Researchers have developed a new framework to assess the reliability of visual predicates used in understanding robotic manipulation. This framework evaluates how well predicates like contact, support, and grasp perform under various degradation conditions such as blur, occlusion, and frame dropping. Experiments on several datasets demonstrated that while static predicates are relatively robust, dynamic and derived predicates are more susceptible to errors, significantly impacting downstream manipulation understanding accuracy. AI
影响 Provides a diagnostic layer for improving robotic manipulation understanding by identifying weaknesses in visual predicate recognition under degraded conditions.
排序理由 This is a research paper detailing a new framework and experimental results. [lever_c_demoted from research: ic=1 ai=1.0]
AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →