Search

In this study, we investigate the utilization of a multi-objective approach in evolving artificial neural networks (ANNs) for an autonomous mobile robot. The ANN acts as a controller for radio frequency (RF)-localization behavior of a Khepera robot simulated in a 3D physics-based environment. The elitist Pareto-frontier Differential Evolution (PDE) algorithm is used to generate the Pareto optimal sets of ANNs that optimize two conflicting objectives of maximizing the virtual Khepera robot's behavior for homing towards a RF signal source and minimizing the number of hidden neurons used in its ANNs controller. Little work has been done on using the evolutionary multi-objective approach in evolving robot controllers.We mainly demonstrate and verify the evolved controllers' performances and robustness in an obstacle-laden environment. Two obstacles are included in the simulation environment to block the most common paths used for the robot to home in towards the signal source. In the testing phase, the robot's robustness was tested with a different positioning of the obstacles from its original position used during evolution. The testing results show that the controllers were still able to navigate successfully, hence demonstrating the evolved controllers' robustness. This study has thus shown that a multi-objective approach to evolutionary robotics in the form of the elitist PDE-EMO algorithm can be practically used to automatically generate robust controllers for RF-localization behavior in autonomous mobile robots. ©2008 IEEE.

Recently a number of researchers have begun exploring the idea of combining Opposition-Based Learning (OBL) with evolutionary algorithms, reinforcement learning, neural networks, swarm intelligence and simulated annealing. However, an area of research that is still in infancy is the application of the OBL concept to coevolution. Hence, in this chapter, two new opposition-based competitive coevolution algorithms for multiobjective optimization called SPEA2-CE-HOF and SPEA2-CE-KR are discussed. These hybrid algorithms are the combination of Strength Pareto Evolutionary Algorithm 2 (SPEA2) with two types of the competitive fitness strategies, which are the Hall of Fame (HOF) and K-Random Opponents (KR), respectively. The selection of individuals as the opponents in the coevolutionary process strongly implements this opposition-based concept. Scalability tests have been conducted to evaluate and compare both algorithms against the original SPEA2 for seven Deb, Thiele, Laumanns, and Zitzler (DTLZ) test problems with 3 to 5 objectives. The experimental results show clearly that the performance scalability of the opposition-based SPEA2-CE-HOF and SPEA2-CE-KR were significantly better compared to the original non-opposition-based SPEA2 as the number of the objectives becomes higher in terms of the closeness to the true Pareto front, diversity maintenance and the coverage level

Many researchers have shown that the crossover operator is essential for genetic algorithm. Crossover in genetic algorithm works to combine short low-order schemata into high-fitness strings. This paper investigates the influences of the variation of crossover rate in genetic algorithms when applied to bidding strategies in online auctions. The proposed bidding strategy is polynomial in nature in which it will suggest the price to bid at a given time depending on some constraints. The best strategy is discovered by evolving the strategies using the traditional genetic algorithm. In this particular setting, GA has shown promising results in exploring the solution in large space and with little priori information. However, the crossover rate was fixed at 0.6 as suggested by the literature. It cannot be ascertained if this is the best value to use. Hence, the objective of this work is to investigate the effect of varying the crossover rate by observing the performance of the bidding strategy in the online auctions based on the average fitness, success rate and the average payoff. An empirical evaluation on the relative performance of the various crossover rates in genetic algorithm in searching for the most effective strategies in the heuristic decision making framework are discussed in this paper.

The objective of this paper is to investigate whether the performance of the self-adaptive the parameters in 3PDE can Improve the performance for function optimization. In this paper, we firstly propose three new algorithms (3PDE-SACr, 3PDE-SAF and 3PDE-SACrF). The preliminary testing is carried out to compare their performance with 3PDE to determine the best algorithm for the next step to self-adapt the population size. Here, the best algorithm from the preliminary testing will be chosen for the testing on self-adapting the population size in absolute and relative encodings. The preliminary testing showed that 3PDE-SAF performed the best for the first three proposed algorithms. So, 3PDE-SAF is chosen for the self-adaptive population size to test in absolute (3PDE-SAF-Abs) and relative (3PDE-SAF-Rel) encodings and the final result showed that 3PDE-SAF-Rel performed slightly better than all the proposed algorithms in terms of its average performance and its stability. © 2007 IEEE.

