COOPER: A Unified Model for Cooperative Perception and Reasoning in Spatial Intelligence

TL;DR

COOPER is a unified multimodal large language model that integrates perception and reasoning to improve 3D-aware spatial understanding, demonstrating significant performance gains in spatial reasoning tasks.

Contribution

This work introduces COOPER, a novel unified model that combines perception and reasoning in a two-stage training process for enhanced spatial intelligence.

Findings

COOPER achieves a 6.91% improvement in spatial reasoning.

A variant trained only for auxiliary modality generation gains 7.92% in distance and size estimation.

Learning auxiliary modalities enhances internal spatial knowledge.

Abstract

Visual Spatial Reasoning is crucial for enabling Multimodal Large Language Models (MLLMs) to understand object properties and spatial relationships, yet current models still struggle with 3D-aware reasoning. Existing approaches typically enhance either perception, by augmenting RGB inputs with auxiliary modalities such as depth and segmentation, or reasoning, by training on spatial VQA datasets and applying reinforcement learning, and thus treat these two aspects in isolation. In this work, we investigate whether a unified MLLM can develop an intrinsic ability to enhance spatial perception and, through adaptive interleaved reasoning, achieve stronger spatial intelligence. We propose \textbf{COOPER}, a unified MLLM that leverages depth and segmentation as auxiliary modalities and is trained in two stages to acquire auxiliary modality generation and adaptive, interleaved reasoning capabilities. COOPER achieves an average \textbf{6.91\%} improvement in spatial reasoning while maintaining general performance. Moreover, even a variant trained only for auxiliary modality generation attains a \textbf{7.92\%} gain on distance and size estimation, suggesting that learning to generate auxiliary modalities helps internalize spatial knowledge and strengthen spatial understanding.