Organizations: State Key Laboratory of Ocean Sensing, Zhejiang University, China · Interdisciplinary Student Training Platform for Marine areas, Zhejiang University, Hangzhou, China · Department of Electronic and Computer Engineering, The Hong Kong University of Science and Technology, Hong Kong · School of AI and Robotics, Hunan University, China · Ocean College, Zhejiang University, China · Institute of Deep-Sea Science and Engineering, Chinese Academy of Sciences, China
Simultaneous localization and mapping using radar sensors has gained increasing attention due to radar's inherent robustness to adverse weather and lighting conditions. However, radar measurements are characteristically sparse and noisy compared to LiDAR and visual data, posing significant challenges in achieving dense, continuous, and consistent map representations. In this paper, we present RICH-SLAM, a radar SLAM framework designed to address these challenges. Our approach features a Rao-Blackwellized particle filter-based back end that employs particle filtering for pose estimation and Kalman filtering for map updates. We propose an incremental Hilbert-space reduced-rank Gaussian process mapping strategy that enables continuous and uncertainty-aware map representations given sparse radar inputs. We further introduce a posterior-aware particle weighting scheme that leverages the full posterior distribution of map parameters for more robust likelihood evaluation. Experiments on self-collected and public ColoRadar datasets show that RICH-SLAM constructs continuous occupancy maps from sparse radar measurements and supports uncertainty-aware planning for mobile robots.