Aerospace and Electronic Systems - March 2019 - 44

ARIES: An Autonomous Controller For Multirobot Cooperation
target point to Navigation by posting a goal to
Waypoint state variable. Then, Navigation updates
the Waypoint state via posting an observation when
the target point is achieved. The Dispatcher is in
charge of communicating with the robotic system.
As can be observed, the communication among
reactors is carried out via observations and goals to
Figure 2.
the state variables.
Translation flow implemented in ARIES between a PDDL Planner using
In conclusion, T-REX is a novel architecture
action-based temporal planning and the Planner Reactor from ARIES using
because it integrates both the deliberation and
timeline-based planning. PDDL-Lib is able to encode an actions plan as a
tokens plan thanks to the domain and effects files.
reaction processes within a single agent to cover
the high and low-level mission management. Nevtemporal planning for three reasons: first, this paper
ertheless, in the current T-REX approach arises
presents an initial version of ARIES; thus, we kept the parathe following scalability limitations in multirobot cooperdigm with, foreseeable, the most convenient formalism for
ation domains: (i) centralized execution, that is, T-REX
simple problem formulations; second, action based is more
was originally built to be deployed in a centralized syscommonly used in the planning community, so there is a
tem which can execute a single or multiple T-REX
broad variety of action-based temporal planners available
agents; (ii) centralized communication, which means that
to be used in ARIES; and third, to open up an initial step to
T-REX does not support the communication among disencode a temporal action-based paradigm such as PDDL
tributed agents executed in different robotic systems.
[26] into a timeline-based paradigm. Bernardini et al. [27]
These limitations make a tough task to bring up a Tpresented a complete translation from a temporal actionREX-based architecture which could deal with any mulbased planning (PDDL2.2 [28]) to a timeline-based plantirobot cooperation scenario. Therefore, the work here
ning (NDDL [29]). In our case, we present initial translapresented is motivated by building a hierarchical archition primitives from PDDL2.1 to the timeline-based
tecture which can give to T-REX the capabilities to be
language used by T-REX: DDL [30].
deployed in any multirobot cooperation domain. For
that, we have integrated the hybrid leader-follower paraContrary to the timeline-based paradigm used by Tdigm into T-REX. In this way, this hybrid paradigm
REX, an action-based temporal language such as
could upgrade the T-REX-based architecture in the
PDDL2.1, uses predicates logic and the world is seen as
following points: distributed execution, that is, capabilian entity that can be in different states. The domain specities to response to individual robot failures without
fies actions that can be performed to change the world
bringing to a complete team failure and to be scalable to
state and only applicable when some particular states are
the problem complexity; hierarchical deliberation, which
set. This paradigm aims to find a sequence of actions that,
centralizes the high-level planning into the leader, gives
from an initial world state, through applying successive
the capability to coordinate changes on the agent's
actions, the system achieves a desired goal state. Thus,
behaviors over time in response to a dynamic environas T-REX and PDDL2.1 are not initially compatible,
ment or changes in the team mission.
a tailored solution was required to merge the action-based
paradigm used by PDDL2.1 to the timeline-based paradigm accepted by T-REX. Therefore, we have developed
FROM TEMPORAL ACTION BASED TO
a common language between PDDL and T-REX that
translates the sequence of actions of a PDDL-based planTIMELINE-BASED PLANNING
ner into a sequence of tokens over their state variables
As we stated in Section "A T-REX OVERVIEW," the Twith temporal constraints. This translation is performed,
REX architecture follows the timeline-based paradigm to
thanks to the PDDL-lib as Figure 2 shows, a PDDL library
model the world of the robotic system. Recent studies about
developed by Mu~
noz et al. [31] to extract the relevant
the timeline-based planning [24], [25] demonstrate that it
information from the PDDL files, execute any PDDLhas not been reached a complete characterization of this
based planner and read the generated plan in order to exeparadigm complexity to model the behavior of robotic syscute it. Particularly, PDDL-lib requires the domain file
tems. For now, these studies show that simplified formulaand effects file to translate a sequence of PDDL actions
tions of timeline-based problems are EXPSPACE(action-based temporal paradigm) as a sequence of tokens
complete and then, can be expressive enough to be stated
(timeline-based paradigm). In the following, we present
as action-based temporal planning problems. Nonetheless,
the functional scope of both configuration files.
it is still unclear where the complexity border between the
The PDDL domain file, as we explained above,
timeline-based planning and the action-based temporal
describes the actions required to change the world
planning is. In this way, we have chosen the action-based
44

IEEE A&E SYSTEMS MAGAZINE

MARCH 2019

Aerospace and Electronic Systems - March 2019

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - March 2019