regelum-control 0.3.3

Regelum is a flexibly configurable framework for agent-environment simulation with a menu of predictive and reinforcement learning pipelines.

About

Regelum-control stands as a framework designed to address optimal control and reinforcement learning (RL) tasks within continuous-time dynamical systems. It is made for researchers and engineers in reinforcement learning and control theory.

A detailed documentation is available here.

Explore regelum-playground repo with ready-to-use examples.

Features

• Run pre-configured regelum algorithms with ease. Regelum offers a set of implemented, ready-to-use algorithms in the domain of RL and Optimal Control. It provides flexibility through multiple optimization backends, including CasADi and PyTorch, to accommodate various computational needs.

• Stabilize your dynamical system with Regelum. Regelum stands as a framework designed to address optimal control and reinforcement learning (RL) tasks within continuous-time dynamical systems. It comes equipped with an array of default systems, alongside a detailed tutorial that provides clear instructions for users to instantiate their own environments.

• Manage your experiment data. Regelum seamlessly captures every detail of your experiment with little to no configuration required. From parameters to performance metrics, every datum is recorded. Through integration with MLflow, Regelum streamlines tracking, comparison and real-time monitoring of metrics.

• Reproduce your experiments with ease. Commit hashes and diffs for every experiment are also stored in Regelum, offering the ability to reproduce your experiments at any time with simple terminal commands.

• Configure your experiments efficiently. Our Hydra fork within Regelum introduces enhanced functionaly, making the configuration of your RL and Optimal Control tasks more accessible and user-friendly.

• Fine-tune your models to perfection and achieve peak performance with minimal effort. By integrating with Hydra, regelum inherently adopts Hydra's powerful hyperparameter tuning capability.

Install regelum-control with pip

pip install regelum-control

Developer setup

1. Clone the repository.
2. Run:

   pip install -e .
   bash scripts/developer-setup.sh
   

3. Check requirements-dev.txt in the root of the repo for additional details.

Licence

This project is licensed under the terms of the MIT license.

Bibtex reference

@misc{regelum2024,
author =   {Pavel Osinenko, Grigory Yaremenko, Georgiy Malaniya, Anton Bolychev},
title =    {Regelum: a framework for simulation, control and reinforcement learning},
howpublished = {\url{https://github.com/osinekop/regelum-control}},
year = {2024},
note = {Licensed under the MIT License}
}