Overview

This project uses robosuite, a modular simulation framework for robot learning developed by Zhu et al. (2020). We have modified robosuite to add new robot models (including the model for SLIM and our custom differential wrist), new tasks, and an Oculus Meta Quest teleoperation interface. The Meta Quest interfaces utilizes oculus_reader by Jedrzej Orbik and Frederik Ebert (2021).

Models

• See the following python class for the SLIM hand which points to the MuJoCo XML for SLIM
• The following includes the python class for the Franka arm appended with our supplemental wrist

Tasks

See the following files for the task environments. In each file, the _check_success() method encodes the definition of success and is used for automatic task completion detection.

• The Cabinet Pick task is defined here.
• The Cabinet Reorient task is defined here.
• The Box Pick task is defined here.
• The Cabinet Reorganize task is defined here.

Teleoperation Interface

The interface for the Oculus Meta Quest controller is found here. This interfaces between robosuite and oculus_reader.

Operational Space Controller

Robosuite provides an operational space controller, which is used here for task-space teleoperation. The file implementing this controller is found here.

Citations

@inproceedings{robosuite2020, title={robosuite: A Modular Simulation Framework and Benchmark for Robot Learning}, author={Yuke Zhu and Josiah Wong and Ajay Mandlekar and Roberto Mart'{i}n-Mart'{i}n and Abhishek Joshi and Soroush Nasiriany and Yifeng Zhu and Kevin Lin}, booktitle={arXiv preprint arXiv:2009.12293}, year={2020} }

@misc{OrbikEbert2021OculusReader, author = {Jedrzej Orbik and Frederik Ebert}, title = {Oculus Reader: Robotic Teleoperation Interface}, year = {2021}, url = {https://github.com/rail-berkeley/oculus_reader}, note = {Accessed: 2025-04-24} }