基於不確定性之地圖更新與再查找

Abstract

在無需訓練的機器人應用中，以視覺語言模型（VLMs）探索構建的預探索語義地圖作為基礎元素已被證明非常有效。然而，現有的方法假設地圖是準確的，並且沒有提供有效的機制來根據錯誤的地圖修正決策。這篇論文引入了不確定概念，並通過不確定性來估計地圖的準確性和品質，使得機器人可以找到更有可能出錯的區域，並在沒有額外標籤的情況下修正由不準確地圖引起的錯誤決策。 我們使用兩個現代地圖的基石模型，VLMaps 和 OpenMask3D，展示了我們所提出方法的有效性，並在這兩者上都顯示了改善。

Pre-Explored Semantic Map, constructed through prior exploration using visual lan guage models (VLMs), has proven effective as a foundational element for training-free robotic applications. However, existing approaches assume the maps accuracy and do not provide effective mechanisms for revising decisions based on incorrect maps. This work introduces Uncertainty-Aware Memory Updating and Re-Proposing, which estimates the accuracy and quality of the map through uncertainty, enabling the agent to locate the areas with higher chance of being incorrect and to revise erroneous decisions stemming from inaccurate maps without additional labels. We demonstrate the effectiveness of our proposed method using two modern map backbones, VLMaps and OpenMask3D, and show the improvements on both of them.