M2-Reasoning: Empowering MLLMs with Unified General and Spatial Reasoning

2025-07-11

AI Summary

Multimodal large language models (MLLMs) generally lack dynamic spatial interaction capabilities and general-purpose spatial reasoning. To address this, we propose a unified spatial-general joint reasoning framework. Our method introduces three core innovations: (1) construction of a high-quality, dynamically annotated spatial interaction dataset; (2) a staged optimization training paradigm integrating cold-start fine-tuning, multi-task learning, and verifiable reward-guided reinforcement learning (RLVR); and (3) task-specific reward signals to precisely steer spatial relation modeling. The approach significantly enhances model robustness in understanding and reasoning about complex, dynamic spatial scenarios. Evaluated on eight diverse benchmarks—including both general multimodal reasoning and spatial reasoning tasks—our framework achieves state-of-the-art performance across all metrics: average improvement of 4.2% on general reasoning tasks and up to 12.7% accuracy gain on spatial interaction tasks.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent advancements in Multimodal Large Language Models (MLLMs), particularly through Reinforcement Learning with Verifiable Rewards (RLVR), have significantly enhanced their reasoning abilities. However, a critical gap persists: these models struggle with dynamic spatial interactions, a capability essential for real-world applications. To bridge this gap, we introduce M2-Reasoning-7B, a model designed to excel in both general and spatial reasoning. Our approach integrates two key innovations: (1) a novel data pipeline that generates 294.2K high-quality data samples (168K for cold-start fine-tuning and 126.2K for RLVR), which feature logically coherent reasoning trajectories and have undergone comprehensive assessment; and (2) a dynamic multi-task training strategy with step-wise optimization to mitigate conflicts between data, and task-specific rewards for delivering tailored incentive signals. This combination of curated data and advanced training allows M2-Reasoning-7B to set a new state-of-the-art (SOTA) across 8 benchmarks, showcasing superior performance in both general and spatial reasoning domains.

Problem

• MLLMs lack dynamic spatial interaction capabilities
• Need unified general and spatial reasoning in MLLMs
• Existing models struggle with real-world spatial applications

Innovation

• Generates 294.2K high-quality data samples
• Uses dynamic multi-task training strategy
• Integrates step-wise optimization and task-specific rewards