Demo data
See more videos at Video Gallery.
Mapping with various range sensors
Tip
The same parameter set was used for all the sensors.
Warning
As can be seen in the video, the quality of point clouds of stereo-based sensors (D455 and ZED2i) is not very good, and GLIM, which is based on point cloud matching, does not always work well with these sensors. We recommend using other vision-based SLAM packages for stereo sensors.
Flat wall experiment
Outdoor driving test with Livox MID360
Indoor mapping with Azure Kinect
Real-time mapping on Jetson Nano
- os1_128_01_downsampled.bag (515MB)
- config_nano_cpu.zip
- config_nano_gpu.zip
Note
- Only odometry estimation was performed, no global optimization.
- Visualization was run on another PC that received points and pose messages via ethernet.
(rviz took about a half of Jetson Nano's computation capability without rendering anything!!) - (2024/07/04) The current version of GLIM does not support CUDA 11 and older. Some minor midifications are expected to be necessary.