SPARO in ICRA 2024
Salience-guided Ground Factor for Robust Localization of Delivery Robots in Complex Urban Environments
Abstract: In urban environments for delivery robots, particularly in areas such as campuses and towns, many custom features defy standard road semantic categorizations. Addressing this challenge, our paper introduces a method leveraging Salient Object Detection (SOD) to extract these unique features, employing them as pivotal factors for enhanced robot loop closure and localization. Traditional geometric feature-based localization is hampered by fluctuating illumination and appearance changes. Our preference for SOD over semantic segmentation sidesteps the intricacies of classifying a myriad of non-standardized urban features. To achieve consistent ground features, the Motion Compensate IPM (MC-IPM) technique is implemented, capitalizing on a motion for distortion compensation and subsequently selecting the most pertinent salient ground features through moment computations. For thorough evaluation, we validated saliency detection and localization performances in real urban scenarios.
Session: Localization 1 / Tuesday, 10:30-12:00 / Paper TuAT25-NT.1
Mesh-based Photorealistic and Real-time 3D Mapping for Robust Visual Perception of Autonomous Underwater Vehicle
Abstract: This paper proposes a photorealistic real-time dense 3D mapping system that utilizes a learning-based image enhancement method and mesh-based map representation. Due to the characteristics of the underwater environment, where problems such as hazing and low contrast occur, it is hard to apply conventional simultaneous localization and mapping (SLAM) methods. Furthermore, for sensitive tasks like inspecting cracks, photorealistic mapping is very important. However, the behavior of Autonomous Underwater Vehicle (AUV) is computationally constrained. In this paper, we utilize a neural network-based image enhancement method to improve pose estimation and mapping quality and apply a sliding window-based mesh expansion method to enable lightweight, fast, and photorealistic mapping. To validate our results, we utilize real-world and indoor synthetic datasets. We performed qualitative validation with the real-world dataset and quantitative validation by modeling images from the indoor synthetic dataset as underwater scenes.
Session: Workshop on Field Robotics (May 13 (Monday), Full-day, Conference Center 302)
ReFeree: Radar-based efficient global descriptor using a Feature and Free space for Place Recognition
Abstract: Radar is highlighted for robust sensing capabilities in adverse weather conditions (e.g. dense fog, heavy rain, or snowfall). In addition, Radar can cover wide areas and penetrate small particles. Despite these advantages, Radar-based place recognition remains in the early stages compared to other sensors due to its unique characteristics such as low resolution, and significant noise. In this paper, we propose a Radar-based place recognition utilizing a descriptor called ReFeree using a feature and free space. Unlike traditional methods, we overwhelmingly summarize the Radar image (i.e. 361.5KB -> 528B). Despite being lightweight, it contains semi-metric information and is also outstanding from the perspective of place recognition performance. For concrete validation, we test a single session from the MulRan dataset and a multi-session from the Oxford Offroad Radar, Oxford Radar RobotCar, and the Boreas dataset. The supplementary materials of our place recognition and SLAM are available at https://github.com/sparolab/Referee.
Session: Workshop on Radar in Robotics / May 17 Friday, Full-day / ??
DiTer++: Diverse Terrain and Multi-modal Datasetfor Multi-Robot Navigation in Multi-session Outdoor Environments
Abstract: Collaborative of multiple field robots are necessary for navigation and mapping of large-scale environments. While traversing, traversability estimation considering each robot's nature is essential for keeping the robot safe and ensuring its performance. Even in a structured environment, driving without considering terrain information can lead to serious damage to the platform such as slipping due to steep slope or falling caused by sudden height change. To address this challenge, we present DiTer++, multi-robot, multi-session, and multi-modal dataset, including ground-level information. The dataset is obtained with a forward-facing RGB camera and ground-facing RGB-D camera, a thermal camera, two types of LiDARs, IMU, GPS, and robot motion sensors.
Session: Workshop on Future of Construction
Monday / Conference Center 301
Uni-Mapper: Unified Mapping Framework for Multi-modal LiDARs in Complex and Dynamic Environments
Abstract: In construction sites, which are large-scale and complex urban environments, it is essential to merge multiple maps obtained from various platforms. However, sensor-modality and dynamic environments remain challenging problems for unified mapping. To address this issue, we present Uni-Mapper, a dynamic-aware 3D point cloud map merging framework for multi-modal LiDAR systems. Our scene descriptor rejects dynamic objects in real-time and is robust to LiDAR modality based on local triangle features. To ensure consistent mapping performance, we adopt centralized pose graph optimization with a two-step registration process. We thoroughly evaluate the superiority of the proposed framework using two datasets: HeLiPR (multi-modal) and INHA (multi-modal, multi-robot).
Session: Workshop on Future of Construction - Best Research Award 3rd prize
Monday, 10:45AM - 11:15AM / Conference Center 301
StaticNeRF: Neural Implicit Static Mapping and Localization in Dynamic Environments
Abstract: Recently, neural implicit representations have been widely introduced for robot mapping to achieve high-resolution maps. Previous approaches perform well in stable, and static environments but encounter difficulties when faced with the challenges posed by moving objects. In this paper, we propose a entire pipeline for neural implicit mapping and robust filter-based localization in dynamic environments. The entire scene can be decomposed into static and transient fields by implicitly learning geometric information, without the need for any external data. Moreover, this separation facilitates robust localization in dynamic environments by integrating a localization pipeline specifically tailored to the static field. Our approach is validated against standard and custom datasets, demonstrating that our implicit neural map has better performance than the other neural rendering methods and that our pipeline is effective in dynamic object removal and accurate in localization, marking a step forward for efficient navigation systems.
Session: Workshop on RoboNeRF Spotlight
Friday, Conference Center 419
12:00PM - 12:30PM (Oral Session)
13:30PM - 14:30PM (Poster Session)
Vision-based UAV geo-localization using satellite images in GNSS-Denied environments
Abstract: Robust and accurate localization is an essential technology for autonomous flight of unmanned aerial vehicles (UAVs). Many current UAV localization methodologies rely on GNSS signals. However, GNSS signals can be easily blocked in urban environments and can be neutralized by jamming and spoofing. Therefore, geo-localization through matching UAV images and satellite images has been actively studied recently, but the difficulty exists due to the large gap between the two images. Therefore, this paper presents a geo-localization methodology that performs matching between nadir-facing UAV images and satellite images using a foundation model in a GNSS-denied environment and estimates the global position of the UAV. To verify the effectiveness of the geo-localization methodology presented in this paper, we conducted experiments on a real aerial dataset.
Session: Late breaking Results Poster VII / Thursday, 10:30-12:00 / Paper ThAL-EX.4 / Exhibition Hall
