Scalable Object Relation Encoding for Better 3D Spatial Reasoning in Large Language Models

TL;DR

This paper introduces QuatRoPE, a scalable 3D positional encoding method for large language models that improves spatial reasoning in 3D scenes by efficiently encoding pairwise relations with high geometric fidelity.

Contribution

We propose QuatRoPE, a linear-length, holistic 3D positional embedding for LLMs, and IGRE, a mechanism to limit its influence, enhancing 3D spatial reasoning capabilities.

Findings

QuatRoPE achieves better spatial relation extraction than previous methods.

The approach maintains scene geometric integrity with high spatial consistency.

Experiments show improved reasoning performance on 3D scene understanding tasks.

Abstract

Spatial reasoning focuses on locating target objects based on spatial relations in 3D scenes, which plays a crucial role in developing intelligent embodied agents. Due to the limited availability of 3D scene-language paired data, it is challenging to train models with strong reasoning ability from scratch. Previous approaches have attempted to inject 3D scene representations into the input space of Large Language Models (LLMs) and leverage the pretrained comprehension and reasoning abilities for spatial reasoning. However, models encoding absolute positions struggle to extract spatial relations from prematurely fused features, while methods explicitly encoding all spatial relations (which is quadratic in the number of objects) as input tokens suffer from poor scalability. To address these limitations, we propose QuatRoPE, a novel positional embedding method with an input length that is linear to the number of objects, and explicitly calculates pairwise spatial relations through the dot product in attention layers. QuatRoPE's holistic vector encoding of 3D coordinates guarantees a high degree of spatial consistency, maintaining fidelity to the scene's geometric integrity. Additionally, we introduce the Isolated Gated RoPE Extension (IGRE), which effectively limits QuatRoPE's influence to object-related tokens, thereby minimizing interference with the LLM's existing positional embeddings and maintaining the LLM's original capabilities. Extensive experiments demonstrate the effectiveness of our approaches. The code and data are available at https://github.com/oceanflowlab/QuatRoPE.