Lane detection is an essential component of autonomous mobile robot applications. Any lane detection method has to deal with the varying conditions of the lane and surrounding that the robot would encounter while moving. Lane detection procedure can provide estimates for the position and orientation of the robot within the lane and also can provide a reference system for locating other obstacles in the path of the robot. In this paper we present a method for lane detection in video frames of a camera mounted on top of the mobile robot. Given video input from the camera, the gradient of the current lane in the near field of view are automatically detected. Randomized Hough Transform is used for extracting parametric curves from the images acquired. A priori knowledge of the lane position is assumed for better accuracy of lane detection.

This paper explores a new approach of using a multiobjective evolutionary algorithm (MOEA) to evolve robot controllers in performing phototaxis task while avoiding obstacles or navigating through a maze in a simulated environment, to overcome problems involving more than one objective, where these objectives usually trade-off among each other and are expressed in different units. Experiments were conducted in six sets within a 10% noise environment with different task environment complexities to investigate whether the MOEA is effective for controller synthesis. A simulated Khepera robot is evolved by a Pareto-frontier Differential Evolution (PDE) algorithm, and learned through a 3-layer feed-forward artificial neural network, attempting to simultaneously fulfill two conflicting objectives of maximizing robot phototaxis behavior while minimizing the neural network's hidden neurons by generating a Pareto optimal set of controllers. Results showed that robot controllers could be successfully developed using the MOEA.

This article describes a simulation model in which a multiobjective approach is utilized for evolving an artificial neural networks (ANNs) controller for an autonomous mobile robot. A mobile robot is simulated in a 3D, physics-based environment for the RF-localization behavior. The elitist Pareto-frontier Differential Evolution (PDE) algorithm is used to generate the Pareto optimal set of ANNs that could optimize two objectives in a single run; (1) maximize the mobile robot homing behavior whilst (2) minimize the hidden neurons involved in the feed-forward ANN. At testing phase, the generated controllers are tested with a different environment. The testing environment is different from the conducted evolution environment. Interestingly however, the testing results showed some of the mobile robots are still robust to the environment used. The controllers allowed the robots to home in toward the signal source with different paths. © 2008 IEEE.

Little work has been done on using the evolutionary multi-objective approach in evolving the robot controllers. In this study, a multi-objective approach is utilized in evolving the artificial neural networks (ANNs) for autonomous mobile robot controller. The neural network acts as a controller for radio frequency (RF)-localization behavior of a Khepera robot simulated in a 3D physics-based environment. The Pareto optimal sets of ANNs are generated with elitist Pareto-frontier Differential Evolution (PDE) algorithm. The algorithm used to optimize two conflicting objectives; (1) minimize the virtual Khepera robot's behavior for homing towards a RF signal source and (2) minimize the number of hidden neurons used in its ANNs. In this paper, we demonstrate and verify the evolved controllers' moving performances, tracking performances and robustness in a random RF localization environment. In the testing phase, the robot's tracking performances and robustness were tested with five different positioning of the RF signal source from its original position used during evolution. The testing results showed that the controllers were still able to navigate successfully to track the signal source with least possible used of permitted hidden neurons, hence demonstrating the evolved controllers' robustness and tracking ability.

This paper investigates the utilization of a multiobjective approach for evolving artificial neural networks (ANNs) that act as a controller for radio frequency (RF)-localization behavior of a virtual Khepera robot simulated in a 3D, physics-based environment. The non-elitist and elitist Pareto-frontier Differential Evolution (PDE) algorithm are used to generate the Pareto optimal sets of ANNs that optimize the conflicting objectives of maximizing the virtual Khepera robot's behavior for homing towards a RF signal source and minimizing the number of hidden neurons used in its feedforward ANNs controller. A new fitness function which involved maximizing average wheels speed and detection of the RF signal source is also proposed. The experimentation results showed that the virtual Khepera robot was able to move towards to the target with using only a small number of hidden neurons. Furthermore, the testing results also showed that the success rate for the robot to achieve the signal source was higher when the elitist PDE-EMO algorithm was used. The path analysis of the Pareto controllers elucidated many different behaviors in terms of providing a successful homing behavior for the robot to attain the RF signal source. © 2008 IEEE.

