World2Mind: Cognition Toolkit for Allocentric Spatial Reasoning in Foundation Models

Researchers introduce World2Mind, a training-free spatial intelligence toolkit that enhances foundation models' 3D spatial reasoning capabilities by up to 18%. The system uses 3D reconstruction and cognitive mapping to create structured spatial representations, enabling text-only models to perform complex spatial reasoning tasks.

• →World2Mind improves frontier models like GPT-5.2 performance by 5-18% in spatial reasoning tasks without requiring additional training.
• →The toolkit uses 3D reconstruction and instance segmentation to create structured spatial cognitive maps for foundation models.
• →Text-only foundation models can achieve complex 3D spatial reasoning using only the AST-structured text representation.
• →The system addresses current limitations where multimodal models either overfit on 3D data or remain confined to 2D perception.
• →World2Mind introduces a three-stage reasoning chain to mitigate 3D reconstruction inaccuracies and improve spatial understanding.