A Library for Real‑World
Deep Reinforcement Learning

Concept

machina (mάkɪnə) is a library designed for applying Deep Reinforcement Learning (DRL) to real world tasks.

Deep Reinforcement Learning has seen a considerable increase in the number of available algorithms and policies. Development for real-world tasks is challenging due to the trial-and-error approach and the possibility of changing requirements for environments, algorithms and policies. Therefore, the feature to change individual components of a DRL-system is strongly desired by developers.

Unlike other benchmark libraries, machina is able to deliver a solution to this composability issue by decomposing the dependencies between each component of the DRL-system.

Features

Composability designed for Reinforcement Learning

Composability is an important property in computer programming, allowing to dynamically switch between program components during execution. machina was built and designed with this principle in mind, allowing for high flexibility on system and program development.

Specifically, the RL-policy interacts with the environment via generated trajectories, making the exchange of either component simple. For example, using machina, it is possible to switch between a simulated and a real-world environment during the training phase.

Easy implementation of mixed environments

It is possible for an agent to learn simultaneously or sequentially in a mix of different environments.

Convenient for combining multiple algorithms

Agents are able to learn by combining different state-of-the-art algorithms.

Possibility of changing hyperparameters dynamically

Users can readily implement a meta-optimization method for automated tuning of hyperparemeters.

Quick Start

machina can be installed using PyPI.

pip install machina-rl

Or you can install machina directly from source code.

Slides

Contact

GitHub

Request features, discuss issues, and more.

General Inquiry

info [at] machina-rl.org

A Library for Real‑World Deep Reinforcement Learning