To address the low efficiency and poor robustness of robot plug-in charging in unstructured environments, this paper proposes a method for plug-in autonomous charging via a wheeled mobile manipulator in unstructured environments. First, a two-stage paradigm target recognition and positioning algorithm named You Only Look Once version 8 (YOLOv8) and Point Pair Features (PPF) pose estimation is constructed based on YOLOv8 and PPF. By integrating 2D image detection and 3D point cloud processing, it can quickly and accurately recognize and locate charging guns in complex environments. Then, according to the application characteristics of unstructured scenarios, an integrated system of the wheeled mobile manipulator is designed and verified through simulation. Finally, based on the custom-developed wheeled mobile manipulator platform, plug-in autonomous charging experiments are carried out in unstructured scenarios. The results show that in three typical experimental scenarios, namely normal illumination without occlusion, dim light with slight occlusion, and dark with large-area occlusion, the method proposed in this paper can achieve rapid recognition and positioning of the charging gun, meeting the precision requirements that the angle error of the charging gun pose estimation does not exceed \(10^{\circ }\) and that the distance error does not exceed 15mm; this demonstrates the effectiveness and robustness of the designed plug-in autonomous charging method for the wheeled mobile manipulator.
Continual LiDAR place recognition requires a descriptor model to learn new environments while retaining retrieval accuracy in previously learned environments under limited access to historical data. Replaying historical retrieval relations and distilling descriptors impose different optimization demands that fixed loss coefficients cannot readily accommodate. We propose adaptive gradient-balanced boundary memory (AGBBM), a framework that operates on global descriptors without changing the backbone architecture. Boundary-aware triplet memory retains complete anchor–positive–negative relations and their historical distance statistics under an environment-balanced triplet budget. Reliability-guided decoupled distillation separates supervised replay, one-sided positive-distance preservation, and reliability-weighted historical-boundary preservation. A two-level optimization mechanism normalizes active historical losses using exponential moving averages and calibrates same-view descriptor distillation using the ratio of current-learning and distillation gradient norms at a shared descriptor parameter, with configurable attenuation under gradient conflict. We evaluate AGBBM in the stage-wise sequence Oxford → MulRan DCC → MulRan Riverside → In-house using PointNetVLAD, MinkLoc3D, LoGG3D-Net with max pooling, and CASSPR. The final mean Recall@1 values are 61.42%, 79.30%, 66.12%, and 82.07%, respectively, with signed average forgetting of 1.83, −0.17, 1.03, and 0.52 percentage points. The results support historical retrieval retention across the evaluated backbones and four-stage sequence. The evaluation addresses sequential descriptor learning rather than online obstacle detection or explicit map-change identification.