الملخص
Spatial coding agents significantly improve spatial reasoning in Multimodal Large Language Models (MLLMs) by using external tools to generate verified execution traces. However, this paradigm inherently suffers from prohibitive inference-time overhead and external dependencies. In this paper, we explore whether an MLLM can internalize this agentic capability to operate entirely tool-free. We begin with a simple observation: prompting an MLLM with summarized execution traces of a spatial coding agent naturally unlocks the model's internal spatial Chain-of-Thought (CoT). Motivated by this, we introduce SpatialOPSD, an on-policy self-distillation framework that internalizes spatial reasoning into a standalone MLLM by formulating verified agent traces as privileged information. To mitigate privileged-information leakage during distillation, we introduce Repetition-Aware Distillation, which combines repetition masking with unlikelihood regularization. Experiments across multiple benchmarks demonstrate that self-distilling SpatialOPSD achieves higher average accuracy than SFT and GRPO on both spatial and OOD datasets, exhibiting superior performance and generalization.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Li, R., Yang, M., Mao, Y., Tao, Y., Hu, L., Yang, B., Xu, Z., Luo, W., & Xu, B. (2026). SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces. https://omanscience.com/ar/articles/spatialopsd-self-distilling-spatial-intelligence-from-verified-coding-agent-traces
MLA 9
Li, Rongxue, et al. "SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces." https://omanscience.com/ar/articles/spatialopsd-self-distilling-spatial-intelligence-from-verified-coding-agent-traces.
شيكاغو (المؤلف–التاريخ)
Li, Rongxue, Meng Yang, Yiru Mao, Yongliang Tao, Lulu Hu, Bin Yang, Zhao Xu, Weihua Luo, and Bowen Xu. 2026. "SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces." https://omanscience.com/ar/articles/spatialopsd-self-distilling-spatial-intelligence-from-verified-coding-agent-traces.
هارفارد
Li, R., Yang, M., Mao, Y., Tao, Y., Hu, L., Yang, B., Xu, Z., Luo, W. and Xu, B. (2026) 'SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces', Available at: https://omanscience.com/ar/articles/spatialopsd-self-distilling-spatial-intelligence-from-verified-coding-agent-traces.
فانكوفر
Li R, Yang M, Mao Y, Tao Y, Hu L, Yang B, et al. SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces. https://omanscience.com/ar/articles/spatialopsd-self-distilling-spatial-intelligence-from-verified-coding-agent-traces
IEEE
R. Li, M. Yang, Y. Mao, Y. Tao, L. Hu, B. Yang, Z. Xu, W. Luo, and B. Xu, "SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces," https://omanscience.com/ar/articles/spatialopsd-self-distilling-spatial-intelligence-from-verified-coding-agent-traces.