Learning Long-Horizon Robot Exploration Strategies for Multi-Object Search in Continuous Action Spaces
Repository providing the source code for the paper "Learning Long-Horizon Robot Exploration Strategies for Multi-Object Search in Continuous Action Spaces", see the project website. Please cite the paper as follows:
@article{fabian22exploration,
title={Learning Long-Horizon Robot Exploration Strategies for Multi-Object Search in Continuous Action Spaces},
author={Schmalstieg, Fabian and Honerkamp, Daniel and Welschehold, Tim and Valada, Abhinav},
journal={Proceedings of the International Symposium on Robotics Research (ISRR)},
year={2022}
}
conda create -f environment.yaml
conda activate igibson
The code does not work without downloading and unzipping the iGibson dataset. First download assets and then, the dataset of the scenes. Unzip the dataset into the iGibson/data/. folder. After the unzipping, the iGibson/data folder should have a ig_dataset folder. Don't forget to download the igibson.key files under https://stanfordvl.github.io/iGibson/dataset.html
python -m igibson.utils.assets_utils --download_assets
https://storage.googleapis.com/gibson_scenes/ig_dataset.tar.gz
After Installing SB3 and iGibson, copy all files which are located in requirements/ , into the respective folders of iGibson/data. You have to overwrite the existing files. These files are only used during training in order to use inflated maps.
run the model with different scenes using
python evaluate.py
change the scene_id name in config.yaml to one of the following scenes:
Test Scenes: Benevolence_1_int, Wainscott_0_int, Pomaria_2_int, Benevolence_2_int, Beechwood_0_int, Pomaria_1_int, Merom_1_int
Training Scenes: Merom_0_int, Benevolence_0_int, Pomaria_0_int, Wainscott_1_int, Rs_int, Ihlen_0_int, Beechwood_1_int, Ihlen_1_int
run training using:
python training.py
This work was funded by the european union's horizon 2020 research and innovation program under grant agreement no 871449-OpenDR.