[论文解读] RadarEye: Robust Liquid Level Tracking Using mmWave Radar in Robotic Pouring

Hongyu Deng, He Chen|arXiv (Cornell University)|Feb 11, 2026

Soft Robotics and Applications被引用 0

一句话总结

RadarEye 使用毫米波雷达与物理感知追踪器在机器人倒水过程中的实时鲁棒液位估计与跟踪，超越视觉与超声基线。

ABSTRACT

Transparent liquid manipulation in robotic pouring remains challenging for perception systems: specular/refraction effects and lighting variability degrade visual cues, undermining reliable level estimation. To address this challenge, we introduce RadarEye, a real-time mmWave radar signal processing pipeline for robust liquid level estimation and tracking during the whole pouring process. RadarEye integrates (i) a high-resolution range-angle beamforming module for liquid level sensing and (ii) a physics-informed mid-pour tracker that suppresses multipath to maintain lock on the liquid surface despite stream-induced clutter and source container reflections. The pipeline delivers sub-millisecond latency. In real-robot water-pouring experiments, RadarEye achieves a 0.35 cm median absolute height error at 0.62 ms per update, substantially outperforming vision and ultrasound baselines.

研究动机与目标

• Motivate robust liquid level sensing for transparent liquids beyond vision-based methods vulnerable to lighting and refraction.
• Introduce RadarEye: a real-time mmWave radar pipeline with high-resolution beamforming for level estimation and a physics-informed tracker to suppress multipath.
• Demonstrate sub-millisecond latency and higher accuracy in real-robot pouring experiments compared to baselines.
• Showcase performance across incremental filling and dynamic pouring tasks with quantitative comparisons.

提出的方法

• High-resolution range–angle beamforming to form a 2D AoA–ToF spectrum for liquid-surface reflections.
• Discretization of the AoA–ToF plane into an N×N grid and coherent summation to build P(i,j).
• A physics-informed tracker that solves a constrained optimal-path problem over time to follow the liquid surface despite multipath (c_t with transition penalty).
• Use of a Q^2 neighborhood to constrain transitions and achieve real-time tracking.
• Implementation with a TI IWR6843 mmWave radar (61.8 GHz, 3.6 GHz bandwidth) on a robotic setup; sub-millisecond latency.

实验结果

研究问题

• RQ1 Can mmWave radar provide robust, direct liquid-level measurements during dynamic pouring despite multipath clutter?
• RQ2 How does RadarEye perform in real-time tracking compared with vision-based and ultrasound baselines?
• RQ3 Does a physics-informed temporal optimization improve stability and accuracy over peak-based or learning-based methods?
• RQ4 What are the latency and accuracy trade-offs of the proposed radar-based pipeline in real-robot pouring?

主要发现

• Median liquid-level error of 0.35 cm with 0.62 ms update latency in real-robot pouring.
• Incremental filling experiments show 0.12 cm median error for step-wise surface tracking.
• RadarEye outperforms vision-based (median error ~2.1 cm) and ultrasound-based (median error ~4.3 cm) baselines.
• A deep learning tracker on the AoA–ToF spectrum yields ~40 ms latency, whereas RadarEye and smoothing achieve ~0.6 ms.
• During incremental filling, the surface Reflection peaks accurately track height changes up to 7.4 cm with low error.