Radarize: Enhancing Radar SLAM with Generalizable Doppler-Based Odometry

TL;DR

Radarize tackles indoor SLAM using a single-chip mmWave radar by exploiting radar-native cues. It introduces Doppler-azimuth heatmaps to estimate translation and range-azimuth heatmaps to estimate rotation, augmented with elevation-aware beamforming and multipath suppression, and fuses these into a Cartographer backend for real-time maps and trajectories. The work presents a CNN-based translation and rotation estimation pipeline, a UNet-based mapper with echo suppression, and a public radar SLAM dataset collected over diverse buildings and platforms, achieving up to ~5x odometry and ~8x end-to-end SLAM improvements over state-of-the-art radar methods. The results demonstrate robust, real-time radar-only SLAM on commodity hardware with good cross-environment generalization, enabling privacy-preserving indoor mapping for wearable, handheld, and mobile robotic platforms.

Abstract

Millimeter-wave (mmWave) radar is increasingly being considered as an alternative to optical sensors for robotic primitives like simultaneous localization and mapping (SLAM). While mmWave radar overcomes some limitations of optical sensors, such as occlusions, poor lighting conditions, and privacy concerns, it also faces unique challenges, such as missed obstacles due to specular reflections or fake objects due to multipath. To address these challenges, we propose Radarize, a self-contained SLAM pipeline that uses only a commodity single-chip mmWave radar. Our radar-native approach uses techniques such as Doppler shift-based odometry and multipath artifact suppression to improve performance. We evaluate our method on a large dataset of 146 trajectories spanning 4 buildings and mounted on 3 different platforms, totaling approximately 4.7 Km of travel distance. Our results show that our method outperforms state-of-the-art radar and radar-inertial approaches by approximately 5x in terms of odometry and 8x in terms of end-to-end SLAM, as measured by absolute trajectory error (ATE), without the need for additional sensors such as IMUs or wheel encoders.

Figures (15)

• Figure 1: Key advantages of radar over conventional SLAM sensors. Radarize leverages both environmental sensing and motion sensing capabilites of radar to perform SLAM.
• Figure 2: Common sources of artifacts in indoor environments. Each image of a scene is paired with a top-down radar heatmap superimposed on a depth camera point cloud. (a) When moving down hallways, featureless flat walls present ambiguities during scan matching. (b) Multipath reflections can cause objects to appear behind surfaces. (c) Limited elevation resolution on most sensors induces artifacts from floors/ceilings.
• Figure 3: The range-azimuth heatmap captures the reflections along different distances and angles. The heatmap is mapped to a cartesian plane for easy visualization. The intensity of reflection varies from blue (low) to red (high). Left. Dense heatmap overlaid on depth camera point cloud. Right. Same heatmap overlaid on sparse radar point cloud. We trace the shadow behind each point for visibility.
• Figure 4: Radarize Overview.Left. Radar frames are processed into heatmaps. Top. The mapping module converts range-azimuth heatmaps into range scans. Bottom. The tracking module (a) regresses relative rotation from successive range-azimuth heatmaps and (b) regresses velocities from doppler-azimuth heatmaps and integrates them into odometry estimates. Right. An optimization-based SLAM backend outputs real-time map and trajectory estimates.
• Figure 5: Left. Physical IWR1843 antenna array. Right. Corresponding virtual antenna array. Azimuth-only heatmaps are derived from subarray enclosed in red. Elevation-aware heatmaps are derived from the subarray enclosed in blue.