Accurate and robust ego-motion estimation is critical for aiding autonomous vehicle localization in environments devoid of map landmarks or accurate GNSS measurements. RADAR measurements provide unique capabilities for ego-motion estimation while also posing unique challenges. This paper presents the novel real-time RADAR-based ego-motion estimation algorithm, RADARODO (RADAR Odometry). RADARODO is differentiated from similar techniques by two novel attributes. First, it decouples sensor translational and rotational motion by estimating them from Doppler and spatial data, respectively. Secondly, it directly analyzes RADAR images to estimate rotational motion via correspondence-free non-linear optimization. Unlike other methods, RADARODO estimates ego-motion along with its predicted uncertainty from a single RADAR sensor without depending on a lever-arm offset or the zero side-slip assumption. The real-world results demonstrate accurate ego-motion estimates that are particularly suited for integration within a 'tightly-coupled' sensor fusion framework.
All Science Journal Classification (ASJC) codes
- Artificial Intelligence
- Automotive Engineering
- Control and Optimization