CAD2SLAM: Adaptive Projection Between CAD Blueprints and SLAM Maps

Martín Bayón-Gutiérrez, Natalia Prieto-Fernández, María Teresa García-Ordás,  José Alberto Benítez-Andrades, Héctor Alaiz-Moretón, Giorgio Grisetti
IEEE Robotics and Automation Letters, Volume: 10, 2025
DOI: 10.1109/LRA.2024.3522838 

Coming Soon

Open-source ROS2 implementation will be available soon

Installation

• Install the SRRG Suite in your catkin workspace

• Clone the project: cd  ~/workspace/src/ & git clone https://github.com/martin-bayon/CAD2SLAM

• Compile the project: cd ~/workspace/ & catkin build

Run the example

• Prepare your terminal: source ~/workspace/devel/setup.bash & cd ~/workspace/src/cad2slam/config

• Run CAD2SLAM projector: rosrun cad2slam launch_projector demo.conf 

Prepare your own environment

• Prepare your terminal: source ~/workspace/devel/setup.bash & cd ~/workspace/src/cad2slam/config 

• Place your .png and .yaml files for the CAD and SLAM maps on ~/workspace/src/cad2slam/config/data/maps/

• Clone the demo config file: cp demo.conf my.conf

• Modify your config file: confviz -c my.conf 

• Create the correspondences set: rosrun cad2slam launch_correspondence_finder my.conf 

• Prepare your environment: ./launcher.srrg my.conf

• Run CAD2SLAM projector: rosrun cad2slam launch_projector my.conf

Replicate our results:

• Prepare your terminal: source ~/workspace/devel/setup.bash & cd ~/workspace/src/cad2slam/config 

• Run the experiments: rosrun cad2slam launch_experiment_1 <environment.conf>

Each environment is composed of 4 files

• 2 .png files for the CAD and SLAM maps of the environments

• 2 .yaml files for the CAD and SLAM maps containing meta information about the png files

4 test environments

1. Demo environment, sintetically generated

2. DIAG (Dipartimento di Ingegneria informatica, automatica e gestionale) 2nd floor, Sapienza Università di Roma, Italy 

3. EIII-0 (Escuela de Ingeniería Industrial, Informática y Aeroespacial) ground floor, Universidad de León, Spain

4. EIII-3 (Escuela de Ingeniería Industrial, Informática y Aeroespacial) 3rd floor, Universidad de León, Spain

5. MIC (Módulo de investigación cibernética) 1st floor, Universidad de León, Spain

Computer-Aided Design (CAD) Maps

Robot-generated (SLAM) Maps

Path planning on CAD maps

Since neither the CAD blueprint nor the SLAM map can be considered as ground truth, we have developed an experiment in which a robot trajectory on the CAD blueprint is projected onto the SLAM map and sinthetic LiDAR measurements in both representations of the environment are compared in a point-to-point fashion.

The results allows to indirectly evaluate the mapping accuracy given that, under perfect conditions, the observations from both maps should be identical.

The steps of the experiment are as follow:

1. Given start and goal positions in the CAD map, each step of the robot trajectory is projected onto the SLAM map using CAD2SLAM

2. For each step of the trajectory, a simulated LiDAR pointcloud is generated, both in the CAD and SLAM maps

3. The pointcloud for the CAD map is projected onto the SLAM map using CAD2SLAM

4. The resulting pointclouds are compared in a point-to-point fashion

5. The mean square error for each LiDAR reading is used to colorize the following plots 

From left to right:

• Path on the CAD/SLAM map (Green)

• Number of outliers in each step of the path

• Point-to-point value 

CAD2SLAM_dataset.zip

The dataset, is compressed in a zip file with the following structure:

├── correspondences

│   ├── demo_correspondences.data

│   ├── diag_correspondences.data

│   ├── eiii-0_correspondences.data

│   ├── eiii-3_correspondences.data

│   └── mic_correspondences.data

│

└── maps

    ├── demo_cad.png

    ├── demo_cad.yaml

    ├── demo_slam.png

    ├── demo_slam.yaml

    ├── diag_cad.png

    ├── diag_cad.yaml

    ├── diag_slam.png

    ├── diag_slam.yaml

    ├── eiii-0_cad.png

    ├── eiii-0_cad.yaml

    ├── eiii-0_slam.png

    ├── eiii-0_slam.yaml

    ├── eiii-3_cad.png

    ├── eiii-3_cad.yaml

    ├── eiii-3_slam.png

    ├── eiii-3_slam.yaml

    ├── mic_cad.png

    ├── mic_cad.yaml

    ├── mic_slam.png

    └── mic_slam.yaml

Where:

• The correspondences folder contains the correspondence points pairs for the CAD blueprint and the SLAM map

• The maps folder contains the .png image of the environment and its correspoding .yaml metadata file

CAD2SLAM_experiment_A.zip

The experiments results are included in a zip file with the following structure:

├── demo_001

│   ├── cad2slam_clouds

│   ├── cad_clouds

│   ├── info.txt

│   ├── results

│   └── slam_clouds

│

├── demo_002

│   ├── cad2slam_clouds

│   ├── cad_clouds

│   ├── info.txt

│   ├── results

│   └── slam_clouds

│

└── env_XXX

     ├── cad2slam_clouds

     ├── cad_clouds

     ├── info.txt

     ├── results

     └── slam_clouds

Where:

• Each experiment is stored in an individual folder containing the name of the environment and the experiment_id 

  • cad2slam_clouds folder contains the CAD blueprint LiDAR readings after been projected onto the SLAM map

  • cad_clouds folder contains the CAD blueprint LiDAR readings

  • results folder contain the experiments results in txt format, as well as the plots of the experiment

  • slam_clouds folder contains the CAD blueprint LiDAR readings after been projected onto the SLAM map