User-Friendly Robot Environment for Creation of Social Scenarios - Foundations on Natural and Artificial Computation - page 216

Behavior such as letting the robot walk requires already a sensory input, for
instance, it needs to be sure that it is standing. In case the robot is lying down
it first needs to get up. In order to decide which command to execute we need
to get sensory information first.
To do so, a 'RobotSense' module is constructed in a way that it reads out
all relevant sensory information from an a-priori known robot at a fixed, cyclic
update speed, which is set by the user. Reading sensory information from a robot
usually is a relatively trivial task, however, sensory information is obtained at
different frequencies. Images in common can be obtained at a maximum rate
of 30 frames per second, while internal data can be obtained at higher rates,
for instance at 10 or 20 milliseconds rather than the 33 for video. Latter might
form a bottleneck on bandwidth which might in turn lead to latency and a lower
frame rate. Hence, audio, video and single valued information are decoupled, as
illustrated in Figure 3.

a) b)

Fig. 3. a) Global processes and flows of a command generator and a robot module
robot. b) TiViPE implementation of a) using a NAO robot and automated getting up
before walking (in a circle).

A robot needs several seconds to get up, while less is needed for reading
out sensory information. To avoid generating commands that the robot is not
able to complete in time, a moderator process, called 'SendCommand' (see also
Figure 3) is introduced. This process sends a command to the robot if all former
commands are accomplished.
This so-called 'RobotCommand' (for NAO it is NaoRobot) processes the given
commands. Switches to busy mode until it is ready signalling the 'readBoolean'
module that its ready to receive.
This concept makes continuous sensing, and providing the latest command
string on this basis possible without the need of specific knowledge on robot
sensing or commanding. Hence, the focus is shifting to scenario's on basis of
textual commands that are driven by internal information gathered over time
and sensory data, hence one can focus on the 'GenerateCommands' type of
modules.