FP-Loc: Lightweight and Drift-free Floor Plan-assisted LiDAR Localization (2203.00292v1)

Abstract: We present a novel framework for floor plan-based, full six degree-of-freedom LiDAR localization. Our approach relies on robust ceiling and ground plane detection, which solves part of the pose and supports the segmentation of vertical structure elements such as walls and pillars. Our core contribution is a novel nearest neighbour data structure for an efficient look-up of nearest vertical structure elements from the floor plan. The registration is realized as a pair-wise regularized windowed pose graph optimization. Highly efficient, accurate and drift-free long-term localization is demonstrated on multiple scenes.

• The paper introduces the Approximate Nearest Neighbour Field (ANNF) to efficiently match floor plan elements, reducing search complexity to O(d).
• The paper presents a drift-free LiDAR localization algorithm that combines single-frame registration with windowed pose graph optimization for robust six DoF tracking.
• The paper demonstrates real-time performance on standard hardware, achieving 5x speed and outperforming conventional SLAM methods in challenging indoor environments.

Overview of FP-Loc: Lightweight and Drift-free Floor Plan-assisted LiDAR Localization

The paper "FP-Loc: Lightweight and Drift-free Floor Plan-assisted LiDAR Localization" by Ling Gao and Laurent Kneip introduces an innovative approach to six degree-of-freedom (DoF) LiDAR localization within indoor environments. This work leverages prior information in the form of floor plans, which are often readily available, to achieve precise localization. The paper builds upon the notion that floor plans are a cost-effective alternative to high-density 3D point clouds regularly used in mapping and localization tasks.

Key Contributions

The authors present several significant contributions:

1. Approximate Nearest Neighbour Field (ANNF): The paper introduces a new data structure called the ANNF to efficiently match the nearest geometric elements (e.g., wall segments) in a floor plan. This innovation reduces computational overhead while maintaining a high success rate for finding nearest neighbors. The ANNF is showcased to have reduced search complexity to $\mathcal{O}(d)$, where $d$ is the maximum depth of ANNF, enhancing performance efficiency compared to traditional dense matching methods.
2. Drift-free LiDAR Localization Algorithm: By using this novel data structure, the authors design a LiDAR-based localization framework that ensures drift-free six DoF tracking performance. The system effectively aligns LiDAR measurements with the floor plan geometry using a robust pose graph optimization strategy, which includes single-frame registration and windowed optimization techniques.
3. Real-time Performance: The proposed method is demonstrated to achieve substantial efficiency, running at five times real-time speed on standard laptop hardware. This level of performance indicates the potential applicability of FP-Loc in embedded systems, providing a practical advantage for deployment in resource-constrained environments.

Implementation and Evaluation

The authors provide rigorous experimental evaluation to corroborate their method's efficacy. The experiments were conducted in environments with minimal texture, such as unfurnished rooms and long corridors, which are typically challenging for traditional SLAM algorithms due to self-similar features causing drift. FP-Loc is compared against traditional SLAM solutions like LeGO-LOAM, demonstrating significant improvements in absolute trajectory accuracy and computational efficiency. Specifically, the method not only provides robust localization in loop-intensive environments but maintains drift-free results compared to other SLAM frameworks under similar conditions.

Implications and Future Directions

The framework presented in this paper capitalizes on the availability of architectural floor plans, which are often underutilized in standard localization strategies. By doing so, FP-Loc offers a feasible alternative to tackle challenges in indoor scenarios, particularly when GPS signals are unavailable or unreliable. The modular design of the system, illustrated through the use of pose graph optimization, presents an adaptable approach that could further benefit from integration with other sensor modalities or machine learning techniques to handle dynamic and cluttered environments more effectively.

In terms of future developments, the authors propose focusing on refining object detection and clutter removal during LiDAR scans, which remains a crucial area in extending the applicability of their work. Additionally, utilizing artificial intelligence to augment the floor plan data with semantic comprehension could further enhance FP-Loc's robustness in complex indoor environments.

In conclusion, FP-Loc offers a compelling framework for precise and efficient indoor LiDAR localization, paving the way for future explorations to bridge the gap between architectural insights and autonomous navigation systems.