面向动态环境的机器人鲁棒长期定位

To address the challenge of positioning failure or accuracy degradation in robotic autonomous inspection systems within dynamic environments, a robust localization framework was proposed.The framework consisted of three core modules, which were robust real-time localization, real-time dynamic point cloud filtering, and high-precision loop closure detection with relocation. During the initial mapping phase, the real-time pose estimated by the simultaneous localization and mapping (SLAM) module was fused with the outputs of the dynamic filtering module to construct a static global map.For the relocation requirement, a point cloud feature descriptor matching method and a pose calculation method were introduced. To furthermore enhance the precision and robustness, a point cloud clustering algorithm was adopted to optimize the constraints of matched point pairs.Experimental results on public datasets show that the proposed method effectively improves the localization accuracy of robots in dynamic settings, and the framework is verified to possess strong adaptability and practical value, which could provide reliable positioning support for robotic autonomous inspection tasks.