cristianrubioa / hilti-challenge

Hilti-challenge_details

Introduction

In the framework of the International Conference on Intelligent Robots and Intelligent Systems 2021, a call for participation in the HILTI-challenge was launched by Prof. Davide Scaramuzza and Giovanni Cioffi from the University of Zürich with the collaboration of Danwei Wang, Christian Laugier, Philippe Martinet and Yufeng Yue. In response, the research group of the University of Ibagué joins this event by means of this document. In response, the research group of the University of Ibagué joins this event by means of this document.

Estimated Trajectories

The 12 trajectories obtained were generated according to the requested formats. Their files can be obtained through the following link: https://www.hilti-challenge.com/dataset.html.

Overview

The challenge was addressed based on the LOAM method and the A-LOAM github repository, in which the trajectory is recovered using a batch optimisation method that processes segmented data sets with added constraints. By decomposing the original problem, a simpler problem such as online motion estimation is solved first. Then, the mapping is performed as a batch optimization (similar to iterative closest point (ICP) methods) to produce high-precision motion estimates and maps. The method used does not report the use of future or predicted information, packet adjustment or loop closure. A-LOAM is an advanced implementation of LOAM (J. Zhang and S. Singh. LOAM: Lidar Odometry and Mapping in Real-time), which uses Eigen and Ceres Solver to simplify the code structure. The code has been modified from the original version to a cleaner and simpler version, without complicated mathematical derivations and redundant operations. Referred to by the authors as a good learning material for SLAM beginners.

 Used Sensors

In this implementation only the LiDAR sensor data was used by subscribing to the topic /os_cloud_node/points.

Execution Environment

Used Hardware

Command	Description
OS	Versión: Ubuntu 18.04.5 LTS - Kernel: Linux 4.15.0-137-generic
CPU	Intel(R) Xeon(R) Gold 6152 CPU @ 2.10GHz 4 Dual-core
RAM	kB: 65947352 kB - GB: 65,95 GB
ROS	Distro: Melodic - Versión: 1.14.10

Evaluation Metric

The integrity and precision of the estimated trajectories were calculated from the evaluation system of the same challenge hiltislamchallenge/evaluation-evo. Where all trajectories have to be provided in the TUM format cristianrubioa/extract_trajectory.

#timestamp tx ty tz qx qy qz qw
1.403636580013555527e+09 0.0 0.0 0.0 0.0 0.0 0.0 0.0

• timestamp (float) gives the number of seconds.
• tx ty tz (3 floats) give the position, where t is the translation vector.
• qx qy qz qw (4 floats) give the orientation, where q is a quaternion describing the rotation.

Note: with each value separated by a space. Names of the provided reference files should not be changed as they indicate which frame the reference is in.

Simulation Results

Sequences	Name_trayectory	Poses	Path_length [m]	Time [s]	ATE
RPG Drone Testing Arena	uzh_tracking_area_run2.txt	893	78.151	89.299	0.492
IC Office	IC_Office_1.txt	2004	171.163	200.301	-
Office Mitte	Office_Mitte_1.txt	2640	214.265	263.913	-
Parking Deck	Parking_1.txt	5823	556.871	582.212	-
Basement	Basement_1.txt	1129	74.981	112.803	0.297
Basement 3	Basement_3.txt	3306	118.793	330.616	-
Basement 4	Basement_4.txt	3503	113.898	350.224	0.237
Lab	LAB_Survey_2.txt	1356	48.089	135.513	0.116
Construction Site Outdoor 1	Construction_Site_1.txt	1995	106.880	199.420
Construction Site Outdoor 2	Construction_Site_2.txt	3991	218.970	399.117	1.596
Campus 1	Campus_1.txt	4298	225.502	429.730	-
Campus 2	Campus_2.txt	3747	194.664	374.725	0.625

Used Parameters

1. Modify the scanRegistration.cpp file.

scanID = int((angle + 45) / 1.428 + 0.2);

// angle: angle of vertical elevation.  
// 45: The vertical filter (lower vertical angle).
// 1.428: the interval between each scan. [vFOV/(num_chanel-1)]
// 0.2: for rounding (keep the integer part).

2. In cristianrubioa/extract_trajectory rename the topic to '/aft_mapped_to_init' for A-LOAM method.
3. In Launch file, rename the name of the topic to the one corresponding to the name of the topic that publishes the LiDAR data (in this case /os_cloud_node/points).

Related Package

• A-LOAM
• laboshinl/LOAM
• daobilige-su/LOAM
• hiltislamchallenge
• extract_trajectory

Acknowledgement

• [1] A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? the KITTIvision benchmark suite,” inProceedings of the IEEE Computer Society Conference onComputer Vision and Pattern Recognition, 2012.
• [2] J. Zhang and S. Singh, “LOAM: Lidar Odometry and Mapping in Real-time", Robotics: Science and Systems , vol. 2, 2014.
• [3] . Zhang, M. Kaess, and S. Singh, “On degeneracy of optimization-based state estimationproblems,” inProceedings - IEEE International Conference on Robotics and Automation,vol. 2016-June, 2016.
• [4] J. Zhang and S. Singh, “Low-drift and real-time lidar odometry and mapping,”Au-tonomous Robots, vol. 41, no. 2, 2017.