APPLICATION DESIGN ISSUES - Hard Real-Time Computing Systems: Predictable Scheduling Algorithms and Applications - page 410

These basic control strategies are essential for executing autonomous tasks in unknown
conditions, and in fact, they are used in the next level to implement more skilled
actions.

Based on the control strategies developed in the Behavior Level, the Action Level
enhances the robot capability by adding more sophisticated sensory-motor activities,
which can be used at the higher level for carrying out complex tasks in unstructured
environments. Some representative actions developed at this level include (1) the abil-
ity of the robot to follow an unknown object contour, maintaining the end-effector in
contact with the explored surface; (2) the reflex to avoid obstacles, making use of vi-
sual sensors; (3) the ability to adapt the end-effector to the orientation of the object to
be grasped, based on the reaction forces sensed on the wrist; (4) visual tracking, to fol-
low a moving object and keep it at the center of the visual field. Many other different
actions can be easily implemented at this level by using the modules available at the
Behavior Level or directly taking the suited sensory information from the functions at
the Device Level.

Finally, the Application Level is the level at which the user defines the sequence of
robot actions for accomplishing application tasks, such as assembling mechanical
parts, exploring unknown objects, manipulating delicate materials, or catching mov-
ing targets. Note that these tasks, although sophisticated in terms of control, can be
readily implemented thanks to the action primitives included in the lower levels of the
hierarchical control architecture.

11.3.1 EXAMPLES OF REAL-TIME ROBOTICS

APPLICATIONS

In this section we describe a number of robot applications that have been implemented
by using the control architecture presented above. In all the examples, the arm trajec-
tory cannot be pre-computed off-line to accomplish the goal, but it must be continu-
ously replanned based on the current sensory information. As a consequence, these
applications require a predictable real-time support to guarantee a stable behavior of
the robot and meet the specification requirements.