D-KalmanNet: Neural Kalman Filtering for Doppler LiDAR Tracking

This repository contains the pre-trained weights for D-KalmanNet, the tracking component of the DPNet framework. D-KalmanNet integrates a structured Gaussian State Space (GSS) model with a recurrent neural network to accurately predict and track the future states of dynamic obstacles using measurements from (FMCW) Doppler LiDAR.

The full framework can be found in the official GitHub repository.

Model Details

• Developed by: UUwei-zuo
• Dataset Trained On: AevaScenes
• Framework: PyTorch
• Associated Code: GitHub: UUwei-zuo/DPNet
• Paper: [RA-L '26]DPNet: Doppler LiDAR Motion Planning for Highly-Dynamic Environments

How to Use

Intructions for loading the pretraining model.pt or training your custom model can be found in GitHub: UUwei-zuo/DPNet.

Citation

@article{zuo2026dpnet,
  author={Zuo, Wei and Ren, Zeyi and Li, Chengyang and Wang, Yikun and Zhao, Mingle and Wang, Shuai and Sui, Wei and Gao, Fei and Wu, Yik-Chung and Xu, Chengzhong},
  journal={IEEE Robotics and Automation Letters}, 
  title={DPNet: Doppler LiDAR Motion Planning for Highly-Dynamic Environments}, 
  year={2026},
  volume={11},
  number={6},
  pages={7190-7197},
  doi={10.1109/LRA.2026.3685933}
}