Towards Capability-Aware Traversability Navigation for Unstructured Environments
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2026
Gianluca Capezzuto · Felipe Tommaselli · Matheus P. Angarola · Ricardo V. Godoy · Marcelo Becker
Mobile Robotics Group, University of São Paulo, São Carlos, SP, Brazil
Pretrained CAT checkpoint. CAT predicts dense traversability from a single RGB-D image, conditioned on the robot that will drive over it.
📦 Files
| File | Contents |
|---|---|
best_model.pth |
CAT checkpoint |
depth_stats.json |
Training depth statistics per sensor |
The frozen DINOv3 and CLIPSeg backbones download on first use. Demo robot scores and example RGB-D frames are in the code repository.
📍 Minimal Usage
Set up the code repository, then download the weights and run the demo on the NaviTrace examples:
uv run --locked --extra cu128 hf download CAT-IROS-2026/cat best_model.pth depth_stats.json --local-dir checkpoints/cat
uv run --locked --extra cu128 python cat_demo.py \
--checkpoint checkpoints/cat/best_model.pth \
--rgb examples/navitrace/*/rgb.png \
--depth examples/navitrace/*/depth.png \
--scores examples/navitrace/scores.yaml \
--sensor ts
See Minimal Usage for the arguments.
License
The checkpoint is trained on DINOv3 features and released under the DINOv3 License. The code is released under the MIT License.
Citation
If you find this work useful in your research, please cite:
@inproceedings{capezzuto2026cat,
title = {Towards Capability-Aware Traversability Navigation for Unstructured Environments},
author = {Gianluca Capezzuto and Felipe Tommaselli and Matheus P. Angarola and Ricardo V. Godoy and Marcelo Becker},
booktitle = {2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year = {2026},
eprint = {2607.20679},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2607.20679}
}
Model tree for CAT-IROS-2026/cat
Base model
CIDAS/clipseg-rd64-refined