IEEE Robotics & Automation Magazine - September 2015 - 164

generating twist commands to the BaseController component to follow a given trajectory. Let us now consider a different application in which the robot has to follow a moving
target (e.g., another robot). The TrajectoryFollower component is replaced with the LeaderFollower component that
periodically generates twist commands for the robot according to the position of the target estimated by the sensors.
The HyperFlex allows
representation of the navigation strategy as a variaA robot control system is
tion point that can be
composed of tens of components. resolved by selecting at
deployment time either
the LeaderFollower component or the TrajectoryFollower component. Figure 5
shows the Feature Model of the navigation system. Green
boxes represent the selected features. The triangles below the
boxes indicate the cardinality of the variation point. In particular, it means that, for each variation point, only one variant can be selected. A black circle indicates that the child
feature is mandatory.
The selection of desired features triggers model-to-model
transformations that configure the template architecture.
Several types of transformations can be defined, such as
removing a component, changing the properties of a component, and changing the connections between components.
For example, the selection of the feature LeaderFollower triggers a transformation that replaces the TrajectoryFollower
component with the LeaderFollower component and configures the connectors. The HyperFlex toolchain offers the possibility to compose hierarchically Feature Models in such a
way that higher-level Feature Models abstract the subsystems
functional features. Here, the idea is to model features at different levels of abstractions for different types of users.
Typically, the expert in robotic functionalities is interested
in a representation of the functional features that highlights
the different algorithms implemented in the robot control
system. On the contrary, the application domain expert is

Listing 3. The Specification of a Differential
Constraint.
BEGIN DifferentialDrive
ACTION: ul, ur;
PARAM: L, r;
CONFIG: x, y, theta;
d(x) = r/2 * (ul + ur) * cos(theta);
d(y) = r/2 * (ul + ur) * sin(theta);
d(theta) = r/L * (ur - ul);
END;

interested in a representation of the control system capabilities that highlight the application requirements. For example,
the robot is able to navigate in an environment, which might
be static or populated by moving obstacles (e.g., people). It
might be an open space with wide passages or a small room
crowded with tiny obstacles. The selection of a feature corresponding to an application requirement triggers the automatic selection of features corresponding to robot capabilities. For
example, if the environment is a static and crowded space, the
robot should be configured with a complete (and likely slow)
motion planner; instead, in dynamic environments, a fast and
approximate motion planner is more effective.

Modeling Robotic Resources
A model of the robot embodiment (the kinematic structure,
the device characteristics, and so on) and a model of the
operational environment (the objects to recognize, avoid,
grasp, and so on) are crucial for object manipulation and
navigation. They act as shared resources among multiple
robot functionalities, such as perception, planning, and control. Most algorithms that implement these functionalities
are parametric with respect to the embodiment and world
models. This means that the robot control system can be
configured at deployment time according to the specific
application requirements and can be reused without modification for different tasks and
different robotic systems.
Blumenthal and Bruyninckx
[49] present a textual domainNavigation
specific modeling language for
Local
modeling the structural aspects
Navigation
of the robot embodiment and
Navigation
environment as a Scene Graph,
Strategy
i.e., a description of relevant obTrajectory
Motion
Kinematic
1f1
jects and the relations among
Adaptation
Behavior
Model
them. These relations are orga1f1
1f1
1f1
nized in a Direct Acyclic Graph.
The Scene Graph can express
prior semantic knowledge
Leader
Trajectory
Differential
Reactive Smooth
Omni
DWA
VFH
Follower
Follower
Drive
about a scene, such as the morphology and appearance of an
object (e.g., a table) that the
Figure 5. The HyperFlex Feature Model of the navigation system. DWA: dynamic window
approach; VFH: vector field histogram.
robot should recognize

164

IEEE ROBOTICS & AUTOMATION MAGAZINE

September 2015

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2015