Evolutionary Robotics: A Survey of Algorithms and Applications

Abstract

This paper reviews the design of autonomous robots that exhibits the given behavior in a given environment by using evolutionary methods. This evolutionary method has been successfully applied to different robots for numerous mannerisms. Evolutionary algorithms consolidate principles from biological population genetics to perform search, optimization, and learning. The control system of the robots is obtained by comparing and improving the various algorithms. An overview of the evolutionary algorithm is given, which explains evolution strategies, evolutionary programming, and genetic algorithms.

Introduction

In recent years, many researchers have pursued overcoming the difficulties of coding mobile robots that have to accomplish the specified task in any possibly changing and unknown circumstances. Due to such difficulties, the impossibility of perceiving the problems the robot will have to solve, and the lack of basic principles upon which human design might rely, these researchers adopted a new method called the evolutionary robotics approach, which is simply known as an automatic design procedure.

The field of robotics offers an infinite problem, which requires an impressive method of solutions. A classic method that has shown its utility on a number of related problems is called Evolutionary Computation [1]

This process calls upon some procedures, such as evolution strategy, evolutionary programming, and genetic algorithm methods. A dedicated fitness function is used to determine the behavior of each individual in the population and to direct the selection properly. In some applications, the evolutionary procedure takes place through changes and progressions performed directly on the robot, and finesses are directly assessed through real-world interactions.

In evolutionary robotics, as in many areas of Artificial Intelligence, there is much exchange between engineering and scientific goals and results. Some of the robots that are often used in the applications are specified here, and they are Khepera, Nao, Tapia, and Pepper.

EVOLUTIONARY ROBOTICS

Evolutionary robotics is a computer-designed method of creating autonomous, intelligent robots with required characteristics based on the principles of Darwin’s theory of evolution. In evolutionary robotics, the robots are treated as structures that will function and develop independently of human instructions[2]

Evolutionary robotics has a target of research within the much larger fields of artificial Intelligence life and completely autonomous robots. One of the primary goals of this is to create automatic methods for developing intelligent, autonomous robot controllers in such a way that it doesn’t require any programming to be given by humans [3]

EVOLUTIONARY ALGORITHMS

The Evolutionary Algorithm is a part extracted from Evolutionary computation[4]. The Evolutionary Algorithm uses mechanisms inspired by biological evolution. Biological evolution has many sub-divisions

• Life and Information Processing
• Meiotic heredity
• Mutations
• Molecular Darwinism [5]

The other three main streams of evolutionary algorithms (EAs) based on the model of natural evolution are:

• Evolution strategies(ESs)
• Evolutionary programming (EP)
• Genetic algorithms (GAs) [6]

EVOLUTION STRATEGIES

Evolution Strategies is a globally used algorithm and is one of the types of Algorithms from the fields of Evolutionary Computations. This is inspired by the theory of evolution by natural selection. This technique is mainly used to enlarge the suitability of the collection of candidate solutions in the context of an aim-oriented function from a domain. This technique is distinguished from other techniques in the aspect of genetic mechanism [7]. Unlike other techniques, this technique is not concerned with a genetic mechanism.

EVOLUTION PROGRAMMING

Evolution programming is a stochastic optimization. It gives priority to the behavioral relationship between parents and their offspring.

This basic programming involves three steps:

1. In this process, it chooses a basic fundamental population at random. The number of solutions in a population is highly compatible with the speed of optimization, but there aren’t any accurate answers available as to how many solutions are relevant and applicable and how many solutions are wasteful.
2. Here, each solution reproduces a new population. Each of
   these next generations is mutated according to the assigned mutation types, ranging from minor to major mutation types. The mutation is based and concluded on the functional change implemented on the parents.
3. In each offspring solution, it is estimated by computing its fitness.

Basically, a stochastic event is held to determine N solutions to be maintained for the population of solutions, even though it is performed occasionally. There is no requirement that the population size is to be the same. However, only a single offspring can be produced from each parent.

It should be noted that Evolutionary Programming typically does not use any crossover as a genetic operator [8]

GENETIC ALGORITHMS

The genetic algorithm is a method for solving both easy and very difficult optimization issues that are based on natural selection. This process gives biological evolution. A continuous modification of the population of individual solutions is done by the genetic algorithm. At each step, this algorithm selects individuals at random from the present population to be parents and uses them to produce the next generation. Over consecutive and successive generations, the population ‘evolves’ to give an optimal solution. Then we can apply this genetic algorithm to solve various optimization problems that are not suited for the standard optimization algorithms.

The genetic algorithm uses three main types of rules to create offspring, they are:

• Here, the Selection rules select the individuals, called parents, that provide the population to the next generation.
• Crossover rules are used to combine two parents to produce children for the next generation.
• Random changes are applied by mutation rules to the individual parents to form offspring.

The offspring gets the same specifications, features, and changes according to what their parents undergo during this mutation process. [9]

Cite this page