LiDAR Loop Closure Detection using Semantic Graphs with Graph Attention Networks

2025-01-11

J. Intell. Robotic Syst.

AI Summary

To address the insufficient robustness of loop closure detection in LiDAR SLAM under cross-view and dynamic scenarios, this paper proposes a semantic graph–based loop closure detection method leveraging Graph Attention Networks (GATs). We introduce object-level semantic graph modeling into LiDAR loop closure detection for the first time: PointPillars is employed to obtain semantic segmentation outputs, from which an explicit scene semantic graph is constructed to encode spatial structure and inter-object relationships. A GAT is then used to aggregate multi-hop semantic neighborhood information, learning interpretable node embeddings; loop closure is determined via graph-level similarity matching. Evaluated on the KITTI and Oxford RobotCar datasets, the method achieves a 12.3% improvement in mean Average Precision (mAP) and reduces false positive rate by 37%, significantly enhancing long-term localization robustness.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

Problem

• Lidar Data
• Loop Closure
• Robotics

Innovation

• Neural Network
• Feature Extraction
• Lidar Data Processing