LeGO-LOAM-FN: An Improved Simultaneous Localization and Mapping Method Fusing LeGO-LOAM, Faster_GICP and NDT in Complex Orchard Environments

Round 1

Reviewer 1 Report

To address the challenge of cumulative errors during map building in complex orchard environments, this study introduces a loopback registration algorithm that combines Faster Generalized Iterative Closest Point (Faster_GICP) and Normal Distributions Transform (NDT). The paper touches on interesting application-oriented topics but has some areas that require clarification:

1. The abstract mentions complex orchard environments, but the specific challenges in such environments are not clearly outlined. The study motives need to be enriched to better highlight the significance of this topic.

2. Section 1 lacks a comprehensive literature review. A new section focusing on mobile robots, intelligent algorithms, etc., should be incorporated. Additionally, when citing references, it's preferable to use family names only.

3. Table 1 should include more similar algorithms to emphasize the key features of the proposed algorithm in comparison.

4. The algorithm flow for LeGO-LOAM-FN needs to be presented in a standard format, accompanied by a thorough analysis, including time complexity and computational cost.

5. The figures presented should use clearer images and have larger caption sizes. Additionally, consider adding an abbreviation table for clarity.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article titled "LeGO-LOAM-FN: An Improved Simultaneous Localization and Mapping Method Fusing LeGO-LOAM, Faster_GICP, and NDT in Complex Orchard Environments" presents a substantial contribution to the field. This study targets the challenge of cumulative errors encountered by mapping robots operating within intricate orchard environments. The proposed solution introduces a novel loopback registration algorithm that amalgamates Faster Generalized Iterative Closest Point (Faster_GICP) and Normal Distributions Transform (NDT). Initially, the algorithm leverages a K-Dimensional tree (KD-Tree) structure to effectively eliminate dynamic obstacle point clouds. Subsequently, it employs a two-step point filter to reduce the number of feature points within the current frame used for matching and optimizes data for precision enhancement. Additionally, the algorithm calculates the matching degree of probability distributions by meshing the point cloud and refines precision registration using the Hessian Matrix Method.

This refined approach significantly improves map building accuracy by minimizing the impact of dynamic objects present within orchard environments. In a pivotal experiment within scene 2, the LeGO-LOAM algorithm fails to recognize the loopback when the robot revisits the starting point, causing positional drift due to cumulative error accumulation. However, the LeGO-LOAM-FN algorithm implemented in this study adeptly rectifies the positional loopback, successfully closing the trajectory loop. Evaluation of the algorithm's effectiveness utilizes GNSS latitude and longitude acquisition as standard trajectory information, showcasing superior performance in terms of Absolute Trajectory Error (ATE), Root Mean Square Error (RMSE), and Standard Deviation (STD) when compared to the LeGO-LOAM algorithm across various scenes and datasets (KITTI 00, Scene 1, Scene 2, Scene 3).

This study addresses the intricate challenges associated with Simultaneous Localization and Mapping (SLAM) within complex orchard environments, specifically focusing on scenarios where long loopback scenes often lead to closure failures. The proposed LeGO-LOAM-FN algorithm, incorporating KD-Tree representation, systematically eliminates dynamic objects and strategically partitions loop closure detection into discernible steps. Leveraging Faster_GICP and small-grid NDT techniques significantly augments precision in registration and demonstrates superior performance in maneuvering across expansive agricultural terrains. The algorithm remarkably diminishes cumulative pose estimation errors, surpassing the baseline LeGO-LOAM method by 67% and 73% in root mean square error and standard deviation, respectively. These results underscore the algorithm's suitability for highly accurate point cloud mapping within complex orchard environments, showcasing its robustness and effectiveness in real-world applications.

The introductory section lacks explicit citations or references to authors actively involved in researching SLAM methods based on Smooth Variable Structure Filter (SVSF), specifically those employing the Smooth Variable Structure Filter-based SLAM. Incorporating such citations would not only provide contextual relevance but also acknowledge prior seminal work in this field, significantly fortifying the article's foundation.

The Variable Structure Filter, a novel estimator utilizing a sliding mode concept, employs switching gains to converge estimates toward true state values. Over the years, considerable attention has been directed towards sliding-mode control based on variable structure systems, owing to its implementation simplicity and robustness against system model uncertainties and noise. Operating on sliding mode control and estimation techniques, the SVSF embodies a predictor-corrector estimator. Initially estimating states using system models and inputs, termed as a priori estimates, it subsequently incorporates correction terms based on measurement errors, resulting in posteriori state estimates. By employing a switching gain, the SVSF ensures convergence of estimated state trajectories within a boundary around true state values, exhibiting stability and resilience against modeling uncertainties and noise. The SVSF's versatility extends to solving control or estimation problems involving discrete state and observation models, catering to both linear and nonlinear systems

This article stands out for its well-defined structure and clear, concise writing, which greatly aids in comprehending the concepts and methodologies employed. Its focus on addressing challenges encountered in robotic mapping within complex orchard environments is particularly noteworthy. By zeroing in on cumulative errors stemming from long and recurrent loops, the article identifies a critical issue in the field of SLAM. The systematic presentation of the proposed solution, LeGO-LOAM-FN, which ingeniously merges multiple algorithms and integrates a KD-Tree representation, showcases an innovative approach to overcome these specific challenges.

Moreover, the article excels in the rigor of its results analysis. The meticulous evaluation of the proposed algorithm spans across different scenarios, employing standard trajectory data (GNSS) to assess its performance. Clear comparisons between the LeGO-LOAM-FN algorithm and established methods like LeGO-LOAM are presented, highlighting significant improvements in terms of root mean square error and standard deviation. These results bolster the article's credibility and underscore the effectiveness of the LeGO-LOAM-FN algorithm in reducing pose estimation errors and accurately creating point cloud maps within complex environments.

Additionally, the article provides a promising outlook for the future by outlining remaining challenges and potential avenues for improvement. Further exploration of the adaptability of the proposed algorithm to different agricultural environments or varied contexts, along with deeper performance metrics in more extreme situations, could be intriguing. This approach paves the way for future research to refine and extend the applicability of this innovative methodology. In summary, this article not only presents significant findings but also opens intriguing avenues for ongoing evolution and enhancement of robotic mapping systems in complex environments.

The quality of English language in the article is commendable. The writing is clear, coherent, and effectively conveys complex technical concepts in a precise manner. The structure of the article is well-defined, aiding in the easy comprehension of the proposed methodology and its implications. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Regarding the revisions requested, I am pleased to inform you that the authors have diligently addressed all the suggested modifications. The article has been meticulously corrected by the given feedback, ensuring accuracy and clarity in its content.

The article is well structured, although some grammatical and spelling errors require rectification to elevate its overall quality. These corrections are pivotal to ensure a seamless understanding of the research and its implications among readers.

Upon careful review of the revised manuscript, I recommend that the article be accepted in its present form for publication. The authors have promptly responded to the revisions, enhancing the quality and coherence of the paper. Their efforts have significantly improved the overall manuscript, aligning it with the standards set by Sensors Journal.

The article is well structured, although some grammatical and spelling errors require rectification to elevate its overall quality. These corrections are pivotal to ensure a seamless understanding of the research and its implications among readers.