The success of the metaverse depends on technological advancements as well as human-centered designs that care for human users in fully immersive virtual environments. Without substantial spatial knowledge of the virtual environment, users can feel lost or disoriented in the metaverse. This can cause distress and poor overall user experience. This article introduces an unstuck feature that aims at guiding users out of irreconcilable situations. The feature delivers persuasive user navigation using automated avatars which were driven by machine learning models. A user experiment with mobility and eye-gaze tracking has been conducted to evaluate the effectiveness of the unstuck feature.