SpatialWorld: Benchmarking Interactive Spatial Reasoning of Multimodal Agents in Real-World Tasks

2026-06-08

AI Summary

Existing benchmarks struggle to evaluate the interactive spatial understanding capabilities of multimodal agents in realistic settings. To address this gap, this work proposes SpatialWorld, a unified benchmark that integrates eight heterogeneous simulation backends and 760 human-annotated tasks, requiring agents to actively explore and complete complex real-world tasks from a first-person, partially observable perspective. SpatialWorld introduces the first simulator-agnostic protocol, enabling cross-domain, long-horizon, and active-perception-based spatial reasoning evaluation, and provides human-validated initial states, reference trajectories, and an automated final-state verification mechanism. Evaluation of 15 state-of-the-art agents reveals that even the strongest model, GPT-5, achieves only a 17.4% success rate, highlighting significant bottlenecks in current systems’ active exploration and planning efficiency.

Abstract

Spatial reasoning is a foundational capability for multimodal large language models (MLLMs) to perceive and operate within the physical world. However, existing benchmarks predominantly rely on passive evaluation (e.g., static VQA) or simulator-specific pipelines, failing to assess general interactive spatial understanding. We introduce SpatialWorld, a unified benchmark designed specifically for evaluating the interactive spatial understanding of multimodal agents in complex real-world tasks. Integrating eight heterogeneous simulation backends under a shared, simulator-agnostic protocol, SpatialWorld features 760 human-annotated tasks across diverse domains (e.g., household routines, travel, social collaboration). Agents must solve tasks under vision-only partial observability, actively gathering egocentric visual evidence and expressing decisions via a unified, text-based action interface native to MLLMs. For reliable evaluation, each task includes a human-validated initial state, a reference trajectory, and a terminal-state verifier. Evaluating 15 advanced agents reveals that robust spatial task solving remains challenging: the strongest model, GPT-5, achieves an average task success rate (TSR) of only 17.4%, while the leading open-source model, Qwen-3.5, reaches 14.1%. Further analysis exposes a clear mismatch between task success and execution efficiency, alongside substantial domain-specific performance variations. These bottlenecks in active exploration and long-horizon planning position SpatialWorld as a rigorous testbed for future spatial agents.

Problem

• spatial reasoning
• multimodal agents
• interactive evaluation
• real-world tasks
• benchmarking

Innovation

• interactive spatial reasoning
• multimodal agents
• simulator-agnostic benchmark
• egocentric vision
• real-world task evaluation