NTFields: Neural Time Fields for Physics-Informed Robot Motion Planning
Ruiqi Ni, Ahmed H Qureshi
Paper | GitHub | arXiv | Published in ICLR 2023.
This repository is the official implementation of "NTFields: Neural Time Fields for Physics-Informed Robot Motion Planning".
Clone the repository into your local machine:
git clone https://github.com/ruiqini/NTFields --recursive
Install requirements:
conda env create -f NTFields_env.yml
conda activate NTFields
Download datasets and pretrained models, exact and put datasets/
Experiments/
to the repository directory:
The repository directory should look like this:
NTFields/
├── datasets/
│ ├── arm/ # 4DOF and 6DOF robot arm, table environment
│ ├── c3d/ # C3D environment
│ ├── gibson/ # Gibson environment
│ └── test/ # box and bunny environment
├── Experiments
│ ├── 4DOF/ # pretrained model for 4DOF arm
│ └── Gib/ # pretrained model for Gibson
• • •
• • •
To prepare the Gibson data, run:
python dataprocessing/preprocess.py --config configs/gibson.txt
To prepare the arm data, run:
python dataprocessing/preprocess.py --config configs/arm.txt
To visualize our path in a Gibson environment, run:
python test/gib_plan.py
To visualize our path in the 4DOF arm environment, run:
python test/arm_plan.py
To sample random starts and goals in Gibson environments, run:
python test/sample_sg.py
To show our statistics result in Gibson environments, run:
python test/gib_stat.py
To train our model in a Gibson environment, run:
python train/train_gib.py
To train our model in the 4DOF arm environment, run:
python train/train_arm.py
Build the docker image ntf-image
by running:
cd docker && bash build-image.sh
Run the docker container as follows (let we define docker run --gpus all -v /path/to/NTFields-repo:/code-wkdir ntf-image
as docker-ntf
for brevity):
# preprocessing
docker-ntf dataprocessing/preprocess.py --config configs/gibson.txt
# training
docker-ntf train/train_gib.py
Please cite our paper if you find it useful in your research:
@inproceedings{
ni2023ntfields,
title={{NTF}ields: Neural Time Fields for Physics-Informed Robot Motion Planning},
author={Ruiqi Ni and Ahmed H Qureshi},
booktitle={International Conference on Learning Representations},
year={2023},
url={https://openreview.net/forum?id=ApF0dmi1_9K}
}
NTFields is released under the MIT License. See the LICENSE file for more details.