Winner

HILTI SLAM Challenge (LiDAR Single-Session)

ICRA 2023 2023
The handheld LiDAR rig on a construction site (left, photo: HILTI) and our map of a site-2 building under construction (right).
The handheld LiDAR rig on a construction site (left, photo: HILTI) and our map of a site-2 building under construction (right).

Overview

Platform
Handheld rig and tracked robot
Sensors
Hesai 32 (handheld), Robosense (robot)
Environment
Construction sites, one underground
Metric
Score by position error, max 1600
Multi-session maps: site 2 before and after alignment (left), four site-3 sessions (right).
Multi-session maps: site 2 before and after alignment (left), four site-3 sessions (right).

My Part

Faster-LIO diverges in a cramped space; AdaLIO does not.
Faster-LIO diverges in a cramped space; AdaLIO does not.

AdaLIO on every sequence

Our adaptive odometry changes its parameters when a scan is degenerate. I tuned and ran it per sequence.

Multi-session alignment

Quatro for a global guess, G-ICP to refine, then a multi-session pose graph.

Award

LiDAR single-session first-place certificate, HILTI SLAM Challenge 2023.
LiDAR single-session first-place certificate, HILTI SLAM Challenge 2023.
The team at KAIST EE with the certificates.
The team at KAIST EE with the certificates.