SafePILCO: a software tool for safe and data-efficient policy synthesis

TL;DR

SafePILCO is a Python-based software tool that enhances the PILCO reinforcement learning algorithm to enable safe, data-efficient policy synthesis, making it more accessible for verification and control applications.

Contribution

It introduces a modular Python implementation of SafePILCO, extending PILCO to support safe learning and broadening its usability across communities.

Findings

Provides a practical, safe reinforcement learning tool

Supports data-efficient policy search in continuous control

Facilitates wider adoption through modular design

Abstract

SafePILCO is a software tool for safe and data-efficient policy search with reinforcement learning. It extends the known PILCO algorithm, originally written in MATLAB, to support safe learning. We provide a Python implementation and leverage existing libraries that allow the codebase to remain short and modular, which is appropriate for wider use by the verification, reinforcement learning, and control communities.