In 2021, the challenge run for the first time. In total, 27 teams participated in the challenge, with 7 commercial companies among them. The challenge was hosted by the IROS workshop.
We congratulate the winners Megvii, Bosch CR and V&R for their outstanding performance.
| Team | Score | ||
|---|---|---|---|
| 1 | Megvii IROS Talk | 461 | |
| 2 | Bosch CR – Advanced Autonomous Systems IROS Talk | 457 | |
| 3 | V&R Vision & Robotics GmbH | 406 | |
| 4 | GeoSLAM | 389 | |
| 5 | VILENS and SLAM, Oxford Robotics Institute 7000 USD | 378 | |
| 6 | VIRAL SLAM, Nanyang Technological University, Singapore 2000 USD | 346 | |
| 7 | CMU Doom | 333 | |
| 8 | Maplab + OKVIS, ETHZ 1000 USD | 288 | |
| 9 | NPM3D Team, MINES ParisTech | 273 | |
| 10 | UC San Diego | 258 | |
| 11 | Cristian F Rubio and Harold F Murica, Universidad de Ibague, Colombia | 228 | |
| 12 | IVISO | 219 | |
| 13 | Spectacular AI (real-time, without lidar) | 216 | |
| 14 | Undisclosed | 216 | |
| 15 | Undisclosed | 147 | |
| 16 | Autonomous Mobile Machines group, Tampere University, Finland | 132 | |
| 17 | Undisclosed | 18 | |
| 18 | Undisclosed | 6 |
The first four places have been taken by commercial algorithms, which all focused on LiDAR-IMU odometry, showing the maturity and robustness of these approaches.
The best team, Megvii, used a variant of FAST-LIO2 and achieved an average error of 9.3 cm on all sequences. Megvii was one of the few teams that merged the Ouster and the Livox LiDAR data, which, together with using all LiDAR points for state estimation, gave them a significant advantage.
The best algorithm that fuses vision with LiDAR and imu ranked 5th, VILENS by the Oxford Robotics Institute. The best vision-only solution ranked 12th, with the majority of errors larger than 50 cm.