dexterity 0.0.2

Software and tasks for dexterous multi-fingered hand manipulation, powered by MuJoCo

The MuJoCo Dexterity Suite (alpha-release)

Software and tasks for dexterous multi-fingered hand manipulation, powered by MuJoCo.

dexterity builds on dm_control and provides a collection of modular components that can be used to define rich Reinforcement Learning environments for dexterous manipulation. It also comes with a set of standardized tasks that can serve as a performance benchmark for the research community.

An introductory tutorial is available as a Colab notebook:

Installation

PyPI (Recommended)

The recommended way to install this package is via PyPI:

pip install dexterity

Source

We use Python 3.8 and Miniconda for development. To create an environment and install dependencies, run the following steps:

conda env create -f environment.yml  # Creates a dexterity env.
conda activate dexterity
pip install -e ".[dev]"

Overview

The MuJoCo dexterity suite is composed of the following core components:

• models: MuJoCo models for dexterous hands and PyMJCF classes for dynamically customizing them.
• inverse_kinematics: Inverse kinematics solver for multi-fingered hands.
• effectors: Interfaces for controlling hands and defining action spaces.

These components in conjunction with dm_control allow you to define and customize rich environments for reinforcement learning.

dexterity also comes pre-packaged with a suite of benchmark RL environments. Our hope is to grow it over time with crowd-sourced contributions from the research community.

Task Name	Description	Preview
reach	Successively controlling the fingers of the hand to reach goal locations in 3D space.
reorient	Reorienting the OpenAI cube to a goal orientation.

Roadmap

• Add support for bi-manual tasks.
• Add more tasks to the suite.

Acknowledgements

A large part of the design and implementation of dexterity is inspired by the MoMa library in dm_robotics.