Towards versatile and generalized SLAM for robotics : the FusionPortable benchmark

Abstract

Simultaneous Localization and Mapping (SLAM) technology has been widely applied in various robotic scenarios, from rescue operations to autonomous driving. However, the generalization of SLAM algorithms remains a significant challenge, as current datasets often lack diversity in terms of platforms and environments. To address this limitation, we present the FusionPortable benchmark, a comprehensive multi-sensor SLAM dataset featuring a diverse range of sensor modalities, varied motion patterns, and a wide spectrum of environmental scenarios, along with a unified evaluation framework and well-defined metrics for assessing localization accuracy, mapping quality, and depth estimation performance. Our dataset comprises 27 sequences, spanning over 2.5 hours and collected from four distinct platforms: a handheld multi-sensor suite, a legged robot, an unmanned ground vehicle (UGV), and a high-speed vehicle. These sequences cover diverse settings, including indoor spaces, buildings, campuses, and urban areas, with a total trajectory length of 38.7 km. Additionally, the dataset includes ground-truth (GT) trajectories and RGB point cloud maps covering approximately 0.3 km². To validate the utility of our dataset in advancing SLAM research, we evaluate the performance of several state-of-the-art (SOTA) SLAM algorithms on the dataset. Furthermore, we demonstrate the dataset’s broad applicability beyond traditional SLAM tasks by investigating its potential for monocular depth estimation. The complete dataset, including raw sensor data, GT trajectories and maps, calibration parameters, and development tools, is publicly available at https://fusionportable.github.io/dataset/fusionportable_v2.

Date of Award	2024
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Shaojie SHEN (Supervisor)