Researchers at Tongji University have developed a LiDAR SLAM back-end constraint construction method that uses an adaptive adjustment mechanism to reduce false loop-closure matches in complex environments. The approach targets scenarios where weak geometric differences and degraded point-cloud features cause mismatches, leading to increased errors in graph optimization and reduced mapping accuracy.
Key takeaways
- Dynamic search window adjustment: An improved Floyd algorithm maintains a shortest-distance matrix among submaps, quantifying matching uncertainty between nodes. This enables real-time adaptation of the search domain during constraint construction.
- Adaptive score threshold: The method quantifies point-cloud specificity from candidate solution scores, dynamically adjusting the scan-matching score threshold to prevent false constraints when features are degraded.
- Improved efficiency: The modified Floyd algorithm reduces time complexity from O(n³) to O(n²), ensuring real-time performance.
- Experimental validation: Physical tests on a mobile robot platform show significant reduction in incorrect constraints and enhanced mapping accuracy and localization robustness compared to mainstream graph-based LiDAR SLAM algorithms.
The method is designed to be generic and applicable to other SLAM modalities, including visual SLAM. By adaptively tuning search parameters, it strengthens the reliability of back-end constraints and improves overall optimization, offering a practical solution for mobile robots operating in industrial and everyday environments.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2026-01-13 · “基于自适应调节机制的激光SLAM后端约束构建方法”