Encouraging Behavioral Diversity in Evolutionary Robotics: An Empirical Study

Evolutionary robotics (ER) aims at automatically designing robots or controllers of robots without having to describe their inner workings. To reach this goal, ER researchers primarily employ phenotypes that can lead to an infinite number of robot behaviors and fitness functions that only reward the achievement of the task—and not how to achieve it. These choices make ER particularly prone to premature convergence. To tackle this problem, several papers recently proposed to explicitly encourage the diversity of the robot behaviors, rather than the diversity of the genotypes as in classic evolutionary optimization. Such an approach avoids the need to compute distances between structures and the pitfalls of the noninjectivity of the phenotype/behavior relation; however, it also introduces new questions: how to compare behavior? should this comparison be task specific? and what is the best way to encourage diversity in this context? In this paper, we review the main published approaches to behavioral diversity and benchmark them in a common framework. We compare each approach on three different tasks and two different genotypes. The results show that fostering behavioral diversity substantially improves the evolutionary process in the investigated experiments, regardless of genotype or task. Among the benchmarked approaches, multi-objective methods were the most efficient and the generic, Hamming-based, behavioral distance was at least as efficient as task specific behavioral metrics.

Download Full-text

Online Discovery of Search Objectives for Test-Based Problems

Evolutionary Computation ◽

10.1162/evco_a_00179 ◽

2017 ◽

Vol 25 (3) ◽

pp. 375-406 ◽

Cited By ~ 8

Author(s):

Paweł Liskowski ◽

Krzysztof Krawiec

Keyword(s):

Evolutionary Algorithms ◽

Search Algorithm ◽

Premature Convergence ◽

Multiple Tests ◽

Multi Objective ◽

Coevolutionary Algorithms ◽

Fitness Functions ◽

Candidate Solution ◽

Complete Order ◽

Theoretical Evidence

In test-based problems, commonly approached with competitive coevolutionary algorithms, the fitness of a candidate solution is determined by the outcomes of its interactions with multiple tests. Usually, fitness is a scalar aggregate of interaction outcomes, and as such imposes a complete order on the candidate solutions. However, passing different tests may require unrelated “skills,” and candidate solutions may vary with respect to such capabilities. In this study, we provide theoretical evidence that scalar fitness, inherently incapable of capturing such differences, is likely to lead to premature convergence. To mitigate this problem, we propose disco, a method that automatically identifies the groups of tests for which the candidate solutions behave similarly and define the above skills. Each such group gives rise to a derived objective, and these objectives together guide the search algorithm in multi-objective fashion. When applied to several well-known test-based problems, the proposed approach significantly outperforms the conventional two-population coevolution. This opens the door to efficient and generic countermeasures to premature convergence for both coevolutionary and evolutionary algorithms applied to problems featuring aggregating fitness functions.

Download Full-text

Augmenting Scalable Communication-Based Role Allocation for a Three-Role Task

INTELIGENCIA ARTIFICIAL ◽

10.4114/intartif.vol22iss64pp152-165 ◽

2020 ◽

Vol 22 (64) ◽

pp. 152-165

Author(s):

Gustavo Martins ◽

Paulo Urbano ◽

Anders Lyhne Christensen

Keyword(s):

Fitness Function ◽

A Priori ◽

Evolutionary Process ◽

Evolutionary Robotics ◽

Cognitive Architectures ◽

Local Conditions ◽

Fitness Functions ◽

Signal Differentiation ◽

Scalable Communication ◽

Role Allocation

In evolutionary robotics role allocation studies, it is common that the role assumed by each robot is strongly associated with specific local conditions, which may compromise scalability and robustness because of the dependency on those conditions. To increase scalability, communication has been proposed as a means for robots to exchange signals that represent roles. This idea was successfully applied to evolve communication-based role allocation for a two-role task. However, it was necessary to reward signal differentiation in the fitness function, which is a serious limitation as it does not generalize to tasks where the number of roles is unknown a priori. In this paper, we show that rewarding signal differentiation is not necessary to evolve communication-based role allocation strategies for the given task, and we improve reported scalability, while requiring less a priori knowledge. Our approach for the two-role task puts fewer constrains on the evolutionary process and enhances the potential of evolving communication-based role allocation for more complex tasks. Furthermore, we conduct experiments for a three-role task where we compare two different cognitive architectures and several fitness functions and we show how scalable controllers might be evolved.

Download Full-text

A Review of Multi-Objective Evolutionary Based Fuzzy Classifiers

Recent Advances in Computer Science and Communications ◽

10.2174/2213275912666190410142052 ◽

2020 ◽

Vol 13 (1) ◽

pp. 77-85

Author(s):

Praveen Kumar Dwivedi ◽

Surya Prakash Tripathi

Keyword(s):

Fuzzy Systems ◽

Evolutionary Process ◽

Fuzzy Rule ◽

Fuzzy Classification ◽

Multi Objective Optimization ◽

Fuzzy Classifier ◽

Rule Based ◽

Trade Off ◽

Multi Objective ◽

Rule Based Systems

Background: Fuzzy systems are employed in several fields like data processing, regression, pattern recognition, classification and management as a result of their characteristic of handling uncertainty and explaining the feature of the advanced system while not involving a particular mathematical model. Fuzzy rule-based systems (FRBS) or fuzzy rule-based classifiers (mainly designed for classification purpose) are primarily the fuzzy systems that consist of a group of fuzzy logical rules and these FRBS are unit annexes of ancient rule-based systems, containing the "If-then" rules. During the design of any fuzzy systems, there are two main objectives, interpretability and accuracy, which are conflicting with each another, i.e., improvement in any of those two options causes the decrement in another. This condition is termed as Interpretability –Accuracy Trade-off. To handle this condition, Multi-Objective Evolutionary Algorithms (MOEA) are often applied within the design of fuzzy systems. This paper reviews the approaches to the problem of developing fuzzy systems victimization evolutionary process Multi-Objective Optimization (EMO) algorithms considering ‘Interpretability-Accuracy Trade-off, current research trends and improvement in the design of fuzzy classifier using MOEA in the future scope of authors. Methods: The state-of-the-art review has been conducted for various fuzzy classifier designs, and their optimization is reviewed in terms of multi-objective. Results: This article reviews the different Multi-Objective Optimization (EMO) algorithms in the context of Interpretability -Accuracy tradeoff during fuzzy classification. Conclusion: The evolutionary multi-objective algorithms are being deployed in the development of fuzzy systems. Improvement in the design using these algorithms include issues like higher spatiality, exponentially inhabited solution, I-A tradeoff, interpretability quantification, and describing the ability of the system of the fuzzy domain, etc. The focus of the authors in future is to find out the best evolutionary algorithm of multi-objective nature with efficiency and robustness, which will be applicable for developing the optimized fuzzy system with more accuracy and higher interpretability. More concentration will be on the creation of new metrics or parameters for the measurement of interpretability of fuzzy systems and new processes or methods of EMO for handling I-A tradeoff.

Download Full-text