Pose-Graph Attentional Graph Neural Network for Lidar Place Recognition
This addresses place recognition for robotics in challenging environments, but it is incremental as it builds on existing encoders and graph neural network concepts.
The paper tackles place recognition in lidar data by proposing P-GAT, a graph neural network that compares sequential and non-sequential sub-graphs instead of frame-to-frame retrieval, showing effectiveness in feature-poor scenes and domain adaptation with performance gains over state-of-the-art methods.
This paper proposes a pose-graph attentional graph neural network, called P-GAT, which compares (key)nodes between sequential and non-sequential sub-graphs for place recognition tasks as opposed to a common frame-to-frame retrieval problem formulation currently implemented in SOTA place recognition methods. P-GAT uses the maximum spatial and temporal information between neighbour cloud descriptors -- generated by an existing encoder -- utilising the concept of pose-graph SLAM. Leveraging intra- and inter-attention and graph neural network, P-GAT relates point clouds captured in nearby locations in Euclidean space and their embeddings in feature space. Experimental results on the large-scale publically available datasets demonstrate the effectiveness of our approach in scenes lacking distinct features and when training and testing environments have different distributions (domain adaptation). Further, an exhaustive comparison with the state-of-the-art shows improvements in performance gains. Code is available at https://github.com/csiro-robotics/P-GAT.