Chapter 6

Question 1

Evolutionary robotics

Answer 1

Generating adaptive robots that evolve through a process analogous to natural evolution

Question 2

Genotypes

Answer 2

The encoding of the characteristics of the robots

Question 3

Phenotypes

Answer 3

The characteristics of the robots

Question 4

Fitness

Answer 4

The evaluation of the performance of the robots with a scalar value

Question 5

Offspring

Answer 5

Robotic copies of the best parent genotypes

Question 6

Variations

Answer 6

Differences of genotypes introduced through mutation or other genetic operators

Question 7

Fitness Function

Answer 7

The function that rates the ability of a robot

Question 8

Developmental process

Answer 8

The rules that determine the relationship between the genotype and phenotype

Question 9

Genetic algorithm

Answer 9

Operates on genotypes constituted by vectors of binary values , selects the reproducing genotypes stochastically with a probability proportional to their relative fitness, and generates variations through recombination and mutations.

Question 10

Recombination

Answer 10

Divides the genotype of two reproducing individuals in two or three parts and by pasting some of the parts copied from the first genotype with complimentary parts copied from the second genotype

Question 11

Evolutionary Strategies

Answer 11

Operates on genotypes formed by vectors of real numbers and generates variations by perturbing all real numbers with small randomly generated values

Question 12

Modern evolutionary strategies

Answer 12

An extended version of evolutionary strategies that combine the matrix of random variations and the fitness received by offspring to estimate either the correlation among variations leading to high fitness or the gradient of the expected fitness.

Question 13

OpenAI-ES algorithm

Answer 13

A natural evolutionary strategy that estimates the flgradient of the expected fitness by using uniform distribited variations and updates the centroid of the population through the Adam stochastic gradient optimizer

Question 14

Categorical policies

Answer 14

Common stochastic policies used in the case of discrete action spaces, i.e. in problems where the agent can perform an action within a finite set of possible alternative actions.

Question 15

Diagonal Gaussian Policies

Answer 15

Common stochastic policies used for problems involving a continuous action space.