The objective of this paper is to implement a multipronged strategy for generating diversity using non-adaptive, adaptive as well as self-adaptive methods for controlling mutation operations in a real-coded genetic algorithm (RCGA). Currently, one of the state-of-the-art RCGAs for function optimization is called the G3-PCX algorithm. However, its performance for solving multimodal problems is known to be poor compared with its performance for unimodal problems. In G3-PCX, the main problem primarily stems from premature convergence to local rather than global optima due to lack of explorative capabilities of the algorithm. As the G3-PCX algorithm relies completely on crossover for promoting diversity, this paper proposes a multilevel mutation operator to augment the algorithm's capability of escaping local optima. The proposed algorithm is called G3M2 (G3-PCX with Multilevel Mutation) and empirical tests on four benchmark multimodal test functions have shown highly competitive performance. The objective of this paper is to investigate whether the proposed multilevel mutation is able to improve the precision accuracy of G3-PCX in solving multimodal function optimization problems. In three of the four problems, G3M2 outperformed the standard G3-PCX algorithm in terms of solution quality. Thus, the multilevel combination of non-adaptive, adaptive and self-adaptive parameter control strategies into a single paradigm is empirically shown to have beneficial effects for enhancing the effectiveness of the G3-PCX algorithm for solving multimodal optimization problems in terms of solution quality. © 2007 IEEE.

Many have treated mutation operators as a supplement operator in genetic algorithm. Researches have shown that the mutation operator plays an important role in genetic algorithm. This paper investigates the influences of the variation of mutation rate in genetic algorithms when applied to bidding strategies in online auctions. The proposed bidding strategy is polynomial in nature in which it will suggest the price to bid at a given time depending on some constraints. Genetic algorithm has shown promising result in the previous work in this particular setting. However, a fixed mutation rate (Pm 0.02) is used in the algorithm as suggested by the literature. It cannot be ascertained if this is the best value to use. Hence, the objective of this work is to investigate the effect of varying the mutation rate by observing the performance of the bidding strategy in the online auctions based on the average fitness, success rate and the average payoff. An empirical evaluation on the relative performance of the various mutation rates in genetic algorithm in searching for the most effective strategies in the heuristic decision making framework are discussed in this paper. © 2008 IEEE.

This chapter will demonstrate the various robotics applications that can be achieved using evolutionary multi-objective optimization (EMO) techniques. The main objective of this chapter is to demonstrate practical ways of generating simple legged locomotion for simulated robots with two, four and six legs using EMO. The operational performance as well as complexities of the resulting evolved Pareto solutions that act as controllers for these robots will then be analyzed. Additionally, the operational dynamics of these evolved Pareto controllers in noisy and uncertain environments, limb dynamics and effects of using a different underlying EMO algorithm will also be discussed.

For function optimization problems in continuous search spaces, one of the main difficulties currently faced is that of locating high quality solutions. This problem is particularly pertinent for continuous multimodal problems where the quality rather than computational efficiency is more important as a test of the solver's ability to escape local optima and finding solutions near the global optimum. Moreover, this difficulty is further compounded when the function involves large numbers of variables, which translates into a highly deceptive fitness landscape with very large numbers of local optima.

A comparison of six basic Ant Colony Optimization (ACO) in dynamic environment was studied in this paper. Dynamic Traveling Salesman Problem (TSP) will be used as a dynamic environment. A number of cities are swap over time to make the TSP environment dynamic. A pheromone equalization strategy was applied in all the six ACO to react to the change. Three sets of TSP are used in this experiment. The result will show which of the six basic ant algorithms work best in dynamic environment.

Recently, numerous Multiobjective Evolutionary Algorithms (MOEAs) have been presented to solve real life problems. However, a number of issues still remain with regards to MOEAs such as convergence to the true Pareto front as well as scalability to many objective problems rather than just bi-objective problems. The performance of these algorithms maybe augmented by incorporating the coevolutionary concept. Hence, in this paper, a new algorithm for multiobjective optimization called SPEA2-CC is illustrated. SPEA2-CC combines an MOEA, Strength Pareto Evolutionary Algorithm 2 (SPEA2) with Cooperative Coevolution (CC). Scalability tests have been conducted to evaluate and compare the SPEA2-CC against the original SPEA2 for seven DTLZ test problems with a set of objectives (3 to 5 objectives). The results show clearly that the performance scalability of SPEA2-CC was significantly better compared to the original SPEA2 as the number of objectives becomes higher.

