robot_sim_dexwm

Robot Simulation for DexWM

This repo is a supporting simulation software for DexWM and is used for data generation, and evaluation of the model. The repo adopts the code from RoboCasa with key changes to incorporate the MURP robot.

Installation steps:

1. Set up conda environment:l

   conda create -c conda-forge -n robot_sim_dexwm python=3.10

2. Activate conda environment:

   conda activate robot_sim_dexwm

3. Clone and setup robosuite dependency:

   cd robosuite_murp
   pip install -e .

4. Setup Robocasa:

   cd robocasa-murp
   pip install -e .
   pip install imageio
   
   (optional: if running into issues with numba/numpy, run: conda install -c numba numba=0.56.4 -y)

5. Install the package and download assets:

   python robocasa/scripts/download_kitchen_assets.py
   python robocasa/scripts/setup_macros.py
   python robosuite/scripts/setup_macros.py

   Aigen objects assets have a higher probability of not getting installed during the process. To avoid any errors , we install it manually :

   cd robocasa-murp/robocasa/models/assets/objects
   wget https://utexas.box.com/shared/static/os3hrui06lasnuvwqpmwn0wcrduh6jg3.zip
   unzip

6. Re-Install Numpy and Numba versions (If faced with error):

   pip install numpy==1.23.3 numba==0.56.4

BibTeX

DexWM:

@article{goswami2025world,
  title={World Models for Learning Dexterous Hand-Object Interactions from Human Videos},
  author={Goswami, Raktim Gautam and Bar, Amir and Fan, David and Yang, Tsung-Yen and Zhou, Gaoyue and Krishnamurthy, Prashanth and Rabbat, Michael and Khorrami, Farshad and LeCun, Yann},
  journal={arXiv preprint arXiv:2512.13644},
  year={2026}
}
}

RoboCasa:

@inproceedings{robocasa2024,
  title={RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots},
  author={Soroush Nasiriany and Abhiram Maddukuri and Lance Zhang and Adeet Parikh and Aaron Lo and Abhishek Joshi and Ajay Mandlekar and Yuke Zhu},
  booktitle={Robotics: Science and Systems},
  year={2024}
}
}