LAPS improves incremental neural LiDAR mapping by actively managing replay samples under a fixed memory and training budget. It consists of two key components: Reliability-based Active Pooling, which retains reliable replay samples while reducing spatial imbalance, and Uncertainty-guided Active Sampling, which allocates online training samples based on uncertainty.

Overview of the LAPS pipeline.

• [June, 2026]: Code released.
• [May, 2026]: LAPS was accepted to IEEE Robotics and Automation Letters (RA-L)!

git clone https://github.com/dongjae0107/LAPS.git

conda create -n laps python=3.10
conda activate laps

conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=12.1 -c pytorch -c nvidia

For different CUDA or PyTorch versions, please refer to the official PyTorch installation guide.

python -m pip install matplotlib open3d scikit-image tabulate
conda install -c conda-forge quaternion

Please download the datasets and organize them as follows:

LAPS/
└── dataset/
    ├── maicity/
    ├── ncd/
    └── spires/

Download the dataset from here or use the following script:

cd dataset/scripts/
bash download_maicity.sh

Download the dataset from here.

Download the dataset from here.

Run LAPS with the corresponding configuration file:

# MaiCity
python run.py config/maicity/maicity.yaml

# Newer College
python run.py config/ncd/ncd.yaml

# Oxford Spires
python run.py config/spires/blenheim_palace.yaml # [blenheim_palace, christ_church, keble_college, observatory_quarter]

If you use this code or find our work useful for your research, please consider citing:

@article{lee2026laps,
  title={LAPS: Improving Incremental LiDAR Mapping using Active Pooling and Sampling for Neural Distance Fields},
  author={Lee, Dongjae and Yang, Wooseong and Tao, Yifu and Fallon, Maurice and Kim, Ayoung},
  journal={IEEE Robotics and Automation Letters},
  year={2026},
  volume={11},
  number={7},
  pages={8584-8591},
  doi={10.1109/LRA.2026.3699255}
}

This codebase builds on ideas and implementations from the following excellent works:

• SHINE_mapping
• PIN_SLAM
• 4dNDF
• N3-Mapping
• SiLVR
• Bayes'Rays