Fusion in Robotics - Information Fusion in Signal and Image Processing - page 51

- the robot is reaching the end of the hallway;
- the vehicle is being overtaken.
The situation is characterized by symbolic data evaluated from usually numeri-
cal data obtained from the sensors. This evaluation requires the use of models that
describe the symbols based on their numerical values:
- geometric description of the pin;
- meaning of “close”;
- geometric description of a hallway;
- meaning of “reaching the end of the hallway”;
- localization of a vehicle in the lanes;
- meaning of “being overtaken”.
The confidence that can be assigned to the various symbolic assertions depends on
the quality of the measurements (accuracy, reliability) and of the symbol description
models.
4.4. Data fusion mechanisms
Data fusion in robotics relies on the use of traditional tools (probabilities, crisp or
fuzzy sets, belief theory). Applications in robotics are different from other applica-
tions, especially when it comes to the structure of the fusion and more generally of the
perception.
First of all, we have to properly identify the objectives of the fusion in order for
them to be consistent with the mission the robot has to achieve. For example, when the
robot is located far away from the elements that comprise its environment, no effort
will be made to precisely localize them since they do not pose an immediate threat.
However, we need to be sure that the space chosen for movement is clear of obstacles.
The objectives of perception can change over time, requiring that the fusion system,
as well as the sensors and the processing algorithms, are able to adapt to the situation.
The aspect of time is fundamental in robotics. Many fusion algorithms use evolu-
tionary models such as differential equations, recursive equations, graphs and sequen-
tial logic rules. It is then possible to fuse the knowledge acquired previously and reg-
istered with the present moment with newly acquired information. Some methods also
use single or multiple target tracking algorithms [BAR 88, KOK 94, NIM 00].
Finally, the security aspect has to be taken into account because the information
obtained from the fusion is generally used for moving the robot. If it is erroneous or not
accurate enough, it can lead to the robot colliding with its environment. Therefore, it