MEGA Hub

SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces

Authors

Do you know Rongxue Li?You can claim authorship or link another user.Do you know Meng Yang?You can claim authorship or link another user.Do you know Yiru Mao?You can claim authorship or link another user.Do you know Yongliang Tao?You can claim authorship or link another user.Do you know Lulu Hu?You can claim authorship or link another user.Do you know Bin Yang?You can claim authorship or link another user.Do you know Zhao Xu?You can claim authorship or link another user.Do you know Weihua Luo?You can claim authorship or link another user.Do you know Bowen Xu?You can claim authorship or link another user.

Abstract

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.

Community

00