中文

SIR: Structured Image Representations for Explainable Robot Learning

机器人学 2026-06-29 v1 计算机视觉与模式识别

摘要

Existing robot policies based on learned visual embeddings lack explicit structure and are sensitive to visual distractions. Thus, the representations that drive their behaviour are often opaque, making their decision-making process difficult to interpret. To address this, we introduce Structured Image Representations (SIR), a method that leverages Scene Graphs (SGs) as an intermediate representation for robot policy learning. Our approach first constructs a fully connected graph, using image-derived features as initial node representations. Then, a module learns to sparsify this graph end-to-end, creating a task-relevant sub-graph that is passed to the action generation model. This process makes our model intrinsically explainable. Evaluations on RoboCasa show that our sparse graph policies outperform image-based baselines on average with 19.5% vs 14.81% success rate. Most importantly, we show that the learned sparse graphs are a powerful tool for model analysis. By analysing when the model's sub-graph deviates from human expectation, such as by including distractor nodes or omitting key objects, we successfully uncover dataset biases, including spurious correlations and positional biases. https://github.com/intuitive-robots/SIR_Model

引用

@article{arxiv.2606.30101,
  title  = {SIR: Structured Image Representations for Explainable Robot Learning},
  author = {Paul Mattes and Jan Schwab and Jens Bosch and Nils Blank and Maximilian Xiling Li and Minh-Trung Tang and Moritz Haberland and Rudolf Lioutikov},
  journal= {arXiv preprint arXiv:2606.30101},
  year   = {2026}
}

备注

Published at CVPR 2026