Abstract:Robust Simultaneous Localization and Mapping (SLAM) is a crucial enabler for autonomous navigation in natural, unstructured environments such as parks and gardens. However, these environments present unique challenges for SLAM due to frequent seasonal changes, varying light conditions, and dense vegetation. These factors often degrade the performance of visual SLAM algorithms originally developed for structured urban environments. To address this gap, we present ROVER, a comprehensive benchmark dataset tailored for evaluating visual SLAM algorithms under diverse environmental conditions and spatial configurations. We captured the dataset with a robotic platform equipped with monocular, stereo, and RGB-D cameras, as well as inertial sensors. It covers 39 recordings across five outdoor locations, collected through all seasons and various lighting scenarios, i.e., day, dusk, and night with and without external lighting. With this novel dataset, we evaluate several traditional and deep learning-based SLAM methods and study their performance in diverse challenging conditions. The results demonstrate that while stereo-inertial and RGB-D configurations generally perform better under favorable lighting and moderate vegetation, most SLAM systems perform poorly in low-light and high-vegetation scenarios, particularly during summer and autumn. Our analysis highlights the need for improved adaptability in visual SLAM algorithms for outdoor applications, as current systems struggle with dynamic environmental factors affecting scale, feature extraction, and trajectory consistency. This dataset provides a solid foundation for advancing visual SLAM research in real-world, natural environments, fostering the development of more resilient SLAM systems for long-term outdoor localization and mapping. The dataset and the code of the benchmark are available under <a class="link-external link-https" href="https://iis-esslingen.github.io/rover" rel="external noopener nofollow">this https URL</a>.

EnvoDat: A Large-Scale Multisensory Dataset for Robotic Spatial Awareness and Semantic Reasoning in Heterogeneous Environments

RoboSense: Large-scale Dataset and Benchmark for Egocentric Robot Perception and Navigation in Crowded and Unstructured Environments

OmniHD-Scenes: A Next-Generation Multimodal Dataset for Autonomous Driving

CoPeD-Advancing Multi-Robot Collaborative Perception: A Comprehensive Dataset in Real-World Environments

Towards Robust Robot 3D Perception in Urban Environments: The UT Campus Object Dataset

DIDLM:A Comprehensive Multi-Sensor Dataset with Infrared Cameras, Depth Cameras, LiDAR, and 4D Millimeter-Wave Radar in Challenging Scenarios for 3D Mapping

ROVER: A Multi-Season Dataset for Visual SLAM

MCD: Diverse Large-Scale Multi-Campus Dataset for Robot Perception

OpenMPD: An Open Multimodal Perception Dataset for Autonomous Driving

WildScenes: A benchmark for 2D and 3D semantic segmentation in large-scale natural environments

MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations

EmbodiedScan: A Holistic Multi-Modal 3D Perception Suite Towards Embodied AI

FusionPortable: A Multi-Sensor Campus-Scene Dataset for Evaluation of Localization and Mapping Accuracy on Diverse Platforms

nuScenes: A Multimodal Dataset for Autonomous Driving

S3E: A Large-scale Multimodal Dataset for Collaborative SLAM

The ADUULM-360 Dataset -- A Multi-Modal Dataset for Depth Estimation in Adverse Weather

Scalability in Perception for Autonomous Driving: Waymo Open Dataset.

INSANE: Cross-Domain UAV Data Sets with Increased Number of Sensors for developing Advanced and Novel Estimators

Zenseact Open Dataset: A large-scale and diverse multimodal dataset for autonomous driving

The INSANE dataset: Large number of sensors for challenging UAV flights in Mars analog, outdoor, and out-/indoor transition scenarios