CausalSpatial: A Benchmark for Object-Centric Causal Spatial Reasoning

TL;DR

CausalSpatial introduces a benchmark to evaluate object-centric causal spatial reasoning in models, revealing significant gaps in current multimodal models' ability to predict consequences in 3D scenes, and proposes a new framework to improve this reasoning.

Contribution

The paper presents CausalSpatial, a new benchmark for causal spatial reasoning, and introduces the COW model that externalizes simulation to enhance model grounding in physical reality.

Findings

Humans score 84% on the benchmark, GPT-5 scores 54%.

Current models rely heavily on textual reasoning, leading to ungrounded hallucinations.

COW improves reasoning by generating videos of hypothetical dynamics.

Abstract

Humans can look at a static scene and instantly predict what happens next -- will moving this object cause a collision? We call this ability Causal Spatial Reasoning. However, current multimodal large language models (MLLMs) cannot do this, as they remain largely restricted to static spatial perception, struggling to answer "what-if" questions in a 3D scene. We introduce CausalSpatial, a diagnostic benchmark evaluating whether models can anticipate consequences of object motions across four tasks: Collision, Compatibility, Occlusion, and Trajectory. Results expose a severe gap: humans score 84% while GPT-5 achieves only 54%. Why do MLLMs fail? Our analysis uncovers a fundamental deficiency: models over-rely on textual chain-of-thought reasoning that drifts from visual evidence, producing fluent but spatially ungrounded hallucinations. To address this, we propose the Causal Object World model (COW), a framework that externalizes the simulation process by generating videos of hypothetical dynamics. With explicit visual cues of causality, COW enables models to ground their reasoning in physical reality rather than linguistic priors. We make the dataset and code publicly available here: https://github.com/CausalSpatial/CausalSpatial