sejong01.md
July 23, 2024 ยท View on GitHub
MulRan: Sejong01
Experiment summary:
- Robot: car
- Sensor: Ouster OS1-64
- Odometry: KISS-ICP
- Trajectory length: 23 km
- Duration: 47 min
- Mean speed: 12.4 m/s
- Max speed: 20.3 m/s
- Results below comptued on: AMD Ryzen 7 PRO 4750U
Graphs' legend from top to bottom:
- timing (better: higher)
- compression (better: higher)
- per-point information in the optimization (better: higher)
- # of ICP iterations till convergence (better: lower)
Pure KISS-ICP
Results: Unstable with a lot of redundancy in the data.
KISS-ICP preceded by RMS
Results: Low and stable timing+compression, which effectively lowers the # of ICP iterations (registration time lowered 12 times on average!). High level of information preserved.
APE and RPE comparison
Graphs:
- blue: KISS-ICP
- orange: KISS-ICP preceded by RMS
Results: High drift reduction by RMS.
APE
| Sampling method | RMSE (m) | mean (m) | median (m) | max (m) |
|---|---|---|---|---|
| Voxelization | 7100 | 6511 | 7815 | 9581 |
| RMS | 2230 | 1953 | 2352 | 3292 |
RPE
| Sampling method | RMSE (m) | mean (m) | median (m) | max (m) |
|---|---|---|---|---|
| Voxelization | 17.9 | 4.3 | 2.7 | 1311.6 |
| RMS | 3.1 | 2.9 | 2.6 | 21.6 |
Example sampling of a single frame
