Image Processing Reference
In the course of exploration, the robot's mission is to maneuver inside the environ-
ment in order to construct a map of it. This map may have different semantic levels,
cells dividing the environment and which have to be recognized as occupied or empty,
or even be used for recognizing and locating objects with more complex structures,
such as a hole in a mechanical structure in which a pin has to be inserted, a vehicle on
a road, or a door in a room.
In the course of localization, the robot has at its disposal a map of the environment
that generally corresponds to reality in terms of accuracy (objects that are displaced
or not well positioned), but also of reliability: the existence of an object that does not
appear in the map (false positive) or the absence of an object listed on the map (false
negative). Sometimes, the environment is comprised of objects that can also move in
the environment. In that case, it is obviously impossible to have a map beforehand
representing the environment at every instant. However, if evolutionary models of
these objects are available, as well as their states at a given time, it is possible to
predict the map's state in a “near” future, i.e. in a short time horizon compared to the
time constant of the observed systems.
In the course of the action, based on the knowledge of its current state and of the
objective it has to accomplish, the robot generates a trajectory that is feasible in its
environment, in other words a control to transmit to its actuators and the observations
needed to “properly” proceed with this trajectory [KOS 93].
4.2. Specific features of fusion in robotics
4.2.1. Constraints on the perception system
By definition, the robot must maneuver and operate inside its environment, which
leads to risks. Therefore, it is necessary for the information obtained with the percep-
tion system to be accurate and reliable enough to ensure the safety of the robot, of the
environment and of the other users in this environment, particularly human beings.
Sensors are embedded in the robot and their numbers are therefore limited, as well
as their performances in time and in space, in size and in energy resources. As a result,
the data obtained from the sensors is necessarily limited and the lack of information
is what is dealt with, rather than redundancy.
4.2.2. Proprioceptive and exteroceptive sensors
Just as a human being needs to know its own state before considering a movement
(sitting, standing, carrying weight, off balance), it is necessary to know a certain num-
ber of parameters regarding the robot's static and dynamic state. In order to do this,