Page 33 - Dynamic Vision for Perception and Control of Motion

1.5 What Type of Vision System Is Most Adequate? 17

one might say); if it can be invoked in corresponding situations in the future, it will
help to better control one’s behavior in similar cases (see Chapter 3).

Past Short - term
memory expec- Future
tations
history

Details of internal S e n s i n g Sparse
Point
representations State ‘here long–term
history of and expecta-
objects now’ tions

Time P r e- Time
d i c-
Abstracted experience
• Model elements for inter- control t i-
pretation of sensor data o n
• Goal functions
• Control mode actuation
• Instantiation control
Long–term memory
Extended presence
Figure 1.6. Symbolic representation of the interactions between the ‘mental-’ and the ‘real
world’ (point ‘here and now’) in closed-loop form

Intelligence, thus, is defined as allowing deep understanding of processes and
the way the “own” body may take advantage of this. Since proper reactions depend
on the situation encountered, recognizing situations early and correctly and know-
ing what to do in these cases (decision-making) is at the core of intelligence. In the
sense of steady learning, all resulting actions are monitored and exploited to im-
prove the internal representations for better use in the future. Figure 1.6 shows a
symbolic representation of the overall interaction between the (individual) “mental
world” as data manipulation activity in a prediction-error feedback loop. It spans
part of the time axis (horizontal line) and the “real world” represented by the spa-
tial point “here” (where the sensors are). The spatial point “here”, with its local en-
vironment, and the temporal point “now”, where the interaction of the subject with
the real world takes place, is the only 4-D point for the autonomous system to
make real-world experience. All interactions with the world take place “here and
now” (see central box). The rest of the world, its extensions in space and time, are
individual constructs in the “mental world” to “make sense” of the sensor data
stream and its invariance properties observed individually, and as a social endeavor
between agents capable of proper information exchange.
The widely varying interpretations of similar events in different human cultures
are an indication of the wide variety of relatively stable interpretation systems pos-
sible. Biological systems had to start from scratch; social groups were content with
interpretations, which allowed them to adjust their lives correspondingly. Inconsis-