IMPRINT: Image-Conditioned Query Enrichment for Long-Tail Object Goal Navigation
IMPRINT:用于长尾目标导航的图像条件查询增强
机构 * University of Padova(帕多瓦大学) ; Fondazione Bruno Kessler (FBK)(布鲁诺·凯斯勒基金会)
专题命中 具身导航 :navigation(title,abstract);embodied AI(abstract);分类 cs.CV
AI总结 研究针对具身人工智能中ObjectNav依赖纯文本查询可靠性低的问题,提出IMPRINT框架,用网络图像丰富文本查询改善定位,无需训练导航策略。通过新基准HSSD-rare评估,显示图像条件查询可提升导航增益,还指出下游检测质量是关键瓶颈。
Comments Accepted at the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Project page: https://github.com/JelinR/IMPRINT