FAST-LIO Loop Closure Test on NUANCE

This was our final project for the EECE 5554 (Robotics Sensing and Navigation) course at Northeastern University.

Data Collection

We collected data using the NUANCE autonomous car of Northeastern University. We utilized the Ouster 3D Lidar (OS1-64), the VectorNav IMU (VN-100), and the GPS sensors of the car.

The Ouster 3D Lidar (OS1-64) & The VectorNav IMU (VN-100) & The NUANCE autonomous car

We drove around Boston for about 10 minutes to collect the data. While collecting the data, we ensured that we had some loop closures in our path so that we could use them later. The exact path followed by us is provided below.

This is the link to our dataset - https://drive.google.com/file/d/1fpwFDyM8jiKTvpuVPF6OtJXaFg9kRt9O/view?usp=sharing

Aim of the project

The aim of our project was to compare the data we recorded on the FAST-LIO and LIO-SAM algorithms and compare those algorithms on the basis of loop-closure.

We started off by running those algorithms on some existing datasets. After that, we tried to run them on our own dataset. We generated good results for FAST-LIO but couldn't run it on LIO-SAM due to some challenges that have been highlighted in detail in our project presentation.

Although the results obtained from FAST-LIO were good, there was a small issue. There was a misalignment in our return trajectory and the trajectory when we commenced our journey. This was due to two reasons:

The length of our path was huge.

FAST-LIO didn't have loop-closure incorporated into it.

Hence, we made use of FAST-LIO LC. It is an open-source repository that has loop-closure incorporated in it. The results we obtained using FAST-LIO LC (LC stands for Loop Closure), although not perfect, were much better as compared to FAST-LIO.

Results

For the Results and problems that we faced, you can check out our presentaion PDF named "Project_Presentation.pdf" in the repository.

Also, here is the link to the 3D point cloud scan file we generated after using FAST-LIO algorithm on the data we collected - https://drive.google.com/file/d/135FwKW1pHvMg_6CV8-F2XXNinwPFRoaE/view?usp=sharing

The Custom_Scripts Folder

Included in this repository is a Custom_Scripts folder used to plot the the gps data from NUANCE onto a satelite map image covering the area data was collected. We have provided a brief description of what each script's function purpose is:

gps_to_csv.py

This script runs through the bag file and extracts the gps data into a csv file (output.csv) for further processing.

gps_kml.py

This script creates a kml file (gps_points.kml) file that can be uploaded to https://kmlviewer.nsspot.net that will automatically show all points onto Google Maps.

gps_plot.py

This script manually plots all of the gps points ontop of a satelite map image in longitude and latitude.

gps_plot_utm.py

This script manually plots all of the gps points ontop of a satelite map image in UTM coordinates, scaled local to the image.

References

• FAST-LIO GitHub Repository
• FAST-LIO Paper
• LIO-SAM GitHub Repository
• LIO-SAM Paper
• FAST-LIO LC GitHub Repository
• Lidar_IMU_Init GitHub Repository
• Ouster 3D Lidar (OS1-64)
• VectorNav (VN-100)

Acknowledgements

We would like to specially thank our course instructor Prof. Hanumant Singh as well as our course teaching assistants Vishnu Rohit Annadanam and Jasen Levoy who helped us immensely not only while doing this project but throught the course. They were great at teaching, managing the course, and were always available and enthusiastic to solve our doubts.

We would also like to thank Northeastern University for providing us with the NUANCE autonomous car so that we could collect our data.

Contributors

• Aryaman Shardul
• Jaykumar Goswami
• Jake Ross
• Rongxuan Zhou

Contact

• Aryaman Shardul - [email protected]
• Jaykumar Goswami - [email protected]
• Jake Ross -
• Rongxuan Zhou - [email protected]