2D Ego-Motion with Yaw Estimation using Only mmWave Radars via Two-Way weighted ICP

TL;DR

The paper tackles yaw-rate estimation for 2D ego-motion using only mmWave radar data, addressing the radar-only perception gap without IMU or multiple radars. It introduces a radar-only pipeline that first estimates 2D linear velocity from Doppler data and then registers heatmap-derived features via a two-way weighted ICP, with the rotation solved in $SO(2)$. Synchronization between single-chip and cascade radars is achieved by velocity interpolation and frame rectification, with intensity-based weights guiding the registration through $R^* = UV^T \in SO(2)$. Experiments on the ColoRadar dataset show that Top-k heatmap preprocessing combined with mwICP yields robust yaw estimates and accurate 2D trajectories, while challenging scenes reveal remaining limitations and point to future extension to single-chip heatmaps.

Abstract

The interest in single-chip mmWave Radar is driven by their compact form factor, cost-effectiveness, and robustness under harsh environmental conditions. Despite its promising attributes, the principal limitation of mmWave radar lies in its capacity for autonomous yaw rate estimation. Conventional solutions have often resorted to integrating inertial measurement unit (IMU) or deploying multiple radar units to circumvent this shortcoming. This paper introduces an innovative methodology for two-dimensional ego-motion estimation, focusing on yaw rate deduction, utilizing solely mmWave radar sensors. By applying a weighted Iterated Closest Point (ICP) algorithm to register processed points derived from heatmap data, our method facilitates 2D ego-motion estimation devoid of prior information. Through experimental validation, we verified the effectiveness and promise of our technique for ego-motion estimation using exclusively radar data.

Figures (5)

• Figure 1: Visualization of the 2D ego-motion estimation process using only mmWave radar, along with the estimated trajectory in an indoor environment.
• Figure 2: Overview of the proposed method to estimate 2D ego-motion divided by two sections.
• Figure 3: Illustration of our key point matching step. From the intensity heatmap, preprocess the Top k points(top left) and rectify the previous frame points by estimated linear velocities(top right). After sampling the points, find the matched pair to perform weighted ICP(bottom left). Aligned points at 1st iteration in current source to previous target(bottom right). Black point on the bottom is the origin which represents the robot position.
• Figure 4: Cumulative squared error of the estimated yaw in (a)EC Hallways 0 sequence and (b)Aspen 5 sequence respect to the preprocessing methods. Through the estimated 2D ego-motion, the trajectories of each sequences (c) and (d) with Top k preprocessing method. Preprocessing with one-way weighted ICP represents using the previous frame as the source and the current frame as the target.
• Figure 5: Challenging scenes, (left)unstable features with different segment size or (right) curvature distortion after the rectification in narrow place. Black point represent the center of the sensor and the gray lines implies the detection range of the heatmap.