arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.01693cs.RO
收藏

GSR：为具身操作学习结构化推理

GSR: Learning Structured Reasoning for Embodied Manipulation

Klmmotion

Kewei Hu, Michael Zhang, Wei Ying, Tianhao Liu, Guoqiang Hao, Zimeng Li, Wanchan Yu, Jiajian Jing, Fangwen Chen, Hanwen Kang

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

GSR通过显式建模世界状态演变，提升具身操作中长周期任务的完成能力和零样本泛化性能。

摘要：

尽管取得了快速进展，具身智能体仍然在需要维持空间一致性、因果依赖性和目标约束的长周期操作中遇到困难。现有方法的关键限制在于任务推理隐含地嵌入在高维潜在表示中，使得难以将任务结构与感知变化区分开来。我们引入了Grounded Scene-graph Reasoning（GSR），一种结构化推理范式，它显式地将世界状态演变建模为在语义上接地的场景图上的转换。通过逐步推理对象状态和空间关系，而不是直接将感知映射到动作，GSR能够在物理上接地的空间中显式推理动作前提、后果和目标满足。为了支持学习此类推理，我们构建了Manip-Cognition-1.6M，一个大规模数据集，共同监督世界理解、动作规划和目标解释。在RLBench、LIBERO、GSR基准测试和现实世界机器人任务上的广泛评估表明，GSR在提示基线上的零样本泛化和长周期任务完成方面有显著提升。这些结果突显了显式世界状态表示作为可扩展具身推理的关键归纳偏置。

英文摘要：

Despite rapid progress, embodied agents still struggle with long-horizon manipulation that requires maintaining spatial consistency, causal dependencies, and goal constraints. A key limitation of existing approaches is that task reasoning is implicitly embedded in high-dimensional latent representations, making it challenging to separate task structure from perceptual variability. We introduce Grounded Scene-graph Reasoning (GSR), a structured reasoning paradigm that explicitly models world-state evolution as transitions over semantically grounded scene graphs. By reasoning step-wise over object states and spatial relations, rather than directly mapping perception to actions, GSR enables explicit reasoning about action preconditions, consequences, and goal satisfaction in a physically grounded space. To support learning such reasoning, we construct Manip-Cognition-1.6M, a large-scale dataset that jointly supervises world understanding, action planning, and goal interpretation. Extensive evaluations across RLBench, LIBERO, GSR-benchmark, and real-world robotic tasks show that GSR significantly improves zero-shot generalization and long-horizon task completion over prompting-based baselines. These results highlight explicit world-state representations as a key inductive bias for scalable embodied reasoning.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

↑