A Robust Robot Perception Framework for Complex Environments Using Multiple Mmwave Radars

Hongyu Chen,Yimin Liu,Yuwei Cheng

DOI: https://doi.org/10.1109/jstsp.2024.3420234

IF: 7.695

2024-01-01

IEEE Journal of Selected Topics in Signal Processing

Abstract:The robust perception of environments is crucial for mobile robots to operate autonomously in complex environments. Over the years, mobile robots mainly rely on optical sensors for perception, which degrade severely in adverse weather conditions. Recently, single-chip millimeter-wave (mmWave) radars have been widely used for mobile perception, owing to their robustness to all-weather conditions, lightweight design, and low cost. However, existing research based on mmWave radars primarily focuses on single radar and single task. Due to the limited field of view and sparse observation, perception based on a single radar may not ensure the required robustness in complex environments. To address this challenge, we propose a novel robust perception framework for robots in complex environments based on multiple mmWave radars, named MMR-PFR. The framework integrates three critical tasks for robots, including ego-motion estimation, multi-radar fusion mapping, and dynamic target state estimation. Multiple tasks collaborate and facilitate each other to improve overall performance. In the framework, we propose a new multi-radar point cloud fusion method to generate a more accurate environmental map. In addition, we propose a new online calibration algorithm for multiple radars to ensure the long-term reliability of the system. To evaluate MMR-PRF, we build a prototype and carry out experiments in real-world scenarios. The evaluation results show the effectiveness and superiority of the proposed framework.