MapEval: Towards Unified, Robust and Efficient SLAM Map Evaluation Framework

Xiangcheng Hu¹ · Jin Wu¹ · Mingkai Jia¹ · Hongyu Yan¹ · Yi Jiang² · Binqian Jiang¹
Wei Zhang¹ · Wei He³ · Ping Tan^1*†

¹HKUST     ²CityU     ³USTB

†Project Lead   *Corresponding Author

Overview

MapEval is a comprehensive framework for evaluating point cloud maps in SLAM systems, addressing two fundamentally distinct aspects of map quality assessment:

1. Global Geometric Accuracy: Measures the absolute geometric fidelity of the reconstructed map compared to ground truth. This aspect is crucial as SLAM systems often accumulate drift over long trajectories, leading to global deformation.

2. Local Structural Consistency: Evaluates the preservation of local geometric features and structural relationships, which is essential for tasks like obstacle avoidance and local planning, even when global accuracy may be compromised.

These complementary aspects require different evaluation approaches, as global drift may exist despite excellent local reconstruction, or conversely, good global alignment might mask local inconsistencies. Our framework provides a unified solution through both traditional metrics and novel evaluation methods based on optimal transport theory.

News

• 2025/07/06: Use TBB to accelerate MME calculation, update parameter settings and add more configuration examples.
• 2025/05/05: Add new test data and remove simulation codes.
• 2025/03/05: Formally published in IEEE RAL!
• 2025/02/25: Paper accepted!
• 2025/02/12: Source code released!
• 2025/02/05: Paper resubmitted.
• 2024/12/19: Paper submitted to IEEE RAL!

Key Features

Traditional Metrics Implementation

• Accuracy (AC): Point-level geometric error assessment
• Completeness (COM): Map coverage evaluation
• Chamfer Distance (CD): Bidirectional point cloud difference
• Mean Map Entropy (MME): Information-theoretic local consistency metric

Novel Proposed Metrics

• Average Wasserstein Distance (AWD): Robust global geometric accuracy assessment
• Spatial Consistency Score (SCS): Enhanced local consistency evaluation

Experimental Results

Simulation Experiments

Noise Sensitivity	Outlier Robustness

Real-World Experiments

Map Evaluation via Localization Accuracy	Map Evaluation in Diverse Environments

Efficiency and Parameter Analysis

Datasets

The following datasets are supported and used for evaluation:

Dataset	Description
MS-Dataset	Multi-session mapping dataset
FusionPortable (FP) and FusionPortableV2	Multi-sensor fusion dataset
Newer College (NC)	Outdoor autonomous navigation dataset
GEODE Dataset (GE)	Degenerate SLAM dataset

Quick Start

Dependencies

• Open3D 0.15.1 (>= 0.11)
• Eigen 3.3.7
• PCL 1.10.0
• yaml-cpp
• TBB 2020.1
• Ubuntu 20.04

Test Data

Download the test data using password: 1

Sequence	Preview	Test PCD	Ground Truth PCD
MCR_slow		map.pcd	map_gt.pcd
PK01		map.pcd	gt.pcd

Installation and Usage

1. Install Open3D

Note: A higher version of CMake may be required.

git clone https://github.com/isl-org/Open3D.git
cd Open3D && mkdir build && cd build   
cmake ..
make install

2. Configure Parameters

Set and review the parameters in config.yaml:

# accuracy_level: vector5d, we mainly use the result of the first element
# For small inlier ratios, try larger values, e.g., for outdoors: [0.5, 0.3, 0.2, 0.1, 0.05]
accuracy_level: [0.2, 0.1, 0.08, 0.05, 0.01]

# initial_matrix: vector16d, the initial transformation matrix for registration
# Ensure correct format to avoid YAML::BadSubscript errors
initial_matrix:
  - [1.0, 0.0, 0.0, 0.0]
  - [0.0, 1.0, 0.0, 0.0]
  - [0.0, 0.0, 1.0, 0.0]
  - [0.0, 0.0, 0.0, 1.0]
  
# vmd_voxel_size: outdoor: 2.0-4.0; indoor: 2.0-3.0
vmd_voxel_size: 3.0

3. Compile MapEval

git clone https://github.com/JokerJohn/Cloud_Map_Evaluation.git
cd Cloud_Map_Evaluation/map_eval && mkdir build && cd build
cmake ..
make

4. Run Evaluation

./map_eval

This evaluates a point cloud map generated by a SLAM system against a ground truth point cloud map and calculates related metrics.