The aim of 3-Parents Differential Evolution (3PDE) is to reduce the parental requirement in the original Differential Evolution (DE). The effectiveness of 3PDE has been reported and is considered as a useful algorithm that performed better convergence to optimality. In general, 3PDE is considered as a useful contribution since it has successfully reduced the parental requirement in DE without significant reductions in absolute optimization performance by gaining better average performance as well as stability. The objective of this paper is to investigate whether certain parameters that are self-adapted in 3PDE can enhance its performance for function optimization. Here, we propose three new algorithms to compare against 3PDE for their performance, which included 3SACr, 3SAF, and 3SAFCr. Fifty run were conducted for each 20 well-known benchmark functions to test all proposed algorithms. The experimental results showed that 3SACr performed the best among the other algorithms in terms of its better average performance as well stability.

In this paper, we propose the integration between Strength Pareto Evolutionary Algorithm 2 (SPEA2) with two types of coevolution concept, Competitive Coevolution (CE) and Cooperative Coevolution (CC), to solve 3 dimensional multiobjective optimization problems. The resulting algorithms are referred to as Strength Pareto Evolutionary Algorithm 2 with Competitive Coevolution (SPEA2-CE) and Strength Pareto Evolutionary Algorithm 2 with Cooperative Coevolution (SPEA2-CC). The main objective of this paper is to compare competitive against cooperative coevolution to ascertain which coevolutionary approach is preferable for multiobjective optimization. The competitive coevolution will be implemented with K-Random Opponents strategy. The performances of SPEA2-CE and SPEA2-CC for solving triobjective problems using the DTLZ suite of test problems are presented. The results show that the cooperative approach far outperforms the competitive approach when used to augment SPEA2 for tri-objective optimization in terms of all the metrics (generational distance, spacing and coverage). © Springer-Verlag Berlin Heidelberg 2007.

Although a number of multi-objective evolutionary algorithms (MOEAs) have been proposed over the last two decades, very few studies have utilized MOEAs for game agent synthesis. Recently, we have suggested a co-evolutionary implementation using the Pareto evolutionary programming (PEP) algorithm. This paper describes a series of experiments using PEP for evolving artificial neural networks (ANNs) that act as game-playing agents. Three systems are compared: (i) a canonical PEP system, (ii) a co-evolving PEP system (PCEP) with 3 different setups, and (iii) a co-evolving PEP system that uses an archive (PCEP-A) with 3 different setups. The aim of this study is to provide insights on the effects of including co-evolutionary techniques on a MOEA by investigating and comparing these 3 different approaches in evolving intelligent agents as both first and second players in a deterministic zero-sum board game. The results indicate that the canonical PEP system outperformed both co-evolutionary PEP systems as it was able to evolve ANN agents with higher quality game-playing performance as both first and second game players. Hence, this study shows that a canonical MOEA without co-evolution is desirable for the synthesis of cognitive game AI agents

Artificial neural networks (ANN) have demonstrated good predictive performance in a wide variety of real world problems. However, there are strong arguments as to why ANNs are insufficient for data mining. The arguments are the poor comprehensibility of the learned ANNs, which is the inability to represent the learned knowledge in an understandable way to the users. In this paper, Artificial Neural Network Tree (ANNT), i.e. ANN training preceded by Decision Tree rules extraction method, is presented to overcome the comprehensibility problem of ANN. Experimental results on three data sets show that the proposed algorithm generates rules that are better than C4.5. This paper provides an evaluation of the proposed method in terms of accuracy, comprehensibility and fidelity.

In this paper, our objective is to conduct comprehensive tests for competitive coevolution using an evolutionary multiobjective algorithm for 3 dimensional problems. This competitive coevolution will be implemented with K-Random Opponents strategy. A new algorithm which integrates Competitive Coevolution (CE) and the Strength Pareto Evolutionary Algorithm 2 (SPEA2) is proposed to achieve this objective. The resulting algorithm is referred to as the Strength Pareto Evolutionary Algorithm 2 with Competitive Coevolution (SPEA2-CE). The performance between SPEA2-CE is compared against SPEA2 to solve problems with each having three objectives using DTLZ suite of test problems. In general, the results show that the SPEA2-CE with KRandom Opponents performed well for the generational distance and coverage but performed less favorably for spacing. © 2007 IEEE.

