Page 337 - Introduction to AI Robotics

320
Part II
5. Supports corrections to the map and re-planning. Path planning requires an a
priori map, which may turn out to be seriously wrong. Therefore, a robot
may start out with one map, discover it is incorrect, and need to update
the map and re-plan. Clearly techniques which permit the existing plan
to be repaired rather than be scrapped and computed from scratch are
desirable.

The Impact of Sensor Uncertainty

Since navigation is a fundamental capability of a mobile robot, researchers
have been investigating navigational techniques since the 1960’s. But as was
seen in Part I, it was only since the early 1990’s that robots became afford-
able, and had on-board sensing and reasonable computational power. As
a result, most researchers in navigation were forced to develop techniques
using simulators and assumptions about how real robots would physically
work.
Two of the most pervasive assumptions of these researchers turned out to
be unfortunate in retrospect. First, it was generally assumed that the robot
could localize itself accurately at each update. This assumption was based in
part on another assumption: that sensors would give an accurate represen-
tation of the world. As was seen just with sonars in Ch. 6, this is often not
true. Sensors are always noisy and have vulnerabilities.
Therefore, a robot has to operate in the presence of uncertainty. In the Re-
active Paradigm, the way in which the sensors were coupled with the actua-
tors accepted this uncertainty. If the sonar or IR returned an incorrect range
reading, the robot may appear to start to avoid an imaginary obstacle. How-
ever, the process of moving often eliminated the source of the noisy data,
and soon the robot was back to doing the right thing. Uncertainty becomes
more serious when dealing with map making and localization; therefore a
new wave of techniques has been developed to smooth over sensor noise
and ascertain the correct state of the world. These methods are mathematical
in nature and are covered in Ch. 11.

Navigation and the Robotic Paradigms

The questions posed call to mind deliberation. Planning, just from the name
alone, is deliberative. Map making and localization imply memory and la-
beling specific locations (room, hall, river, canyon); these are symbolic rep-
resentations and so also fit the notion of deliberation from the Hybrid Para-