Spatial context prediction requires visual grounding and spatial reasoning to identify vacant regions relative to objects in a scene. Spatial relationships and physical constraints remain challenging for large vision-language models (VLMs). Recursive reasoning has demonstrated strong performance on discrete and abstract reasoning tasks, motivating its use for improving spatial context prediction in natural images. Weight sharing enables higher accuracy with a small trainable parameter count. We introduce Spatial-TRM, a compact recursive model with frozen vision-language features, a convex region decoder, and just 11.2M trainable parameters. This weight-tied reasoner supports six spatial relationships, three reference frames, and open-vocabulary anchor objects. On the RoboSpatial-Home benchmark, Spatial-TRM achieves 29.51% context accuracy, a gain of 9.02 percentage points (44.0% relative) over a separately trained single-step model with the same parameter count. Including the frozen encoders, the model has 877.8M total parameters, approximately 15× fewer than the reported 13B size of RoboPoint. These results show that repeated refinement improves spatial context prediction from pretrained vision-language features at a fixed trainable parameter count on this benchmark.
Figures & tables
Figure 1: Example of a spatial-context query and its region target. The prompt specifies the bed as the anchor and asks for vacant space to its left in the object’s reference frame. The mask is constructed from the annotated target points.
Figure 2: (a) The decoder predicts an interior point and distances to half-plane boundaries at fixed normal directions, whose intersection defines a convex polygon. (b) Each pixel is assigned a logit based on its minimum margin across the half-planes.
Model
Trainable parameters
Accuracy (%)
Feedforward
4.46M
17.21
Single-step
11.2M
20.49
Recursive (ours)
11.2M
29.51
Table 1: Reasoning ablation on RoboSpatial-Home. The separately trained single-step and recursive models both have 11.2M trainable parameters; repeated refinement improves context accuracy by 9.02 percentage points. All variants use the same frozen encoder.
Figure 3: Through iterative updates, the recursive model updates its current answer state over a fixed budget. Foreground regions added at each step are represented in green and removed regions in red. The final prediction at step n is evaluated using the interior point marked by a star.
Model
Accuracy (%)
Parameters
Qwen2.5-VL-3B ( Bai et al., 2025 )
0.00
3B
Molmo-7B ( Deitke et al., 2025 )
0.82
7B
Qwen2.5-VL-32B ( Bai et al., 2025 )
3.28
32B
GPT-4o ( OpenAI, 2024 )
6.60
-
Gemini 2.5 Flash ( Gemini Team, 2025 )
7.90
-
Gemini 2.5 Pro ( Gemini Team, 2025 )
8.30
-
Table 2: RoboSpatial-Home context accuracy (higher is better). Baseline parameter counts are nominal model sizes; the SpaceTools count excludes external tools. Our count includes the frozen encoders. ∗ denotes fine-tuning on RoboSpatial; a dash indicates an unreported parameter count.
Encoder
Accuracy (%)
IoU
Dice
Parameters
DINOv3 ViT-L/16
29.51
0.11
0.16
877.8M
CLIP ViT-L/14
26.23
0.08
0.12
438M
SigLIP-SO400M/14
16.39
0.07
0.12
888M
Table 3: Encoder ablation on RoboSpatial-Home. CLIP halves the total parameter count relative to DINOv3, with a 3.28-percentage-point accuracy reduction. Parameter counts include the full model. Bold marks the best value in each numeric column.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 4: Context accuracy on the RoboSpatial-Home benchmark at 25-epoch intervals over the entire training duration. The checkpoint selected based on validation performance is marked with a red star.
Figure 5: Predictions whose representative points fall outside the annotated target masks. Ground truth in green, predicted mask in red, and representative point marked with a star.
Figure 6: Examples of predicted regions with limited overlap with the annotation-derived targets. Ground truth masks shown in green, predicted masks shown in red.
Large Vision-Language Models (LVLMs) have made remarkable progress across visual perception tasks, yet spatial reasoning remains a persistent weakness, especially for questions that require reasoning over visual space. Recent spatial-reasoning methods incorporate generated grounding, where models predict bounding boxes, masks, or other localization outputs for task-relevant objects as part of their reasoning trace. However, these approaches typically optimize final-answer correctness alone, allowing correct answers to be rewarded even when the model does not reason from confidently localized task-relevant objects. We introduce SpatialCORE (Spatially COnfident REasoning), a post-training framework that turns the model's own confidence in generated grounding into a learning signal for spatial reasoning. Its central idea is to reinforce grounding that is both accurate and confident, encouraging the model to reason from confidently localized task-relevant objects. SpatialCORE realizes this through a self-regulating spatial reward that weights each predicted bounding box's matching quality by its coordinate-token confidence. An answer gate further ties grounding optimization to final-answer correctness. SpatialCORE achieves state-of-the-art results among open-source and specialized spatial reasoning models across diverse benchmarks, and transfers effectively in zero-shot settings to unseen data distributions. The source code is available at https://github.com/rafiibnsultan/SpatialCORE.
Rafi Ibn Sultan, Xiangyu Zhou, Md. Sajid Alam Chowdhury +4
Department of Computer Science, Wayne State University · Department of Radiation Oncology, Henry Ford Health · Department of Electrical and Computer Engineering, The Ohio State University +1
Existing spatial reasoning benchmarks mainly test spatial perception: reading off relations already visible in the input. Yet real-world spatial intelligence demands predictive spatial reasoning: constructing a scene from observations, anticipating how an intervention changes it, and reasoning about the unseen outcome. We introduce SpaceCast-Bench, the first benchmark to directly and diagnostically evaluate this capability. Built around an observe-transform-infer framework, its 3,862 questions from 182 real-world scenes span 16 task types at three levels: static perception, local prediction, and global prediction, progressively requiring scene understanding, spatial state updating, and relational inference over unobserved outcomes. Evaluating 21 models exposes a stark gap: the strongest model reaches only 58.0% against 87.2% human performance, while spatially specialized models remain near random chance. Controlled analyses further reveal that bridge views are critical for integrating distributed observations, and that explicit 3D evidence benefits models more reliably than generated outcome images or videos. Fine-tuning on our programmatically generated data lifts Qwen3-VL-4B from 34.0% to 65.7% with macro-average gains across six out-of-domain benchmarks.
Predicting future spatial states supports collision avoidance and timely decision-making in dynamic environments. However, existing vision-language models (VLMs) and benchmarks for spatial reasoning primarily focus on observed scenes, leaving predictive spatial reasoning beyond the observed interval underexplored. To this end, we introduce SpatialMind, a metric-scale VLM for spatial reasoning and future prediction. Its metric depth adapter anchors spatial reasoning to real-world scale, while its progressive state chain establishes current spatial states and observed dynamics as the foundation for future prediction. Given a video prefix, SpatialMind predicts distances, motion directions, and spatial relations in both observed and unseen future frames. For training and evaluation, we build a scalable data engine that grounds entity descriptions in metric geometry to generate question-answer pairs and state supervision. Using this engine, we construct the SpatialMind-30K dataset and the SpatialMind-2K benchmark, both covering driving and everyday egocentric scenes. The benchmark spans eight tasks across three levels: current-state understanding, observed-dynamics understanding, and future prediction. Experiments show that SpatialMind substantially outperforms both general and spatially specialized models on our benchmark while achieving competitive zero-shot performance on VSI-Bench, OSI-Bench, and VLM4D.
Feiran Wang, Xiaoqi Wang, Ziwei Li +3
University of Illinois at Chicago · Bosch Research North America · Bosch Center for Artificial Intelligence (BCAI)