Comparison between Various SLAM Algorithms¶
LOAM: LIDAR Odometry and Mapping in Real-time¶
Paper Link: here
GitHub Repository Link: There is no direct implementation available online but an advanced version can be found at HKUST-Aerial-Robotics/A-LOAM
Sensor Suite¶
- IMU
- LIDAR
Cons¶
- Suffers from the issues of ghosting based on LIDAR Scan Error resulting in reduced dimensional accuracy.
- Pitch and Roll rotations tends to break the Mapping.
- Close Loop Algorithm (non-existent) suffers an issue with distortion in low feature areas such as hallways. Results in incorrect geometry.
Pros¶
- Real-Time implementation due to small sensor suite and higher approximations for reduced sampling.
LIOM: Tightly Coupled 3D LIDAR Inertial Odometry and Mapping¶
Paper Link: here
GitHub Repository Link: hyye/lio-mapping
Website: here
Sensor Suite¶
- IMU
- LIDAR
Cons¶
- Improved dimensional accuracy than LOAM but still contains significant ghosting.
- Improved Distortion in narrow hallways but still suffers from mapping error for long hallways which are featureless.
LIO-SAM: Tightly-coupled LIDAR Inertial Odometry via Smoothing and Mapping¶
Paper Link: here
GitHub Repository Link: TixiaoShan/LIO-SAM
Sensor Suite¶
- IMU
- LIDAR
- GPS
Cons¶
- Still suffers from issue with featureless areas which uses GPS for correction. This results in failure in GPS constraint or noisy readings.
VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator¶
Paper Link: here
GitHub Repository Link: HKUST-Aerial-Robotics/VINS-Mono
Sensor Suite¶
- IMU
- 1 x Monocular Camera
Cons¶
- Since this algorithm depends on feature-based SLAM, it fails for long hallways or open areas which lacks features.
- High Paced Motion results in distortion and Mapping Failure due to limitation of the processor to find features in images.
- Dependence on camera requires heavy calibration for intrinsic and extrinsic parameters of sensors.
EMV-LIO: An Efficient Multiple Vision aided LiDAR-Inertial Odometry¶
Paper Link: here
GitHub Repository Link: BingqiShen/EMV-LIO
Sensor Suite¶
- IMU
- Cameras
- LIDAR
Cons¶
- Open-Source Codebase suffers from high number of Magic Numbers for transforms and values. Needs heavy modification to be used for your application.
- Algorithm Fails for Big Maps due to ROS Node not being able to handle data higher than 1 GB on a node.
- Built using the combination of VINS-Mono and LIO-SAM which has a bad intertwined Transform Tree. Modifying the code requires a lot of effort.