PIPE Planner: Pathwise Information Gain with Map Predictions for Indoor Robot Exploration

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2025

Seungjae Baek* · Brady Moon* · Seungchan Kim*
Muqing Cao · Cherie Ho · Sebastian Scherer · Jeong hwan Jeon

Paper | Project Page | Video

Preliminary Setup

Clone the github repository

Clone the repository as below.

git clone --branch init-import --single-branch https://github.com/castacks/pipe-planner.git
cd pipe-planner

Set up Conda Environment

Create environment with the name 'pipe' from lama's conda_env.yml

conda env create -n pipe -f lama/conda_env.yml
conda activate pipe

Build from Source to install 'range_libc'

cd range_libc/pywrapper

# Install build dependencies (if needed)
conda install -y cython

# Build and install
python setup.py install

# Verify installation
cd ../..
python -c "import range_libc; print('range_libc installed successfully')"

Download pretrained prediction models (KTH dataset)

You can download pretrained models from this link. Place the zip file under pretrained_models directory and unzip the file.

mv ~/Downloads/weights.zip ~/pipe-planner/pretrained_models/
cd ~/pipe-planner/pretrained_models/
unzip weights.zip

The pretrained_model directory and its subdirectories should be organized as below:

pipe-planner
├── pretrained_models
    ├── weights
        ├── big_lama
            ├── models
                ├── best.ckpt
        ├── lama_ensemble
            ├── train_1
                ├── models
                    ├── best.ckpt
            ├── train_2
                ├── models
                    ├── best.ckpt
            ├── train_3
                ├── models
                    ├── best.ckpt    

Experiments

Customize your own experiment

In configs/base.yaml, you can manually select map, starting pose, and planning method. All map information and starting points available at here.

log_iou

If true, your algorithm runs until reaching the maximum time step budget (1500 for small maps, 3000 for medium maps, and 6000 for large maps), or reaching the 95% IoU. It saves the IoU score per 20 time steps and when reaching 90% and 95% IoU. If false, the algorithms for designated 'mission_time' time step.

Run the script

Run the 'explore.py' script as below:

cd ../scripts/
python3 explore.py

Citation

If you find our paper or code useful, please cite us:

@inproceedings{baek2025pipe,
  author={Baek, Seungjae and Moon, Brady and Kim, Seungchan and Cao, Muqing and Ho, Cherie and Scherer, Sebastian and Jeon, Jeong Hwan},
  booktitle={2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, 
  title={PIPE Planner: Pathwise Information Gain with Map Predictions for Indoor Robot Exploration}, 
  year={2025},
  pages={7684-7691},
  doi={10.1109/IROS60139.2025.11246190}}