Spatial Reasoning
Overview
Spatial reasoning concerns how people and AI systems represent where things are, how they relate, and how those relations change with viewpoint or movement. Common approaches use grids, graphs, coordinate frames, and 3D cognitive maps; others add geometric tools, visual imagination, active viewing, or reinforcement learning with verifiable rewards. Evaluations range from grid mazes and multi-view question answering to egocentric video, simulated navigation and manipulation, and GIS workflows. A central challenge is that success on isolated spatial questions often fails to transfer to consistent world models, perspective transformations, and reliable action over time.
Latest papers
165 papers8 papers in the last four weeks, down 67% on the four weeks before. 0.1% of all new papers.
Beyond Spatial Benchmarks: From Spatial Reasoning to Navigation
Metric-Bench: Exploring In-context Spatial Metric Reasoning in VLMs for Indoor Scenes
Spatial-Interactor: Learning Spatial Reasoning through Interaction with the Observable Physical World
A Qualitative Model for Reasoning about Path and Support
VABench: Measuring Embodied Spatial Intelligence through Visual Demonstrations, Active Perception, and Metric Control
Enhancing Human Mobility Prediction with Spatially Aware LLM-based Multi-Agent Systems
MV-STRIDE: Enabling MLLMs to Master Multi-View Spatial Reasoning via Hierarchical Capability Modeling
Geometry-Aware Test-Time Learning for Quantitative Spatial Reasoning
Unfold The World: Factorize 4D Properties in Reinforcing Spatial Reasoning
flat'' when reasoning about the physical world. We argue that this spatial bottleneck stems from a profound dimensional mismatch: while VLMs are trained to interpret 2D projections, true spatial reasoning demands the recovery of latent 3D geometry and temporal continuity. To conquer this high-dimensional complexity, we advocate a shift from monolithic learning to a divide and conquer'' paradigm. We present FactoSR, a factorized reinforcement learning framework that explicitly interpret the dimensions collapsed by visual projection. At its core, FactoSR decomposes the monolithic problem of world-consistent reasoning into three orthogonal, geometric sub-objectives: planar correspondence (), depth consistency (), and temporal reversibility (). By optimizing these verifiable constraints within a unified policy learning mechanism, we effectively transform an ill-posed projection recovery problem into a series of tangible reasoning steps. Extensive evaluations on multi-view and video benchmarks demonstrate that this elegant decomposition yields substantial gains in 3D and 4D reasoning, achieving a 5.9% boost on VSI-Bench and 4.5% on All-Angles-Bench. Our findings suggest that reinforcing explicit, factorized 4D consistency is a critical step toward evolving VLMs into robust, world-aware reasoners.