SpatialDreamer: Incentivizing Spatial Reasoning via Active Mental Imagery
This addresses the need for human-like active spatial mental simulation in MLLMs, representing a critical advancement in the field.
The paper tackles the problem of limited spatial reasoning in Multi-modal Large Language Models by proposing SpatialDreamer, a reinforcement learning framework that uses active exploration and visual imagination, achieving competitive results on multiple benchmarks.
Despite advancements in Multi-modal Large Language Models (MLLMs) for scene understanding, their performance on complex spatial reasoning tasks requiring mental simulation remains significantly limited. Current methods often rely on passive observation of spatial data, failing to internalize an active mental imagery process. To bridge this gap, we propose SpatialDreamer, a reinforcement learning framework that enables spatial reasoning through a closedloop process of active exploration, visual imagination via a world model, and evidence-grounded reasoning. To address the lack of fine-grained reward supervision in longhorizontal reasoning tasks, we propose Geometric Policy Optimization (GeoPO), which introduces tree-structured sampling and step-level reward estimation with geometric consistency constraints. Extensive experiments demonstrate that SpatialDreamer delivers highly competitive results across multiple challenging benchmarks, signifying a critical advancement in human-like active spatial mental simulation for MLLMs.