提出安全写入机制，解决传感器模糊与定位漂移导致的流场地图伪影问题。
Write-Safe Flow Field Mapping under Ambiguous Onboard Sensing and Localization Drift
- 基于地图参考的保守融合框架，动态评估写入安全性。
- 合成环境中鬼影污染减少42%，真实硬件测试再降39%。
- 适合在定位不稳、感知模糊的复杂流场中部署机器人。
移动机器人可从机载传感器推断局部流场结构，但局部合理估计未必安全写入全局地图。相似流场结构可能产生模糊观测，而定位漂移会导致预测块被错误写入位置，重复错误更新会累积成持续存在的鬼影结构。本文提出一种地图参考感知的保守融合框架，模型同时预测局部速度块和学习得到的写入安全评分，持续抑制不确定的地图更新，同时在无可靠地图参考时仍可初始化。在合成喷流与横流环境中的实验显示，该方法相较无门控融合平均减少42%的鬼影污染；使用真实推进器尾迹的压力与光流数据进行零样本硬件回放，进一步降低39%鬼影污染，同时保持81%的地图覆盖率。结果表明，在感知模糊与定位漂移条件下，安全地图写入对流场建图至关重要。
原文摘要 · Abstract (English)
Mobile robots can infer local flow structure from onboard sensing, but a locally plausible estimate is not always safe to write into a global map. Similar flow structures may produce ambiguous observations, while localization drift causes predicted patches to be written at incorrect locations. Repeated misregistered updates then accumulate into persistent ghost structures. We address this failure mode with a map-reference-aware conservative fusion framework. The model predicts a local velocity patch and a learned write-safety score that continuously attenuates uncertain map updates while permitting initialization when no reliable map reference is available. Across synthetic jet and crossflow environments, the proposed method reduces average ghost contamination by 42% relative to ungated fusion. A zero-shot hardware replay using real pressure and optical-flow measurements from a thruster wake further reduces ghost contamination by 39% while retaining 81% map coverage. These results show that safe map writing is critical for flow mapping under ambiguous sensing and localization drift.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。