Aerospace and Electronic Systems - March 2019 - 49

Ropero et al.
[see Figure 7(b)]: the Recover-Reactor (R-Reactor) in the
deliberative layer, the COOP Reactor in the cooperative
layer, and the GER [implemented by Ropero et al.,
(2018)] in the executive layer. As we did with the Leader
Agent, the functional scope of each layer is depicted in
the following.
At the top level, the deliberative layer is formed by
the R-Reactor, a T-REX reactor focused on managing
the communication failures. R-Reactor is called only
when the communication between the Leader and Follower Agent is lost (Follower Failure in Figure 8). As
well as the Planner Reactor of the Leader Agent, RReactor uses a PDDL to model behaviors for recovering
the robotic system from failures, e.g., communication
problems. Also, it uses the PDDL-lib to translate the
sequence of actions into tokens over state variables with
temporal constraints ready for execution. The R-Reactor
declares every state variable of the Follower Agent as
external, in order to be aware of opportunistic system
failures and to act accordingly (C.2 in Figure 8) when a
system recovery is required.
The PDDL planner used by the R-Reactor is UP2TA.
The domain file contains the description of every possible
failure and the sequence of tokens required to recover
from them. The problem file includes the initial facts
defining the initial state of the robotic system and it does
not contain goals because its initial state does not have
failures. A goal into the problem file means that the
robotic system is stuck in a failure state and it requires the
planning of a sequence of tokens to deal with it. Thus, it is
a deliberative reactor, so it may require a certain latency
> 0 and look-ahead p > 1 for the deliberation process.
These values are a matter of design and its dispatching
window HD can be defined as follows:
HD ¼ ½t þ ; t þ þ p:

(3)

At the middle level, the cooperative layer is formed by
the COOP Reactor, whose objective is to enable the coordination with the Leader Agent. It deploys a TCP/IP client
with a unique identifier into the Leader Agent and it
declares the state variables of the Follower as externals.
The COOP Reactor receives the goals from the Leader
Agent (A.2 in Figure 8) and it forwards them to the GER
(A.3 in Figure 8) for their execution. Then, it waits (Follower Waiting in Figure 8) until a goal has been executed
properly in the Follower Agent. Thus, when the COOP
Reactor has received the observation (A.4 in Figure 8)
from the GER, it forwards the observation through the
TCP/IP link (A.5 in Figure 8) to the Leader Agent. If the
communication with the Leader Agent is lost, it reports
the failure to the R-Reactor (Follower Failure in Figure 8)
for triggering the communication failure recovery.
At the bottom level, the executive layer is formed by
the GER, implemented in the same way as the Leader
MARCH 2019

Agent but with different settings, i.e., it declares the state
variables of the Follower Agent as internals (see Figure 8).
Its role is to dispatch the tokens over its internal state variables, as goals requested by the COOP Reactor or the RReactor, to the functional layer of the robotic platform.
Also, it has the capability to monitor the tokens execution
over its internal state variables. If the unachieved goal failure arises, the GER sends every failure to the COOP Reactor to forward them to the Leader Agent (as in A.4, A.5,
and A.6 in Figure 8). Then, the Follower waits for a new
plan from the Leader.

DEMONSTRATION
In this section, we present the ARIES demonstration for a
study case. As we found in the literature (see Section
"STATE OF THE ART"), the evaluation of autonomous
controllers entails to complete autonomous missions by
reaching a set of targets. However, there are no general
problems neither robotic platforms over which we can perform similar comparisons. Furthermore, the experiments
of such autonomous controllers are usually carried out
under hardly reusable and reproducible platforms running
under proprietary hardware/or software. Also, it is often
required the study of several technologies to deploy a fully
operative controller, which can take over a year to build a
proper one depending on the person's experience. Therefore, in practice, is costly to implement an autonomous
controller from the literature for comparison. Also, we
want to remark the time required to model the world
knowledge of the study case and to build the simulation in
a general robotic simulation framework. In order to clarify
this aspect, it took us four months to model the world
knowledge (domain and problem) of the cooperative
mission proposal suited for timeline-based translation.
Thus, we evaluate ARIES over this single mission
proposal depicted for multirobot cooperation. This mission is framed on the future Mars explorations discussed
in the introduction. That is, it is located on the Mars surface and it is performed by a hybrid UGV-UAV system
that cooperates to reach a set of scientific-selected targets.
For the demonstration, we have tested ARIES over one
nominal scenario (i.e., there are no anomalous events during the mission) following the mission proposal, in order
to provide a first technology assessment. To conclude
this section, we present some computational results of the
ARIES execution in several scenarios arisen from this
mission proposal, considering different complexity levels
in terms of tasks number and energy capacity.

STUDY CASE
The study case is based on a mission proposal that arises
from the scientific needs to reach complex targets on

IEEE A&E SYSTEMS MAGAZINE

Aerospace and Electronic Systems - March 2019

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