• Mac OS X or Linux machine

  • Python >= 3.6 (recommended 3.7.9)

  • conda

    • virtualenv is also an acceptable alternative, but we assume you have conda installed in our examples below

Install robomimic#

1. Create and activate conda environment

$ conda create -n robomimic_venv python=3.7.9
$ conda activate robomimic_venv

2. Install PyTorch

PyTorch reference

Option 1: Mac

# Can change pytorch, torchvision versions
# We don't install cudatoolkit since Mac does not have NVIDIA GPU
$ conda install pytorch==1.6.0 torchvision==0.7.0 -c pytorch

Option 2: Linux

# Can change pytorch, torchvision versions
$ conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch

3. Install robomimic

Option 1: Install from source (recommended)

$ git clone
$ cd robomimic
$ pip install -e .

Option 2: Install via pip

$ pip install robomimic

Warning! Additional dependencies might be required

This is all you need for using the suite of algorithms and utilities packaged with robomimic. However, to use our demonstration datasets, you may need additional dependencies. Please see the datasets page for more information on downloading datasets and reproducing experiments, and see the simulators section below.

Optional Installations#

Downloading datasets and reproducing experiments#

See the datasets page for more information on downloading datasets and reproducing experiments.

Install simulators#

If you would like to run robomimic examples and work with released datasets, please install the following simulators:


Required for running most robomimic examples and released datasets. Compatible with robosuite v1.2+. Install via:

# From source (recommended)
$ git clone
$ cd robosuite
$ pip install -r requirements.txt
# Via pip
$ pip install robosuite

(Optional) to use our released datasets and reproduce our experiments, switch to our offline_study branch (requires installing robosuite from source):

git checkout offline_study

mujoco-py dependency!

Robosuite requires mujoco-py. If you are on an Ubuntu machine with a GPU, you should make sure that the GPU version of mujoco-py gets built, so that image rendering is fast (crucial for working with image datasets!).

An easy way to ensure this is to clone the repository, change this line to Builder = LinuxGPUExtensionBuilder, and install from source by running pip install -e . in the mujoco-py root directory.


Useful for running some of our algorithms on the D4RL datasets.

Install via the instructions here.

Test your installation#

This assumes you have installed robomimic from source.

Run a quick debugging (dummy) training loop to make sure robomimic is installed correctly:

$ python examples/ --debug

Run a much more thorough test of several algorithms and scripts (Warning: this script may take several minutes to finish!):

$ bash

To run some easy examples, see the Getting Started section.

Install documentation dependencies#

If you plan to contribute to the repository and add new features, you must install the additional requirements required to build the documentation locally:

$ pip install -r requirements-docs.txt

You can test generating the documentation and viewing it locally in a web browser:

$ make clean
$ make apidoc
$ make html

There should be a generated _build folder - navigate to _build/html/ and open index.html in a web browser to view the documentation.