acceptodds
Withdrawn from ICLR 2027

VALERANT: An Automatic Navigable Game Map Generator via Action-Conditioned World Model Exploration

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

10 views

Abstract

World Action Models (WAMs) couple predictive world modeling with action generation, allowing anticipated future states to guide agent behavior. Although WAMs are rapidly advancing embodied AI, general-purpose counterparts remain largely unexplored in games. Existing game-oriented approaches often combine action-conditioned world models with external policies and reward functions to realize WAM-like decision-making, yet they operate mainly in 2D visual observation space and do not instantiate persistent 3D geometry. Extending this paradigm to 3D games introduces a distinct challenge. In autonomous driving and robotics, the physical environment exists independently of the model, providing a persistent 3D world in which selected actions can be executed. Games have no such external substrate; the virtual world itself must be instantiated. Most playable games require a persistent and navigable space, while 3D games additionally require explicit geometry that supports movement and interaction. Action-conditioned video rollouts provide visual observations but not this spatial representation. We present VALERANT, a training-free framework that transforms a pretrained action-conditioned world model into a WAM for exploring and constructing 3D game maps. By coupling predictive visual rollouts with SLAM-based spatial reconstruction and exploration-driven action selection, VALERANT progressively transforms a single image into a persistent 3D game map. This framework extends WAM-based interaction beyond 2D visual simulation and offers a new approach to reducing manual effort in 3D game-map creation.

Primary area:
generative models
Keywords:
World Model, World Action Model, 3D Generation

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_86AJOpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
3
Soundness
1
Read review
Paper summary

VALERANT automatically explores and constructs a 3D point-cloud map from a single game screenshot. At each step, a frozen Matrix-Game 3.0 generates videos for five candidate actions. MASt3R-SLAM reconstructs camera motion and scene geometry, and the system selects a branch that improves coverage and forward progress while passing collision checks. The selected result is integrated into the map. Floor estimation, complementary collision detection, and checkpoint rewind address geometric drift, wall penetration, and local stagnation. Experiments include qualitative results, module ablations, and a visual-preference comparison with HY-World 2.0.

Critical strengths
  • The evaluation includes a relevant baseline. The paper compares against HY-World 2.0, renders both methods' outputs as 3DGS scenes, and reports VLM and human preferences.
  • The system requires no additional training, and the appendices provide substantial implementation information.
Critical weaknesses
  1. Quantitative results are limited to VLM and human visual-preference comparisons with a single baseline, HY-World 2.0. The numbers of test inputs and comparisons are not reported, so the evaluation scale cannot be assessed. The main paper primarily presents a few qualitative examples, without quantitative validation of final-map navigability, collisions, or exploration coverage.
  2. Consistency when revisiting generated regions is not evaluated. The world model uses the latest 16 frames as visual context. Although the paper claims to construct a persistent map, it does not specifically test whether scene structure remains consistent upon revisiting generated regions or whether the reconstructions before and after a revisit align correctly.
  3. The module ablations provide examples without quantitative results. Figure 3 only shows one scene for each of floor estimation, collision detection, and rewind.
Questions for authors
  1. After generation, can authors conduct relevant experiments for path planning to prove "Navigable"?
  2. How many wins and comparisons produce the 80.00% and 83.33% entries in Table 1? How many inputs were used, how were they selected, which VLM version was used, and how many human participants took part?
  3. Can the three modules be disabled individually on the existing input set, reporting collision rates, stalled steps, and coverage under the same exploration budget?
  4. When continued exploration brings the agent back to a previously generated region, are newly generated frames consistent with the existing map? How is this consistency evaluated, and how are any conflicting reconstructions handled?
Reviewer_MsxMOpenReview ↗
Rating
2
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper proposes a method by leveraging world models to generate 3D game environments. It uses SLAM to reconstruct the scene and choose the best action with some additional engineering efforts to improve the plausibility and consistency of the generation results.

Critical strengths
  1. Applying world models to game map design is a novel direction with some real-world application potential.
Critical weaknesses
  1. Lack of novelty. The proposed method uses pretrained world models and existing SLAM methods to reconstruct the environment, which is a straightforward combination with only some engineering contributions to improve the reconstruction quality. The whole pipeline is just an incremental modification upon the Hy-World 2.0 baseline.
  2. The method claims to tackle game map generation. However, the map is expressed as a sparse slam map with only the ground plane extracted, while a complete scene mesh and other interactable objects is often required for gaming applications.
  3. Lack of quantitative and qualitative results (especially video results of the generated map) and comparisons to other state-of-the-art world models.
  4. The scale of experiments is limited, which only tests on a few images based on the Valorant game.
  5. There is no section discussing the limitations / analyzing failure modes.
  6. The action space is discretized into only 5 atomic actions, which maybe too few for more complex exploration trajectories.
Questions for authors
  1. Can the authors provide results with more diverse input images?
  2. Is it possible for the method to also extract a high-quality mesh of the generated scene rather than sparse key points?
Reviewer_47dbOpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
1
Soundness
2
Read review
Paper summary

This paper presents VALERANT, a training-free framework for constructing explorable 3D game maps from a single image. It integrates a pretrained action-conditioned video world model with visual SLAM to generate and reconstruct multiple candidate futures. An exploration policy selects actions based on spatial coverage, forward progress, and collision constraints. The framework further incorporates robust floor estimation, complementary collision detection, and checkpoint-based rewind to improve exploration reliability. Experiments demonstrate its potential for progressive 3D scene reconstruction and exploration.

Critical strengths

1,The paper presents a training-free framework that integrates a pretrained video world model, visual SLAM, and an exploration policy to progressively construct explorable 3D game environments from a single image.

2,The proposed robustness mechanisms address practical challenges such as geometric drift, collision detection, and dead-end exploration. The qualitative results demonstrate the potential of the framework for autonomous 3D scene exploration and reconstruction.

Critical weaknesses

1、The images corresponding to “Single Image,” “Candidate Geometry,” and “3D Game Map” in Figure 1 should be replaced with visualizations directly derived from the experiments. The current illustrative images are not closely related to the paper’s research context or experimental results, making it difficult to intuitively understand the actual input, candidate geometry reconstruction, and final map construction process.

2、This paper lacks a sufficient quantitative comparison with existing methods. Although the authors presented qualitative results and reported the evaluation scores and user preferences based on VLM, the experimental evaluation was still limited to a single baseline comparison.

3、This paper combines the pre-trained video world model, visual SLAM and exploration strategies to achieve 3D scene construction starting from a single image. However, the core technologies employed in this approach have already been studied in detail in previous research. The paper has not fully clarified the substantial innovations of this framework in terms of core mechanisms or algorithms compared to existing methods.

Questions for authors

See the weakness.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.