This is the PyTorch implementation of Particle Filter Recurrent Neural Networks (PF-RNNs).

Xiao Ma, Peter Karkus, David Hsu, Wee Sun Lee: Particle Filter Recurrent Neural Networks. AAAI Conference on Artificial Intelligence (AAAI), 2020.

Above is the network structures for PF-LSTM and PF-GRU. In PF-RNNs, we maintain a set of latent particles and update them using particle filter algorithm. In our implementation, PF-LSTM and PF-GRU update particles in a parallel manner which benefit from the GPU acceleration.

pip install -r requirements.txt

The training parameters are specified in configs/train.conf. To run the robot localization experiment, use

python main.py -c ./configs/train.conf

If you find this work useful, please consider citing us

@inproceedings{ma2020particle,
  author    = {Xiao Ma and
               P{\'{e}}ter Karkus and
               David Hsu and
               Wee Sun Lee},
  title     = {Particle Filter Recurrent Neural Networks},
  booktitle = {The Thirty-Fourth {AAAI} Conference on Artificial Intelligence, {AAAI}, 2020},
  pages     = {5101--5108}
}