IEEE - Aerospace and Electronic Systems - August 2021 - 33

Credit: Image licensed by Ingram Publishing
The latest research (e.g., Smith et al. [13]) sought to
increase the language expressiveness, but kept their syntax
aimed at temporal planners. Despite being of great value
to the planning community, the added power of expressiveness
can be costly since the models become less intelligible
and more prone to errors. In this regard, Chien
et al. [14] reported that with more general classes of constraints,
finding a plan can be computationally expensive.
In addition to the aforementioned side effects, which are
particularly harmful to the satellite onboard domain, such
an approach is closer to how to program the problem using
manifold automated planning concepts in Artificial Intelligence
(AI) than to properly model the domain for realworld
applications. This takes space system engineers and
operations personnel (who are not AI experts) away from
the modeling process.
In this context, our article proposes a domain definition
language that seeks a lightweight and practical
approach to model the space system behavior and at the
same time meets the demands of an autonomous satellite
(greater performance, scalability, and intelligibility).
Instead of requiring deep knowledge about the planning
language, space engineers can focus on the hierarchical
domain and the spacecraft model to deal with system-level
operational requirements. This issue is especially important
for complex systems, such as satellites, that require
multidisciplinary engineering knowledge to model their
behavior. To increase the model's intelligibility and
improve performance, we propose a language based on,
hierarchical tasks network (HTN), but not limited to. We
describe an extended HTN representation that unifies
planning and scheduling tasks with inference points. It
also includes the description of " acting " and " sensing " to
ensure online planning. To achieve compatibility between
the onboard model and the satellite flight software, we
embrace insights from European Cooperation for Space
Standardization (ECSS). In order to demonstrate the
modeling ability of the proposed language, a detailed case
study focused on a remote-sensing satellite is presented.
An onboard HTN planning approach is implemented to
AUGUST 2021
verify in practice the usability of the model. Realistic
experimental scenarios run on a typical setup of satellite
onboard computer demonstrate that our language is promising.
The remainder of this article is arranged as follows.
First, it reports the related work and brings the background.
Then, we outline the proposed language and its
grammar. Subsequently, our work describes a case study
and shows the experimental study and results. Finally, we
present our conclusions.
RELATED WORK
Planning and scheduling are arguably at the core of autonomy
capability [15], representing the " brain " of an autonomous
satellite. Planning builds a sequence of tasks to
achieve the intended goals, while scheduling allocates
time and resources to activities. Typically, classical planners
based on the Stanford Research Institute Problem
Solver (STRIPS) language use action schemes to decompose
the world into logical conditions and represent a state
as a conjunction of positive literals. As in classical planning,
a language is needed to describe the space domain.
It should describe the behavior of the spacecraft during its
operation from representations of its actions, states, and
goals. However, Smith et al. [13] reported that the propositional
nature and other features of classical planning languages
such as PDDL make it difficult to use for serious
applications such as space systems. As a result, specific
languages have been devised by space agencies in general
for ground automated operation. They are typically based
on temporal intervals, constraints imposed over state values,
and resource consumption. A pioneering language is
ASPEN modeling language (AML) used in the ground
system automated scheduling and planning environment
(ASPEN [16]). Its representation is focused on actions,
states, and resources. However, most languages are
inspired by the notion of timelines, rich in temporal models,
and aimed at CSP planners, as seen in the examples
listed in Table 1.
IEEE A&E SYSTEMS MAGAZINE
33
IEEE - Aerospace and Electronic Systems - August 2021

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - August 2021
