This is the project page for the paper "Language-Embedded 6D Pose Estimation for Tool Manipulation", accepted to IEEE RA-L 2025.
You can download the rendered RGB-D dataset with functional part pose annotations from the following link:
Download Dataset via Google Drive
This dataset is rendered from 3D models obtained from TurboSquid. Due to the licensing terms of the TurboSquid 3D Model License, we are not allowed to redistribute the original mesh files. The released dataset contains only rendered RGB-D images and annotated 6D poses of functional parts.
⚠️ Important: According to the TurboSquid license (Section II-6-d), you may NOT use this dataset for training any machine learning model without prior written authorization from TurboSquid.
If you would like to use the dataset for training or any machine learning purposes, please contact TurboSquid to obtain permission by sending an email to: use@turbosquid.com.
You are welcome to use the dataset for visualization, evaluation, or academic discussion under appropriate non-training conditions.
The dataset was generated using NVIDIA Isaac Gym for physics simulation and rendering. The generation scripts and simulation tools will be released soon.
The training code for Language Embedded 6D Pose Estimation will be made publicly available in the near future.