FAST-LIO2: Fast Direct LiDAR-inertial Odometry

FAST-LIO2 is computationally-efficient (e.g., up to 100 Hz odometry and mapping in large outdoor environments), robust (e.g., reliable pose estimation in cluttered indoor environments with rotation up to 1000 deg/s), versatile (i.e., applicable to both multi-line spinning and solid-state LiDARs, UAV and handheld platforms, and Intel and ARM-based processors), while still achieving higher or comparable accuracy with existing methods.
Compared with FAST-LIO1, we added many new features:
1. Incremental mapping using ikd-Tree, achieve faster speed and over 100Hz frame rate.
2. Direct odometry on Raw LiDAR points (no feature extraction), achieving better accuracy.
3. Due to the lack of feature extraction, FAST-LIO2 can support different LiDAR Types including spinning (Velodyne, Ouster) and solid-state (Livox Avia, horizon).
4. Support LiDAR built-in or external IMUs.
5. Support ARM-based platforms including Khadas VIM3, Nivida TX2, Raspberry 4B with 8G RAM.
FAST-LIO2: github.com/hku...
ikd-Tree: github.com/hku...
Paper: github.com/hku...

[@sjoythehawk]

Excellent, the result is so accurate even without loop-closing.

[@zhaoweiguo2884]

哈哈！已经安装完毕，唯一遗憾的是目前无法得到Avia，只能用Horizon实验啦！期待了三个月，终于等到，非常感谢！

[@zyzxxw]

谢谢，有什么问题随时告诉我

[@trollenz]

Fantastic ! Results are outstanding ! Are you gonna implement a loop closure algorithm ?

[@rexinsightrobotics]

Please try to use the algorithm for lidar scans in a tunnel in Hong Kong. e.g. Eagle's nest tunnel + Shatin Heights Tunnel + Tai Wai tunnel with 60km/h + in 1 go. This had been a challenge for all famous SLAM algorithms. Many algorithms work well in feature rich urban area but when they enter tunnels with speed, with repeating patterns and flat walls, they all lost track.

[@DevelopESPrtk]

Good scan result !

[@jasonzou8508]

屌爆了

[@francescoperasso]

Fastlio2 is available for installation on your github also on jetson tx2 or only on Intel Manifold?

[@zyzxxw]

installation on arm board is same as Intel board

[@francescoperasso]

hello I have been using Fastlio for a few days with the Avia sensor I just bought, before I used MID40 with low value external Imu. Analyzing with Cloud Compare, the cloud created in the PCD folder with the Avia sensor, I noticed, that it only has two returns. My question is: Fastlio takes advantage of the triple return of the sensor especially if you map in areas with many trees or vegetation?
Is there a system or parameter that activates the triple return function? However even if the Avia sensor has a lower FOV and an internal IMU synchronized with the data, the cloud in the presence of vegetation distorts, would a quality IMU with gnss RTK and PPS synchronism improve the result? Thanks.