RoboCasa

RoboCasa is a large-scale simulation framework for training generally capable robots to perform everyday tasks. It was originally released (opens in a new tab) in 2024 by UT Austin researchers. The latest iteration, RoboCasa365, builds upon the original release with significant new functionalities to support large-scale training and benchmarking in sim. Four pillars underlie RoboCasa365:

• Diverse tasks: 365 tasks created with the guidance of large language models
• Diverse assets: including 2,500+ kitchen scenes and 3,200+ 3D objects
• High-quality demonstrations: including 600+ hours of human demonstrations in addition to 1,600+ hours of robot datasets created with automated trajectory tools
• Benchmarking support: popular policy learning methods including Diffusion Policy, pi, and GR00T, plus user-submitted models on the leaderboard (opens in a new tab)

This guide contains information about installation and setup. Please refer to the following resources for additional information:

[Home page] (opens in a new tab)   [Documentation] (opens in a new tab)   [RoboCasa365 Paper] (opens in a new tab)   [Original RoboCasa Paper] (opens in a new tab)   [Leaderboard] (opens in a new tab)

---

Updates

• [10/28/2026] The leaderboard now accepts agentic submissions (opens in a new tab): new models or novel agentic harnesses built around a base policy.
• [10/14/2026] Released the subtask annotation pipeline (opens in a new tab), with labelers covering all 300 composite tasks. It generates per-frame subtask labels directly from each demo's simulation state.
• [7/7/2026] Our target composite task datasets have been updated to include per-frame subtask annotations. Every timestep is labeled with a subtask index, atomic-skill name, stage (i.e. pick / place / navigate), and a natural-language instruction, to support hierarchical policy learning.
• [5/12/2026] v1.0.1: Updated horizon lengths (1.5x increase) across all tasks for consistency. Please update to the latest version for running evals.
• [4/6/2026] RoboCasa365 Leaderboard (opens in a new tab) published, comparing generalist robot policies on the multi-task learning benchmark and open to community model submissions.
• [2/18/2026] v1.0: RoboCasa365 release, with 365 tasks, 2500+ kitchen scenes, 2200+ hours of robot demonstration data, and benchmarking support.
• [10/31/2024] v0.2: using RoboSuite v1.5 as the backend, with improved support for custom robot composition, composite controllers, more teleoperation devices, photo-realistic rendering.

Table of Contents

• Updates
• Installation
• Basic Usage
• Tasks, datasets, policy learning, and additional use cases
• Questions
• License
• Citation

---

Installation

RoboCasa works across all major computing platforms. The easiest way to set up is through the Anaconda (opens in a new tab) package management system. Follow the instructions below to install:

1. Set up conda environment:

   conda create -c conda-forge -n robocasa python=3.11
   

2. Activate conda environment:

   conda activate robocasa
   

3. Clone and setup robosuite dependency (important: use the master branch!):

   git clone https://github.com/ARISE-Initiative/robosuite
   cd robosuite
   pip install -e .
   

4. Clone and setup this repo:

   cd ..
   git clone https://github.com/robocasa/robocasa
   cd robocasa
   pip install -e .
   pip install pre-commit; pre-commit install           # Optional: set up code formatter.
   
   (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 -m robocasa.scripts.setup_macros              # Set up system variables.
   python -m robocasa.scripts.download_kitchen_assets   # Caution: Assets to be downloaded are around 10GB.
   

---

Basic Usage

Gym wrapper

You can create environments using gym wrappers and run rollouts:

import gymnasium as gym
import robocasa
from robocasa.utils.env_utils import run_random_rollouts

env = gym.make(
    "robocasa/PickPlaceCounterToCabinet",
    split="pretrain", # use 'pretrain' or 'target' kitchen scenes and objects
    seed=0 # seed environment as needed. set seed=None to run unseeded
)

# run rollouts with random actions and save video
run_random_rollouts(
    env, num_rollouts=3, num_steps=100, video_path="/tmp/test.mp4"
)

Play back sample demonstrations of tasks

(Mac users: for these scripts, prepend the "python" command with "mj": mjpython ...)

Select a task and play back a sample demonstration for the selected task:

python -m robocasa.demos.demo_tasks

Explore kitchen scenes

Explore 2500+ kitchen scenes:

python -m robocasa.demos.demo_kitchen_scenes

Explore library of 2500+ objects

View and interact with both human-designed and AI-generated objects:

python -m robocasa.demos.demo_objects

Note: By default, this demo shows objaverse objects. To view AI-generated objects, add the flag --obj_types aigen.

Teleoperate the robot

Control the robot directly, either through a keyboard controller or spacemouse. This script renders the robot semi-translucent in order to minimize occlusions and enable better visibility.

python -m robocasa.demos.demo_teleop

Note: If using SpaceMouse, you may need to modify the product ID to your appropriate model, setting SPACEMOUSE_PRODUCT_ID in robocasa/macros_private.py.

---

Tasks, datasets, policy learning, and additional use cases

Please refer to the documentation page (opens in a new tab) for information about tasks, datasets, benchmarking, and more.

---

Questions

If you have any questions about RoboCasa, reach out at robocasa@googlegroups.com (opens in a new tab).

---

License

Code: MIT License (opens in a new tab)

Assets and Datasets: CC BY 4.0 (opens in a new tab)

---

Citation

RoboCasa365:

@inproceedings{robocasa365,
  title={RoboCasa365: A Large-Scale Simulation Framework for Training and Benchmarking Generalist Robots},
  author={Soroush Nasiriany and Sepehr Nasiriany and Abhiram Maddukuri and Yuke Zhu},
  booktitle={International Conference on Learning Representations (ICLR)},
  year={2026}
}

RoboCasa (Original Release):

@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 (RSS)},
  year={2024}
}