RaNDT SLAM: Radar SLAM Based on Intensity-Augmented Normal Distributions Transform

TL;DR

RaNDT SLAM tackles radar-based SLAM in vision-denied environments by augmenting the Normal Distributions Transform (NDT) with radar intensity, enabling robust scan-to-submap registration. The approach fuses a motion model, IMU yaw, and NDT registration within a sliding window, supports loop closures via ScanContext, and optimizes a pose graph for global consistency. Key contributions include intensity-augmented NDT, a submap-based representation, adaptive robust loss with multi-scale optimization, and a publicly released dataset (HM23) and code, with validation on indoor/outdoor scenes and the Oxford Radar RobotCar dataset. The work demonstrates real-time performance and competitive accuracy, advancing radar SLAM applicability for rescue robotics and multi-environment operation.

Abstract

Rescue robotics sets high requirements to perception algorithms due to the unstructured and potentially vision-denied environments. Pivoting Frequency-Modulated Continuous Wave radars are an emerging sensing modality for SLAM in this kind of environment. However, the complex noise characteristics of radar SLAM makes, particularly indoor, applications computationally demanding and slow. In this work, we introduce a novel radar SLAM framework, RaNDT SLAM, that operates fast and generates accurate robot trajectories. The method is based on the Normal Distributions Transform augmented by radar intensity measures. Motion estimation is based on fusion of motion model, IMU data, and registration of the intensity-augmented Normal Distributions Transform. We evaluate RaNDT SLAM in a new benchmark dataset and the Oxford Radar RobotCar dataset. The new dataset contains indoor and outdoor environments besides multiple sensing modalities (LiDAR, radar, and IMU).

Figures (6)

• Figure 1: Architecture of the RaNDT SLAM framework and robot used for evaluation in Section \ref{['ssec:randt_slam_data']}. The robot is equipped with an indurad iSDR-c pivoting radar, a phidgets IMU, and a pair of Sick TiM LiDARs.
• Figure 2: Effect of the filter on the radar scan. Intensities are denoted red/low to purple/high; radome is circular entity in \ref{['fig:unfiltered']}.
• Figure 3: Factor graph optimizing over $l=3$ states.
• Figure 4: RaNDT SLAM scenarios (green: LiDAR trajectory, blue: potential loop closures); maps generated using LiDAR.
• Figure 5: Estimated trajectories in different environments, using specific and mixed parameter configurations.