RaLL: End-to-end Radar Localization on Lidar Map Using Differentiable Measurement Model

Video: YouTube | Bilibili

**NOTE: This repo is built in 2020. Please use an old PyTorch version to run it, e.g., torch==1.5.0**

• data/maps (>80Mb)

point cloud maps and images with resolution of 0.25m/pixel

We have uploaded the radar scan images of Seq02 on Google Drive.

• data/gt_poses

groud truth poses for evaluation

• data/odom

odometry data via ICP

• ekf_filter

differetiable ekf implementation

• loader

dataLoader for training and inference

• loss

cross entropy loss (L1) and squared error loss (L2)

• models

Pre-trained models of RaLL

• network

feature extraction network and patch network

• test_py

test pose tracking on RobotCar and MulRan

Please use the train_rall_L12.py and train_rall_L3.py. Please modify the data path in the python files.

If you use the data or code in an academic work, or inspired by our method, please consider citing the following:

@article{yin2021rall,
  title={RaLL: End-to-end Radar Localization on Lidar Map Using Differentiable Measurement Model},
  author={Yin, Huan and Chen, Runjian and Wang, Yue and Xiong, Rong},
  journal={IEEE Transactions on Intelligent Transportation Systems},
  year={2021},
  publisher={IEEE}
}

We also propose a heterogeneous place recognition method with radar and lidar. Please refer to Radar-to-Lidar for details.