Skip to content

6. Tutorials & FAQs

Jump to bottom
Vikash Kumar edited this page Jul 25, 2023 · 14 revisions

Installation

How to install RoboHive using pipy?

We recommend installation within a conda environment. If you don't have one yet, create one using

conda create -n robohive python=3
conda activate robohive

It is possible to install RoboHive with:

pip install -U robohive
How to install RoboHive in editable mode?

RoboHive uses git submodules to resolve dependencies. Please follow the steps exactly as below to install it correctly.

  1. Clone this repo with pre-populated submodule dependencies as -

    git clone --recursive https://github.com/vikashplus/robohive.git

    Note: RoboHive aggressively uses git-submodules. Therefore, it is important to use the exact command above for installation. A basic understanding of how to work with submodules is expected.

  2. Install the package

    cd robohive
pip install -e .[a0] #with a0 binding for realworld robot
pip install -e .     #simulation only

  3. Export the pythonpath

    export PYTHONPATH="<path/to/robohive>:$PYTHONPATH"

OR Add repo to pythonpath by updating ~/.bashrc or ~/.bash_profile

  1. You can visualize the environments with random controls using the below command 
    $ python robohive/utils/examine_env.py -e FrankaReachRandom-v0
How to install RoboHive on a colab?

RoboHive has a detailed colab tutorial to walk you through various steps of installation to env-interactions. Give it a try.

Which MuJoCo bindings are supported in RoboHive? How can I pick MuJoCo bindings for RoboHive?

RoboHive supports both MuJoCo bindings - MuJoCo bindings from DeepMind as well as MuJoCo_Py bindings form OpenAI.

Set the environment variable sim_backend=MUJOCO or sim_backend=MUJOCO_PY to choose which binding to use.

GLFW errors during visualization?
  1. If the visualization results in a GLFW error, this is because mujoco-py does not see some graphics drivers correctly. This can usually be fixed by explicitly loading the correct drivers before running any scripts. See this page for more details.
  2. If FFmpeg isn't found then run apt-get install ffmpeg on linux and brew install ffmpeg on osx (conda install FFmpeg causes some issues)
How to pick OpenGL rendering backends for onscreen/offscreen?

Rendering in RoboHive is directly handled by the simulation backend --

  • mujoco_py: You can check the backend using python -c "import mujoco_py; print(mujoco_py.cymj)"
  • mujoco: follow instructions here
How to install hardware drivers to work with robots?

Please follow instructions in the Robohive's robot class for hardware install instructions.

Explore Environments

How to explore an environment onscreen using random actions? RoboHive exposes a powerful utility script to examine environments. The following command can be used to examine an environment with random actions. 
  $ python robohive/utils/examine_env.py --env_name FrankaReachRandom-v0

Also try the python robohive/utils/examine_env.py --help to view all available options.

How to explore an environment onscreen using actions from specified policy?

RoboHive exposes a powerful utility script to examine environments using actions from a policy. The following command can be used

  $ python robohive/utils/examine_env.py --env_name FrankaReachRandom-v0 --policy_path <path_to_policy>

the policy is expected to be a pickle file. The policy class is expected to expose an act = policy.get_actions(observations) method. See robohive/utils/examine_env.py for more details.
Also, try the python robohive/utils/examine_env.py --help to view all available options.

How to explore an environment offscreen?

RoboHive exposes a powerful utility script to examine environments. The following command can be used to examine a policy offscreen

  $ python robohive/utils/examine_env.py --env_name FrankaReachRandom-v0 --render offscreen

Also, try the python robohive/utils/examine_env.py --help to view all available options for onscreen/offscreen rendering.

Register Environments

How to register a RoboHive environment?

RoboHive environment follows OpenAI's gym API. The environment can be registered like any gym environment. We provide a detailed notebook explaining the process of registering and instantiating a RoboHive environment at - robohive/tutorials/1_env_registration_and_customization.ipynb

How to customize an existing RoboHive environment?

RoboHive environment follows OpenAI's gym API. We provide a detailed notebook explaining the process of customizing RoboHive's environments at - robohive/tutorials/1_env_registration_and_customization.ipynb

Interact with Environments

How to interact with a RoboHive environment?

RoboHive environment follows OpenAI's gym API. Environment interactions are similar to any gym environment. We provide a detailed notebook at - robohive/tutorials/2_env_interactions.ipynb explaining with examples of how to interact with RoboHive's environments.

How to access details (observation/ proprioception/ exteroception) from the RoboHive environment?

RoboHive supports three ways to query and access the environment based on the nature of the information required -- observation/ proprioception/ exteroception We provide a detailed notebook at - robohive/tutorials/3_get_obs_proprio_extero.ipynb explaining with examples how to access these features.

Explore Simulations

How to render images offscreen from a RoboHive simulation?

RoboHive simulations are pure MuJoCo models. We provide a tutorial (robohive/tutorials/render_cams.py) showing ways to render images offscreen and save them using a mujoco model

Explore Robots

How to interact with a RoboHive's Robot without the environment?

RoboHive provides an explicit robot class that hides the differences between the sim and the real allowing users to seamlessly switch between the two. We provide a tutorial (/robohive/tutorials/examine_robot.py) to demonstrate how to use the RoboHive's Robot class in isolation. This use-case is common in scenarios where the entire env definitions aren't required for experiments. In this tutorial, we demonstrate how we can use the RoboHive's Franka Robot in the real world using specifications available in a hardware config file.

    # python tutorials/examine_robot.py --sim_path PATH_robot_sim.xml --config_path PATH_robot_configurations.config
    $ python tutorials/examine_robot.py --sim_path envs/arms/franka/assets/franka_reach_v0.xml --config_path envs/arms/franka/assets/franka_reach_v0.config

Inverse Kinematics

How to use RoboHive's inverse kinematics (with min jerk)?

RoboHive provides support for inverse kinematics. We provide a tutorial (/robohive/tutorials/ik_minjerk_trajectory.py) to demonstrate how to use RoboHive's inverse kinematics with a min-jerk trajectory.

python tutorials/ik_minjerk_trajectory.py --sim_path envs/arms/franka/assets/franka_busbin_v0.xml

Tele-Operation

How to use RoboHive's teleoperation capabilities with keyboard/ spacenav?

RoboHive provides support for teleoperation. We provide a tutorial (/robohive/tutorials/ee_teleop.py) to demonstrate how to use RoboHive's teleoperation capabilities over an environment to collect data

python robohive/tutorials/ee_teleop.py -e rpFrankaRobotiqData-v0
How to use RoboHive's teleoperation capabilities with Oculus controller?

RoboHive provides support for teleoperation. We provide a tutorial (robohive/tutorials/ee_teleop_oculus.py) to demonstrate how to use RoboHive's teleoperation capabilities over an environment to collect data

python tutorials/ee_teleop.py -e rpFrankaRobotiqData-v0
How do I set up RoboHive's teleoperation in the sim/ real?

RoboHive provides support for teleoperation. We provide a video tutorial to walk you through RoboHive's

  • installation
  • teleoperation capabilities over an environment to collect data
  • Easy of switching between simulation and hardware teleoperation

RoboHive: A unified framework for robot learning

Clone this wiki locally