Answers are grounded in this paper and cite their sources.
Purpose This study aims to improve localization accuracy and robustness for autonomous docking of reconfigurable ground unmanned systems in complex industrial environments, where vision-based localization is sensitive to illumination variations, partial occlusions and observation noise. Design/methodology/approach A multi-sensor fusion localization framework integrating visual and laser range measurements is proposed. A CNN-assisted EKF-based PnP pose estimation framework is developed to enhance motion state estimation under degraded visual conditions. To improve short-range docking accuracy, a laser-based multi-point ranging strategy is introduced. Furthermore, a grey relational analysis–guided Dempster–Shafer evidence fusion framework is developed to handle conflicting sensor information and improve fusion consistency. Findings Experimental results in complex industrial environments demonstrate that the proposed fusion method achieves stable and accurate pose estimation for autonomous docking. Compared with vision-only localization, Kalman filter-based fusion, and weighted averaging fusion schemes, the proposed method exhibits superior robustness under illumination variations and partial occlusions. The system achieves a real-time update rate of 20 Hz, satisfying closed-loop docking control requirements, and provides adaptive down-weighting of unreliable sensor measurements to ensure docking accuracy within a 1° error threshold. Originality/value This study presents an application-oriented multi-sensor localization framework that combines learning-based visual estimation, laser geometric measurements and conflict-aware evidence fusion. The proposed method provides a practical and reliable solution for autonomous docking of reconfigurable unmanned vehicles in industrial environments.
No takes yet. Share an insight, caveat, or question.
Zhan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: