INDIVIDUAL LEARNING POST-EVALUATIONS (SEEDING CHROMOSOMES)

The following tables show the evaluation results of the 3 seeding chromosomes, each has been used to create 4 different generations of 80 individuals, used as starting point for the evolution of social learning behaviour (see section 7.2 of the paper for more details). Columns labelled with Si indicates the different kind of sets of 8 trials: from S1 with sound played the first trial in the set to S8, with sound played the last one. After 8 trials the controller is reset (neuron's states set to 0) and the robots forgets what learned. A phototaxis trial is successful if the robot reaches a distance from the light smaller than 5cm, antiphototaxis trials are successful when the robt get farther than a time and an half the initial distance. Column S indicates the success rate, E1 and E2, indicate respectively the type 1 error rate (the robot should do phototaxis but doesn't) and the type 2 error rate (the robot should perform antiphototaxis but follows the light). Finally F and R indicate the robot relative orientation with respect to the light source: F indicates situations in wich the robot starts by facing the light, B indicates that the light is behind the robot when the trial starts. For more details see section 8.1 of the paper. Each set Si has been repeated 1000 times, so for each set the robot goes through 8000 trials.

Demonstrator 1

Demonstrator 2

Demonstrator 3

Top

SECOND EVOLUTIONARY PHASE

The following graphs show the fitness score of the best individuals of all the second phase evolutionary runs (see section 7.2 of the paper). Each plot refers to the 4 evolutionary runs generated using the same seeding chromosome.

Evolutionary runs 1-4
Generated from Demonstrator 1

Evolutionary runs 5-8
Generated from Demonstrator 2

Evolutionary runs 9-12
Generated from Demonstrator 3

Top

INDIVIDUAL LEARNING EVALUATION (LEARNERS)

The following tables show the post-evaluation results of all the 12 learners, in the individual learning task. The post-evaluation procedure and the meanings of the labels are described here. Data about learners 1, 2, 3, and 4 can be found also in the paper (section 8.1).

Learner 1
(best individual of evolutionary run 1)

Learner 2
(best individual of evolutionary run 2)

Learner 3
(best individual of evolutionary run 3)

Learner 4
(best individual of evolutionary run 4)

Learner 5
(best individual of evolutionary run 5)

Learner 6
(best individual of evolutionary run 6)

Learner 7
(best individual of evolutionary run 7)

Learner 8
(best individual of evolutionary run 8)

Learner 9
(best individual of evolutionary run 9)

Learner 10
(best individual of evolutionary run 10)

Learner 11
(best individual of evolutionary run 11)

Learner 12
(best individual of evolutionary run 12)

Top

SOCIAL LEARNING EVALUATION (LEARNERS)

The following tables show the post-evaluation results of the 12 learners, in the social learning task. The post evaluation procedure works as follow: initially the learner and the demonstrator are placed in the arena togheter and the demo-trial is run. After the removal of the demonstrator the learner performs 8 copy-trials, after which the network is reset (neurons' states set to 0) and the robot forgets what learnt. In the demo-trials the demonstrator can be positioned on the left of the learner (rows labelled with L) or on its right (rows labelled with R). The light can be positioned in front of the robots (columns labelled with F) or behind them (columns labelled with B). Labels Pi and Ai indicate the number (i) of the copy-trial in the sequence of 8, when the demonstator performed phototaxis (labels Pi) or antiphototaxis (labels Ai). With this convention P1 indicates the first copy-trial following a demo trial in which the demonstrator performed phototaxis, P2 the second one and so on till P8. The same for the antiphototaxis behavior. A phototaxis trial is successful if the robot reaches a distance from the light smaller than 5cm, antiphototaxis trials are successful when the robt get farther than a time and an half the initial distance. Each cell of the table shows the error rate for the corresponding setup. Each individual has been evaluated 1000 times (1000 demo-trials and 8000 copy-trials) for each condition. More details can be found in section 8.2 of the paper.

Learner 1
(phototaxis performances)

Learner 1
(antiphototaxis performances)

Learner 2
(phototaxis performances)

Learner 2
(antiphototaxis performances)

Learner 3
(phototaxis performances)

Learner 3
(antiphototaxis performances)

Learner 4
(phototaxis performances)

Learner 4
(antiphototaxis performances)

Learner 5
(phototaxis performances)

Learner 5
(antiphototaxis performances)

Learner 6
(phototaxis performances)

Learner 6
(antiphototaxis performances)

Learner 7
(phototaxis performances)

Learner 7
(antiphototaxis performances)

Learner 8
(phototaxis performances)

Learner 8
(antiphototaxis performances)

Learner 9
(phototaxis performances)

Learner 9
(antiphototaxis performances)

Learner 10
(phototaxis performances)

Learner 10
(antiphototaxis performances)

Learner 11
(phototaxis performances)

Learner 11
(antiphototaxis performances)

Learner 12
(phototaxis performances)

Learner 12
(antiphototaxis performances)

Top