SanD-Planner: Sample-Efficient Diffusion Planner in B-Spline Space for Robust Local Navigation
Paper • 2602.00923 • Published
Error code: RowsPostProcessingError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
dataset_avoid
The dataset_avoid subset (obstacle-avoidance runs) used to train the
SanD-Planner local trajectory
planner.
A single archive dataset_avoid.zip (~14 GB) that extracts to:
dataset_avoid/
└── run_XXXX/ # 152 runs, variable length
├── traj_xyz.npy # (T, 3) trajectory positions (x, y, z), metres
├── traj_yaw.npy # (T,) heading (yaw)
├── traj_pitch.npy # (T,) pitch
└── depth/
└── depth_NNNN.png # per-frame depth images (uint16, mm)
hf download WJCUCL/sandplanner-dataset-avoid dataset_avoid.zip --repo-type dataset --local-dir .
unzip dataset_avoid.zip -d dataset/ # -> dataset/dataset_avoid/run_XXXX/
DATASET_ROOT=dataset bash run_train.sh # run inside the SanD-Planner repo
See the SanD-Planner repository for the training and inference code.
@article{wang2026sand,
title = {SanD-Planner: Sample-Efficient Diffusion Planner in B-Spline Space for Robust Local Navigation},
author = {Wang, Jincheng and Bao, Lingfan and Yang, Tong and Plasencia, Diego Martinez and Jiao, Jianhao and Kanoulas, Dimitrios},
journal = {arXiv preprint arXiv:2602.00923},
year = {2026}
}