[2025.11.13] 🎉 Our dataset is now open for public access.
[2025.12.11] 🎊 The Super4DR paper has been posted on arXiv.
In this paper, we propose Super4DR, a radar-centric framework specifically designed for the unique challenges of 4D radar. It comprises a learning-based odometry for pose estimation coupled with a gaussian-based map optimizer to generate dense and complete structure.
Through experiments with diverse scenes and radar data on public datasets and a self-collected dataset from our multi-sensor handheld platform, we demonstrate Super4DR’s superior performance across multiple tasks.
Download link: Super4DR-bag. Code for extracting netdisk data if needed: jr8s
The visualization of our handheld equipment and the projection results among different sensors. Multi-sensor data are collected across various campus scenes under both daytime and nighttime conditions.