STAR-R1: Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMs

TL;DR

This paper introduces STAR-R1, a novel reinforcement learning framework that significantly improves spatial reasoning in multimodal large language models by efficiently exploring and accurately identifying object transformations across views.

Contribution

STAR-R1 integrates a single-stage RL paradigm with a fine-grained reward mechanism, advancing spatial reasoning capabilities in multimodal models beyond traditional fine-tuning methods.

Findings

STAR-R1 outperforms supervised fine-tuning by 23% in cross-view spatial reasoning.

Achieves state-of-the-art results across 11 evaluation metrics.

Exhibits human-like reasoning behaviors and object comparison abilities.

Abstract

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities across diverse tasks, yet they lag significantly behind humans in spatial reasoning. We investigate this gap through Transformation-Driven Visual Reasoning (TVR), a challenging task requiring identification of object transformations across images under varying viewpoints. While traditional Supervised Fine-Tuning (SFT) fails to generate coherent reasoning paths in cross-view settings, sparse-reward Reinforcement Learning (RL) suffers from inefficient exploration and slow convergence. To address these limitations, we propose STAR-R1, a novel framework that integrates a single-stage RL paradigm with a fine-grained reward mechanism tailored for TVR. Specifically, STAR-R1 rewards partial correctness while penalizing excessive enumeration and passive inaction, enabling efficient exploration and precise reasoning. Comprehensive evaluations demonstrate that STAR-R1 achieves state-of-the-art performance across all 11 metrics, outperforming SFT by 23% in cross-view scenarios. Further analysis reveals STAR-R1's anthropomorphic behavior and highlights its unique ability to compare all objects for improving spatial reasoning. Our work provides critical insights in advancing the research of MLLMs and reasoning models. The codes, model weights, and data will be publicly available at https://github.com/zongzhao23/STAR-R1.