Visualization

Error Visualization

The framework generates rendered distance-error maps with color coding:

• Raw distance-error map (10cm): Shows error for all points
• Inlier distance-error map (2cm): Shows error for matched points only
• Color scheme: R→G→B represents distance error levels from 0-10cm

Evaluation Without Ground Truth

If ground truth is not available, only Mean Map Entropy (MME) can be evaluated. Lower values indicate better consistency. Set evaluate_mme: false in config.yaml.

Mesh Reconstruction

A simple mesh can be reconstructed from the point cloud map:

5. Output Files

The evaluation generates the following result files:

6. Voxel Error Visualization

For detailed voxel error visualization, use error-visualization.py:

pip install numpy matplotlib scipy
python3 error-visualization.py

Frequently Asked Questions

How to Obtain Initial Pose?

Use CloudCompare to align the LiDAR-Inertial Odometry (LIO) map to the ground truth map:

1. Roughly translate and rotate the LIO point cloud map to align with the GT map
2. Manually register the moved LIO map (aligned) to the GT map (reference)
3. Extract the terminal transform output T as the initial pose matrix

Difference Between Raw and Inlier Rendered Maps

• Raw rendered map (left): Color-codes error for all points in the estimated map. Points without correspondences in the ground truth map are assigned maximum error (20cm) and rendered in red.

• Inlier rendered map (right): Excludes non-overlapping regions and colors only inlier points after point cloud matching. Contains only a subset of the original estimated map points.

Credit: John-Henawy in issue #5

Applicable Scenarios

1. With Ground Truth Map

All metrics (AC, COM, CD, MME, AWD, SCS) are applicable.

2. Without Ground Truth Map

Only Mean Map Entropy (MME) can be used for evaluation.

Important considerations:

• Maps must be on the same scale
  • Cannot compare LIO maps with LIO SLAM maps that have undergone loop closure optimization. Loop closure modifies local point cloud structure, leading to inaccurate MME evaluation.
  • Can compare MME between different LIO maps

Credit: @Silentbarber, ZOUYIyi in issue #4 and issue #7

Citation

If you find this work useful for your research, please cite our paper:

@article{hu2025mapeval,
  title={MapEval: Towards Unified, Robust and Efficient SLAM Map Evaluation Framework}, 
  author={Xiangcheng Hu and Jin Wu and Mingkai Jia and Hongyu Yan and Yi Jiang and Binqian Jiang and Wei Zhang and Wei He and Ping Tan},
  journal={IEEE Robotics and Automation Letters},
  year={2025},
  volume={10},
  number={5},
  pages={4228-4235},
  doi={10.1109/LRA.2025.3548441}
}

@article{wei2024fpv2,
  title={Fusionportablev2: A unified multi-sensor dataset for generalized slam across diverse platforms and scalable environments},
  author={Wei, Hexiang and Jiao, Jianhao and Hu, Xiangcheng and Yu, Jingwen and Xie, Xupeng and Wu, Jin and Zhu, Yilong and Liu, Yuxuan and Wang, Lujia and Liu, Ming},
  journal={The International Journal of Robotics Research},
  pages={02783649241303525},
  year={2024},
  publisher={SAGE Publications Sage UK: London, England}
}

Related Works

The following research works have utilized MapEval for map evaluation:

Work	Description	Publication	Metrics Used	Preview
LEMON-Mapping	Multi-Session Point Cloud Mapping	arXiv 2025	MME
CompSLAM	Multi-Modal Localization and Mapping	arXiv 2025	AWD/SCS
GEODE	SLAM Dataset	IJRR 2025	-
ELite	LiDAR-based Lifelong Mapping	ICRA 2025	AC/CD
PALoc	Prior-Assisted 6-DoF Localization	T-MECH 2024	AC/CD
MS-Mapping	Multi-Session LiDAR Mapping	arXiv 2024	AC/CD/MME
FusionPortableV2	SLAM Dataset	IJRR 2024	COM/CD

Contributors

We thank all contributors to this project:

Star History

License

This project is licensed under the MIT License - see the LICENSE file for details.