The Pareto-based Differential Evolution (PDE) algorithm is one of the current state-of-the-art Multi-objective Evolutionary Algorithms (MOEAs). This paper describes a series of experiments using PDE for evolving artificial neural networks (ANNs) that act as game-playing agents. Three systems are compared: (i) a canonical PDE system, (ii) a co-evolving PDE system (PCDE) with 3 different setups, and (iii) a co-evolving PDE system that uses an archive (PCDE-A) with 3 different setups. The aim of this study is to provide insights on the effects of including co-evolutionary techniques on a well-known MOEA by investigating and comparing these 3 different approaches in evolving intelligent agents as both first and second players in a deterministic zero-sum board game. The results indicate that the canonical PDE system outperformed both co-evolutionary PDE systems as it was able to evolve ANN agents with higher quality game-playing performance as both first and second game players. Hence, this study shows that a canonical MOEA without co-evolution is desirable for the synthesis of cognitive game AI agents. © Springer-Verlag Berlin Heidelberg 2007.

One of the most efficient Real-Coded Genetic Algorithms (RCGAs) for function optimization currently is the G3-PCX (Parent-Centric Crossover) algorithm. However, its performance for solving complex multimodal problems with highly deceptive fitness landscapes is known to be poor compared with its performance for unimodal problems. The problem primarily stems from premature convergence to local rather than global optima due to lack of explorative capabilities of the algorithm. In this study, a hybrid approach combining both adaptive as well as non-adaptive mutation in a cooperative manner is investigated in the hope of improving the explorative capabilities of the G3-PCX algorithm for solving complex multimodal problems. The proposed algorithm is called G3HM (G3-PCX with Hybrid Mutation) and empirical tests on four benchmark complex multimodal test problems have shown highly competitive performance. In two of the four problems, G3HM dramatically outperformed the standard G3-PCX algorithm in terms of solution quality. Thus, the concept of combining adaptive with non-adaptive mutation as a hybrid genetic operator is shown to have beneficial effects for cooperatively exploring complex multimodal search spaces.

The integration between Strength Pareto Evolutionary Algorithm 2 (SPEA2) and Competitive Coevolution (CE) concept is presented in this paper, a strategy for solving an optimization problem with three scalable objectives. The Hall of Fame (HOF) competitive fitness function is used to implement the CE. This proposed algorithm referred to as SPEA2-CE-HOF. The performance between SPEA2-CE-HOF is compared against original SPEA2 in solving problems in the DTLZ suite having three to five objectives. The results showed that the SPEA2-CE-HOF performed better than SPEA2 in most of the DTLZ test problems for the generational distance. However the proposed algorithm performed average for the coverage metric. ©2007 IEEE.

This paper compares the implementation of the non-adaptive, adaptive, and self-adaptive co-evolution for evolving artificial neural networks (ANNs) that act as game players for the game of Tic-Tac-Toe (TTT). The objective of this study is to investigate and empirically compare these three different approaches for tuning strategy parameters' in co-evolutionary algorithms in evolving the ANN game-playing agents. The results indicate that the non-adaptive and adaptive co-evolution systems performed better than the self-adaptive co-evolution system when suitable strategy parameters were utilized. The adaptive co-evolution system was also found to possess higher evolutionary stability compared to the other systems and was also successful in synthesizing ANNs with high TTT playing strength both as the first as well as second players

This paper presents an evolving swarm system of flying agents simulated as a collective intelligence within the Breve auto-constructive artificial life environment. The behavior of each agent is governed by genetically evolved program codes expressed in the Push programming language. There are two objectives in this paper, that is to investigate the effects of firstly code size limit and secondly two different versions of the Push genetic programming language on the auto-constructive evolution of artificial life. We investigated these genetic programming code elements on reproductive competence using a measure based on the self-sustainability of the population. Self-sustainability is the point in time when the current population's agents are able to reproduce enough offspring to maintain the minimum population size without any new agents being randomly injected from the system. From the results, we found that the Push2 implementation showed slightly better evolvability than Push3 in terms of achieving self-sufficiency. In terms of code size limit, the reproductive competence of the collective swarm was affected quite significantly at certain parameter settings.

The Generalized Generation Gap (G3) algorithm is one of the most efficient and effective state-of-the-art real-coded genetic algorithms (RCGAs) for unconstrained global optimization. However, its performance on multimodal optimization problems is known to be poor compared to unimodal optimization problems. The G3 algorithm currently relies on crossover operations only. The objective of this paper is to augment the G3 algorithm with adaptive mutation operations which are dynamically activated according to some explicit feedback during the evolutionary optimization process in order to improve its performance for solving multimodal optimization problems. The performance of the enhanced algorithm is compared with its original version based on the global optimization accuracy and the evolutionary dynamics of the optimization process. The proposed algorithm is tested using five benchmark test problems with highly deceptive fitness landscapes. It was found that the performance of the G3 algorithm with adaptive mutation improved significantly in two of the five test problems. In one of these test problems, no optimal solutions could be found previously by the G3 algorithm but can now be solved by the proposed G3 algorithm with augmented adaptive mutation operations.

