VTire: A Bimodal Visuotactile Tire with High-Resolution Sensing Capability

Environment

• Python version: python=3.8
• Cuda version (or you can just use CPU): cuda=11.8 (if you have another version of cuda, please install torch, torchvision and torchaudio that match your cuda version)
• Install required packages by

pip install -r requirements.txt

Dataset

Install terrain classification dataset VisualTactileDataset and breakage detection dataset BrokenDetectionDatset from Google Drive. The structure of dataset is included in meta.txt.

Train

python train.py --cfg [config_file_name].yaml

Test and Visualize the Results

python plot_results.py --path [log_dir]

Available Resources

• Visualized results are presented in ./media/.
• Configurations for all experiments are included in ./config/ for replicating.

Acknowledgements

• Our implementation of multimodal transformer is based on Visuo-Tacttile-Transformer-for-Manipulation