A Biologically Inspired Autonomous Robot Control Based on Behavioural Coordination in Evolutionary Robotics - Advancing Artificial Intelligence Through Biological Process Applications - page 118

Figure 5. Behaviours coordination to yield a more complex one in the process known as scaling-up

Table 1. Specifications parameters for the experimental tests of the evolutionary approach to path gen-
eration in an autonomous robot

8 plus 1 additional input

Number of input neurons

Number of hidden neurons 6

Number of output neurons 2

10

Number of runs

Number of iterations per run 200

300 Number of generations

Number of objects in the
environment

0 (phototaxis|learning) and 10 (obstacle
avoidance)

random Initial population

[-1; 1]

Synaptic weight range

[0; 0.01]

Initial range of synaptic weight

Selection percentage

50%

5% Mutation rate

Light sensors range [0; 1]

Sensors proximity range [0; 1]

Type of activation function sigmoid

Type of mutation Uniform

Type of selection Ranking

To implement the neurocontrollers we used a
simulator of the Khepera © robot called YAKS. The
tests consisted of the robot moving on a labyrinth-
like environment. Since in reactive path generation
security is essential, it is possible to distinguish
two situations to analyze: (a) if an obstacle is de-
tected in the robot proximity (sides or front), the
obstacle avoidance behaviour has priority over the
attraction to the goal behaviour; and (b) the actua-
tors' action arises as a combination between the
obstacle avoidance and the attraction toward the
objective. Therefore, two possible situations are
necessary to study: (1) convex obstacle avoidance;

(2) concave obstacle avoidance. Robustness of the
final controller with regard to sensor imperfections
(e.g., robustness to environmental variations and
morphologic variations) was also analyzed. The
obtained results showed that the controller was
capable of overcoming environmental variations,
but no morphologic variations in most tests.
Once behaviours were implemented, tests
were carried out on a real robot Khepera © . The
environment consisted of a 105x105 cm. arena
delimited by walls, with different internal walls of
white plastic and a compact green rubber floor. A
light source equipped with a bulb of 75 Watts was