Abstract:
Autonomous behavior-based mobile robots react directly to sensor information
from the environment. Their behavior-based control system, which interacts with the
environment is interesting and attractive. For the robot to avoid obstacles in the
environment, it autonomously executes actions based on the data from the sensors. It is
challenging to successfully solve problems in autonomous navigation in real time and
make the mobile robot react to changes in its environment. One of the challenges as far
as this study is concerned is selecting the suitable programming language to cover the
formulation of the behaviors. Another challenge is which to choose of two promising
solutions: the fuzzy logic-based solution or the neural networks solution?
This thesis presents a fuzzy logic-based behavior system that includes the fuzzy
sets and the behaviors which are built on its basis. The behaviors that are designed in
fuzzy logic are goal-seeking behavior on one hand and obstacle avoidance behavior on
the other hand. For this purpose suitable membership functions for inputs and outputs
are used.
For comparison purposes a neural network-based goal-seeking and obstacle avoiding
behaviors using back-propagation algorithm in a sort of supervised learning are
designed and computer simulated.
Also a real mobile robot is found to succeed in testing code for wandering
behavior and for avoiding obstacles behavior. Results are also obtained by running
simulations which show robot reaching the goal in neural networks and fuzzy logic
paradigms.
Results also show that futility back-propagation algorithm is capable of
demonstrating goal seeking behavior, but it requires a number of trials. These trials,
include experimenting with number of layers, number of neurons per layer, and initial
weights. In fuzzy logic the behaviors have been designed depending on analysis of the
robot movement and actions. In addition, fuzzy rules have been used to represent the
knowledge base for these applications.
