Specifications

10
perform this operation. The proposed algorithm is intended to create a logical path for a
unmanned vehicle to take by use of sensors, videos and imaging data. ‘Our focus is to
develop a object avoidance algorithm called SmartAvoidT that extracts multiple
objects/targets out of video/imagery data, establishes individual tracks for each object
and maps a path around each object to avoid collisions.’ (Tunnel D 2004). This
algorithm would then be implemented into the vehicles navigation system in order to
follow the calculated path. The algorithm will be designed to work in all weather
conditions such as day, night, rain, smoke or any other condition.
Koren Ward suggests that a successful method of controlling object avoidance is for the
robot to ‘learn’ methods and trends associated with traversing different situations. This
concept still utilises sensors and logic however has another much more complex
learning ability. The unit will record its previous experience with regard to the readings
from its sensors and so will be able to make a decision based on what occurred last
time. For example if sensor 1 was blocked and sensor 2 was clear then it will perform
in the same way as it did last time as long as last time encountered no errors. If errors
occurred it will try a different method and compare the results. This is a very complex
concept however it would most likely produce the least amount of errors in the long run.
A basic logic system may encounter the same error repeatedly, whereas this use of a
fuzzy logic learning system should prevent this occurrence.
While there are several methods of controlling a robot to avoid objects many are much
too complicated for the general purpose of this dissertation. This dissertation aims to
create a base model of an autonomous robot and so the programming of object
avoidance will be kept to a simple level. The concept described by the technology
student above is the initial concept decided upon for this dissertation. Giving the robot
a set of hard coded rules once it reaches an object should give satisfactory performance
while maintaining a simpler approach.
The coded instructions might look something close to the following:
START Drive in a straight line until interrupt from sensor occurs
Is left sensor clear?