cff-version: 1.2.0
title: >-
  BSK-RL: Modular, High-Fidelity Reinforcement Learning
  Environments for Spacecraft Tasking
message: >-
  If you use this software, please cite it using the
  metadata from this file.
type: software
authors:
  - given-names: Mark
    family-names: Stephenson
    email: Mark.A.Stephenson@colorado.edu
    affiliation: 'University of Colorado, Boulder'
    orcid: 'https://orcid.org/0009-0004-3438-8127'
  - given-names: Hanspeter
    family-names: Schaub
    orcid: 'https://orcid.org/0000-0003-0002-6035'
    affiliation: 'University of Colorado, Boulder'
    email: Hanspeter.Schaub@colorado.edu
identifiers:
  - type: url
    value: 'https://hanspeterschaub.info/Papers/Stephenson2024c.pdf'
repository-code: 'https://github.com/AVSLab/bsk_rl/'
url: 'https://avslab.github.io/bsk_rl/'
abstract: >-
  Reinforcement learning (RL) is a highly adaptable
  framework for generating autonomous agents across a wide
  domain of problems. While RL has been successfully applied
  to highly complex, real-world systems, a significant
  amount of the literature studies abstractions and
  idealized versions of problems. This is especially the
  case for the field of spacecraft tasking, in which even
  traditional preplanning approaches tend to use highly
  simplified models of spacecraft dynamics and operations.
  When simplified methods are tested in a full-fidelity
  simulation, they often lead to conservative solutions that
  are suboptimal or aggressive solutions that are
  infeasible. As a result, there is a need for a
  high-fidelity spacecraft simulation environment to
  evaluate RL-based and other tasking algorithms. This paper
  introduces BSK-RL, an open-source Python package for
  creating and customizing reinforcement learning
  environments for spacecraft tasking problems. It combines
  Basilisk --- a high-speed and high-fidelity spacecraft
  simulation framework --- with abstractions of satellite
  tasks and operational objectives within the standard
  Gymnasium API wrapper for RL environments. The package is
  designed to meet the needs of RL and spacecraft operations
  researchers: Environment parameters are easily
  reproducible, customizable, and randomizable. Environments
  are highly modular: satellite state and action spaces can
  be specified, mission objectives and rewards can be
  defined, and the satellite dynamics and flight software
  can be configured, implicitly introducing operational
  limitations and safety constraints. Heterogeneous
  multi-agent environments can be created for more complex
  mission scenarios that consider communication and
  collaboration. Training and deployment using the package
  are demonstrated for an Earth-observing satellite with
  resource constraints.
license: MIT
version: 1.0.1
date-released: '2024-08-27'