Self-adaptation in evolutionary computation refers to the encoding of parameters into tire chromosome to allow for the self-organization process to act on the parameters in addition to the design variables. This paper investigates the feasibility of introducing a self-adaptive mutation operator into a real-coded evolutionary algorithm called the Generalized Generation Gap (G3) algorithm. G3 is currently one of the most efficient as well as effective state-of-the-art real-coded genetic algorithms (RCGAs) but the drawback is that its performance on multimodal optimization problems is known to be poor compared to unimodal optimization problems. In this research, our objective is to introduce a self-adaptive mutation operator into G3, of which the mutation decision parameter is evolved along with the search variables during the evolutionary optimization process. The proposed algorithm is tested using four well-known multimodal benchmark test problems with many local optima surrounding their global optimum. It was found that the performance of the modified G3 algorithm with self-adaptive mutation outperformed the original G3 algorithm in two out of the four test problems in terms of the solution precision achieved.

Proper diagnosis, classification and prediction of diabetes are essential due to the increasing prevalence ofthe disease and the increasing cost to control it. Appropriate discovery ofknowledge from historical data for this disease would be a valuable tool for clinical researchers. The main purpose of data mining is to gain insight ofthe data, and extract knowledge (inter-relational patterns) from the data. Applying data mining techniques in diabetic data can facilitate systematic analysis. Artificial Neural network (ANN) has already been applied in a variety ofdomains with remarkable success. However, it has not has been well utilized in data mining because of he "black box" nature. In this paper we present a method of using ANN in data mining and overcoming the "black box" nature using Decision Tree.

This paper investigates the use of a multi-objective approach for evolving artificial neural networks that act as controllers for the legged locomotion of a quadrupedal robot simulated in a 3-dimensional, physics-based environment. The Pareto-frontier Differential Evolution (PDE) algorithm is used to generate a Pareto optimal set of artificial neural networks that optimize the conflicting objectives of maximizing locomotion behavior and minimizing neural network complexity. In this study, insights are provided on how the controller generates the emergent walking behavior in the creature by analyzing the evolved artificial neural networks in operation. A comparison between Pareto optimal controllers showed that ANNs with varying numbers of hidden units resulted in noticeably different locomotion behaviors. It was also found that a much higher level of sensory-motor coordination was present in the best evolved controller.

Although the Differential Evolution (DE) algorithm has been shown to be a simple yet powerful evolutionary algorithm for optimizing continuous functions, users are still faced with the problem of preliminary testing and hand-tuning of the evolutionary parameters prior to commencing the actual optimization process. As a solution, self-adaptation has been found to be highly beneficial in automatically and dynamically adjusting evolutionary parameters such as crossover rates and mutation rates. In this paper, we present a first attempt at self-adapting the population size parameter in addition to self-adapting crossover and mutation rates. Firstly, our main objective is to demonstrate the feasibility of self-adapting the population size parameter in DE. Using the F1-F5 benchmark test problems proposed by De Jong, we showed that DE with self-adaptive populations produced highly competitive results compared to a conventional DE algorithm with static populations. In addition to reducing the number of parameters used in DE, the proposed algorithm actually outperformed the conventional DE algorithm for one of the test problems. It was also found that that an absolute encoding methodology for self-adapting population size in DE produced results with greater optimization reliability compared to a relative encoding methodology.

This study explores the use of a multi-objective evolutionary algorithm for the automatic synthesis of neural controllers for the quadrupedal locomotion of an artificial creature in a 3-dimensional, physics-based environment. The Paretofrontier Differential Evolution (PDE) algorithm is used to generate a Pareto optimal set of artificial neural networks that optimize the conflicting objectives of maximizing locomotion behavior and minimizing neural network complexity. The focus of this artificial life experiment is to firstly evolve embodied locomotion controllers for a physically simulated quadrupedal creature under terrestrial conditions (i.e. simulating Earth's gravity) and then to investigate the performance of the best evolved controller in this physically simulated creature under different extraterrestrial environments (i.e. simulating gravity on planets other than Earth). It was found that under all extraterrestrial conditions the artificial creature was still able to perform the required locomotion task while in the worst case, some minimal locomotion behavior was still achieved.

