Aerospace and Electronic Systems - March 2019 - 43

Ropero et al.
current position in real time. The task of a planner
is to find a sequence of tokens that brings the timelines into a final desired state, satisfying the rules
and special conditions called goals.
In T-REX, the state variables are distributed
into modules called reactors. A reactor manipulates specific information, such as the navigation
system managing just the positioning and navigation policy of a rover. State variables are exclusive
property of only one reactor. There are two classes
of state variables:
Internal state variables are controlled by the
owner reactor. The only reactor that defines
a state variable as internal has sole responsibility to update it via observations. Also,
goals posted by other reactors to this state
variable can be considered by the reactor to
plan ahead on any goal state.

Figure 1.
External state variables provides the world
T-REX agent is formed by multiple reactors (red boxes) which are connected
information to a reactor that enables to
through state variables provided by one reactor as internals (solid lines) and
define its own internal state, but the reactor
available for other reactors as externals (dashed lines).
has no control over them. However, the
reactor can request a specific state on an
This implies that a reactor A dispatches a goal to a
external variable posting goals to the owner reactor.
reactor B (owner of the state variable) as soon as
the start time of the goal intersects HD of the reactor
Each T-REX reactor encapsulates the planning and
B. The dispatching window guarantees that each
execution processes into a control loop, such as a state
reactor has enough time, thanks to ; and enough
machine. The frequency at which the state update happens
information, thanks to p, to deliberate on goals
is dictated by a central clock, which uses the tick as the
provided by other reactors.
time unit. The duration of a tick is a matter of design. The
control loop of a T-REX reactor is based on three ordered
Deliberation is used by every reactor to plan a
steps:
sequence of tokens aiming to achieve the received
goals.
Synchronization is used by each reactor to keep
updated its internal state variables by monitoring
Therefore, a T-REX agent is formed by an ordered set
the external state variables. That is, a reactor update
of reactors, whose objective is to change the environment
may produce a goal execution over an internal state
of the robotic platform through actions (or tokens) on their
variable of another reactor.
state variables. This division enables the problem to be distributed into several reactors in different abstraction levels.
Dispatching guarantees that reactors dispatch
Typically, we can differentiate deliberative and executive
goals among them in a timely manner. Each reacreactors. Deliberative reactors are aware of the high-level
tor defines its own dispatching window to idenmission goals and could require larger temporal scopes ()
tify when tokens (following the token start time)
for problem solving. Instead, executive reactors are in a
on an external state variable should be translower abstraction level and commonly implement reactive
formed into goals for the reactors that declares
and/or reflexive behaviors. Then, the T-REX engine is in
the state variable as internal. The dispatching
charge of synchronizing all reactors, so they can be exewindow, denoted as HD , is defined as in (1),
cuted concurrently and all are notified of the evolution of
where t is the execution frontier that expresses
those (external) state variables that affect them, being able
the current tick of the execution, is the latency
to deliberate and dispatch appropriated plans.
defined as the maximum number of ticks that a
Figure 1 shows an instance of a T-REX agent formed
reactor has to deliberate, and p is the planning
by two deliberative reactors (Path-Planner and Taskhorizon of a reactor representing the look-ahead
Planner), two executive reactors (Navigation and Device
for deliberation
Controller), and the Dispatcher reactor. The PathPlanner, which has the Waypoint state variable as exterHD ¼ ½t þ ; t þ þ p:
(1)
nal, computes the routing for the vehicles and reports each
MARCH 2019

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