Abstract
Reliable and long-term positioning is a fundamental requirement for deploying robots across various real-world scenes. Although multi-source fusion has been widely used to enhance positioning robustness, most existing methods are designed for specific scenes and lack generalizability. This reveals that the limitations of current approaches do not stem from a lack of technical tools, but rather from the robot's inability to adaptively select suitable positioning strategies under different spatiotemporal conditions. Inspired by human thinking process-such as information filtering, memory reliability evaluation, and knowledge-based reasoning when faced with multiple potential positioning cues-this paper systematically rethinks the design of multi-source fusion positioning. We propose a novel framework with self-assessment capabilities, aiming to make real-time decisions on the optimal source combination scheme to provide continuous and robust positioning results. Experimental results further verify the robustness of the proposed method under environmental uncertainty, demonstrating its potential in unifying the source selection process in multi-source fusion positioning.