=\frac{1}{n}\sum_{i=1}^n \frac{\text{Area}(R_i\setminus B_{\text{room}})}{\text{Area}(R_i)}.
$$
Semantic coherence is evaluated with GPT-4o scores for Positional Coherency and Rotational Coherency [2510.02178].

The main quantitative results reported for the average over the nine categories are summarized below.

| Method | Pos. / Rot. | Col. / OOB |
|---|---:|---:|
| DisCo-Layout | 67.89 / 66.88 | 0.00 / 0.00 |
| LayoutGPT | 66.78 / 63.38 | 5.10 / 18.01 |
| Holodeck | 64.8 / 58.2 | 0.00 / 0.25 |
| LayoutVLM | 63.31 / 58.74 | 9.09 / 8.54 |

The paper highlights zero physical violations across all nine categories for DisCo-Layout, with Collision $=0.00$ and OOB $=0.00$ [2510.02178]. It also reports category-specific results including Dining Room with Pos. $=87.8$, Rot. $=88.5$, Collision $=0.00$, OOB $=0.00$, and Living Room with Pos. $=69.6$, Rot. $=74.2$, Collision $=0.00$, OOB $=0.00$ [2510.02178].

Ablation studies attribute improvements to the disentangled refinement design. Removing SRT yields Collision $=0.73$, OOB $=1.08$, Pos. $=67.27$, Rot. $=64.67$; removing PRT yields Collision $=12.21$, OOB $=10.21$, Pos. $=66.22$, Rot. $=63.13$; removing both yields Collision $=10.76$, OOB $=12.22$, Pos. $=66.97$, Rot. $=64.16$; replacing structured VQA with open-ended VQA yields Collision $=1.24$, OOB $=0.99$, Pos. $=67.6$, Rot. $=64.4$; the full system achieves Collision $=0.00$, OOB $=0.00$, Pos. $=67.89$, Rot. $=66.88$ [2510.02178]. The paper’s interpretation is that disentangled refinement coordinated by structured VQA yields the best overall quality [2510.02178].

Qualitatively, the framework is described as correcting orientations such as chairs facing a table, enforcing relationship types like near, aligned, and opposite, creating functional zones such as a bar cart adjacent to a dining table, resolving collisions, pulling large objects against walls, and ensuring in-bounds placement [2510.02178]. When PRT cannot find a feasible grid position, it deletes the offending object, which prevents persistent physical violations [2510.02178].

## 7. Relation to decoupled layout generation, limitations, and significance

The name “DisCo-Layout” refers specifically to the 2025 framework for 3D indoor layout synthesis that disentangles semantic and physical refinement [2510.02178]. The data also identifies a conceptual connection to LDGM, a 2023 method for 2D layout generation in formatted scenes. LDGM treats any coarsely specified or partially observed layout as an intermediate diffusion state and uses a decoupled diffusion process across category, position, and size, while learning a single joint reverse model with a transformer backbone and relation-aware attention [2303.05049]. The paper summary states that LDGM and “DisCo-Layout” share the core idea of decoupling diffusion across layout attributes, but also states explicitly that LDGM does not reference “DisCo-Layout” and should be regarded as a different method that embodies the same decoupling principle in a discrete diffusion framework [2303.05049].

This comparison is significant because the two works instantiate decoupling at different levels. In LDGM, the separation occurs in the forward corruption process over layout attributes, followed by joint reverse denoising [2303.05049]. In DisCo-Layout, the separation occurs between semantic correction and physical correction, mediated by a multi-agent orchestration loop [2510.02178]. This suggests a broader design pattern in layout generation: decomposing heterogeneous constraints into specialized operators while retaining some coordinating mechanism.

The limitations of DisCo-Layout are stated directly. The framework depends on GPT-4o for reasoning and is sensitive to prompt design; open-ended VQA is less reliable than structured VQA [2510.02178]. The discrete grid and restricted rotation set may limit placement resolution and nuanced orientations [2510.02178]. PRT may delete objects when no grid cell satisfies all constraints [2510.02178]. The method contains no explicit training or learned physical model, so the heuristic PRT may be sensitive to grid resolution and asset bounding-box quality [2510.02178]. The approach also relies on manual asset screening and Objaverse, so domain shift may affect semantic judgments [2510.02178]. Future work is identified as enhancing generalization to more complex scenarios and improving training for embodied AI applications, potentially with learned refinement components [2510.02178].

Within the scope defined by the paper, DisCo-Layout’s main contribution is the claim that semantic and physical layout quality benefit from being refined separately but coordinated explicitly. The reported zero collision and zero out-of-bounds rates, together with higher positional and rotational coherency than the listed baselines, support that claim on the 45-scene benchmark used in the study [2510.02178].

Source: https://www.emergentmind.com/topics/disco-layout