Recently, there has been a lot of interest in evolving controllers for both physically simulated creatures as well as for real physical robots. However, a range of different ANN architectures are used for controller evolution, and, in the majority of the work conducted, the choice of the architecture used is made arbitrarily. No fitness landscape analysis was provided for the underlying fitness landscape of the controller's search space. As such, the literature remains largely inconclusive as to which ANN architecture provides the most efficient and effective space for searching the range of possible controllers through evolutionary methods. This represents the motivation for this paper where we compare the search space for four different types of ANN architecture for controller evolution through an information-theoretic analysis of the fitness landscape associated with each type of architecture.

This chapter investigates the use of a multi-objective approach for evolving artificial neural networks that act as controllers for the legged locomotion of a 3-dimensional, artificial quadruped creature simulated in a physics-based environment. The Pareto-frontier Differential Evolution (PDE) algorithm is used to generate a Pareto optimal set of artificial neural networks that optimizes the conflicting objectives of maximizing locomotion behavior and minimizing neural network complexity. The evolutionary and operational dynamics of controller evolution is analyzed to provide an insight into how the best controller emerges from the artificial evolution and how it generates the emergent walking behavior in the creature. A comparison between Pareto optimal controllers showed that artificial neural networks (ANNs) with varying numbers of hidden units resulted in noticeably different locomotion behaviors. We also found that a much higher level of sensory-motor coordination was present in the best evolved controller. Finally we investigated the effects of environmental, morphological and nervous system changes on the artificial creature's behavior and found that certain changes are detrimental to the creature's locomotion capability.

The automatic synthesis of embodied creatures through artificial evolution has becomea key area of research in robotics, artificial life and the cognitive sciences.However, the research has mainly focused on genetic encodings and fitness functions.Considerably less has been said about the role of controllers and how theyaffect the evolution of morphologies and behaviors in artificial creatures. Furthermore,the evolutionary algorithms used to evolve the controllers and morphologiesare pre-dominantly based on a single objective or a weighted combination of multipleobjectives, and a large majority of the behaviors evolved are for wheeled orabstract artifacts.In this thesis, we present a systematic study of evolving artificial neural network(ANN) controllers for the legged locomotion of embodied organisms. A virtual butphysically accurate world is used to simulate the evolution of locomotion behaviorin a quadruped creature. An algorithm using a self-adaptive Pareto multi-objectiveevolutionary optimization approach is developed.The experiments are designed to address five research aims investigating: (1) thesearch space characteristics associated with four classes of ANNs with different connectivitytypes, (2) the effect of selection pressure from a self-adaptive Pareto approachon the nature of the locomotion behavior and capacity (VC-dimension) ofthe ANN controller generated, (3) the effciency of the proposed approach againstmore conventional methods of evolutionary optimization in terms of computationalcost and quality of solutions, (4) a multi-objective approach towards the comparisonof evolved creature complexities, (5) the impact of relaxing certain morphologicalconstraints on evolving locomotion controllers.The results showed that: (1) the search space is highly heterogeneous with bothrugged and smooth landscape regions, (2) pure reactive controllers not requiringany hidden layer transformations were able to produce sufficiently good legged locomotion,(3) the propos

Evolutionary methods are now commonly used to automatically generate autonomous controllers for physical robots as well as for virtually embodied organisms. Although it is generally accepted that some amount of redundancy may result from using an evolutionary approach, few studies have focused on empirically testing the actual amount of redundancy that is present in controllers generated using artificial evolutionary systems. Network redundancy in certain application domains such as defence, space, and safeguarding, is unacceptable as it puts the reliability of the system at great risk. Thus, our aim in this paper is to test and compare the redundancies of artificial neural network (ANN) controllers that are evolved for a quadrupedal robot using four different evolutionary methodologies. Our results showed that the least amount of redundancy was generated using a self-adaptive Pareto evolutionary multi-objective optimization (EMO) algorithm compared to the more commonly used single-objective evolutionary algorithm (EA) and weighted sum EMO algorithm. Finally, self-adaptation was found to be highly beneficial in reducing redundancy when compared against a hand-tuned Pareto EMO algorithm.