STAR-R1：通过强化多模态大模型实现多视图空间变换推理
STAR-R1: Multi-View Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMs
把可验证奖励的强化学习（RLVR）与过程监督 SFT 两阶段结合，用于多视图空间变换推理；其「指代感知奖励」设计可直接迁移到需要跨视图物体对应的 VLM 空间理解任务（如 MMSI-Bench、MindCube 类基准）。
原文依据 a two-stage framework that combines process-supervised SFT with a referential-aware RL paradigm
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论多模态大语言模型（MLLMs）在多视图空间推理上仍远未达到人类水平，该任务要求模型在不同视图之间建立物体对应关系并推断连贯的场景语义。我们通过变换驱动的视觉推理（TVR）任务分析这一局限，发现监督微调（SFT）无法捕捉跨视图一致性，而强化学习（RL）无法可靠地识别关键指代物体。为弥合这一差距，我们提出多视图空间变换推理（STAR-R1），一个将过程监督 SFT 与指代感知 RL 范式相结合的两阶段框架。STAR-R1 首先从高质量思维链中学习结构化的空间推理轨迹，随后利用针对指代选择与答案正确性的细粒度奖励，鼓励有效探索与稳健的场景解读。尽管仅使用少量高质量训练数据，STAR-R1 在多视图空间理解基准 TVR、MMSI-Bench、MindCube-Bench 和 SPAR-Bench 上超越了使用远多训练数据的最优模型。我们的研究揭示了 RL 在多视图空间理解中被忽视的潜力，并为 MLLMs 实现更接近人类的空间推理指明了路径。
英文原文摘要 1305 字符
Multimodal Large Language Models (MLLMs) remain far from human-level performance in multi-view spatial reasoning, where models must establish object correspondences across view and infer coherent scene semantics. We analyze this limitation through the Transformation-Driven Visual Reasoning (TVR) task and find that Supervised Fine-Tuning (SFT) fails to capture cross-view consistency, whereas reinforcement learning (RL) fails to reliably identify key referential objects. To bridge this gap, we introduce multi-View Spatial TrAnsformation Reasoning (STAR-R1), a two-stage framework that combines process-supervised SFT with a referential-aware RL paradigm. STAR-R1 first learns structured spatial reasoning trajectories from high-quality CoTs and then uses fine-grained rewards on referential selection and answer correctness to encourage effective exploration and robust scene interpretation. Despite using only a small amount of high-quality training data, STAR-R1 surpasses state-of-the-art models with far more training data on the multi-view spatial understanding benchmarks TVR, MMSI-Bench, MindCube-Bench, and SPAR-Bench. Our study reveals the overlooked potential of RL in multi-view spatial understanding and points a way toward potentially achieving more human-like spatial reasoning in MLLMs.