Mathys C. du Plessis

Differential evolution for dynamic environments with unknown numbers of optima

This paper investigates optimization in dynamic environments where the numbers of optima are unknown or fluctuating. The authors present a novel algorithm, Dynamic Population Differential Evolution (DynPopDE), which is specifically designed for these problems. DynPopDE is a Differential Evolution based multi-population algorithm that dynamically spawns and removes populations as required. The new algorithm is evaluated on an extension of the Moving Peaks Benchmark. Comparisons with other state-of-the-art algorithms indicate that DynPopDE is an effective approach to use when the number of optima in a dynamic problem space is unknown or changing over time.

Towards an artificial neural network-based simulator for behavioural evolution in evolutionary robotics

It is not apparent to employ an Artificial Neural Network (ANN) as simulator structure during the Evolutionary Robotics (ER) process. Consequently, the potential for the use of an ANN in this regard has been investigated in this paper. This simulator Neural Network (NN) was trained to predict changes in the position and orientation resulting from arbitrary motor commands sent to a Lego Mindstorms NXT robot moving in a two-dimensional plane. The proposed NN simulator was employed as an alternative to conventional physics-engine based simulators to evolve simple navigation behaviour in said robot and encouraging results were obtained.

A comparison of neural networks and physics models as motion simulators for simple robotic evolution

Robotic simulators are used extensively in Evolutionary Robotics (ER). Such simulators are typically constructed by considering the governing physics of the robotic system under investigation. Even though such physics-based simulators have seen wide usage in ER, there are some potential challenges involved in their construction and usage. An alternative approach to developing robotic simulators for use in ER, is to sample data directly from the robotic system and construct simulators based solely on this data. The authors have previously shown the viability of this approach by training Artificial Neural Networks (ANNs) to act as simulators in the ER process. It is, however, not known how this approach to simulator construction will compare to physics-based approaches, since a comparative study between ANN-based and physics-based robotic simulators in ER has not yet been conducted. This paper describes such a comparative study. Robotic simulators for the motion of a differentially-steered mobile robot were constructed using both ANN-based and physics-based approaches. These two approaches were then compared by employing each of the developed simulators in the ER process to evolve simple navigation controllers for the experimental robot in simulation. Results obtained indicated that, for the robotic system investigated in this study, ANN-based robotic simulators offer a promising alternative to physics-based simulators.

Self-adaptive competitive differential evolution for dynamic environments

Competitive Differential Evolution (CDE) [1] is a multi-population Differential Evolution (DE) algorithm for optimization in dynamic environments. As such, the control parameters present in DE, are also present in CDE. This paper investigates incorporation of three approaches to self-adapting control parameters into CDE. A comparative evaluation of the performance of each approach is used to determine the most appropriate self-adaptive model for incorporation into CDE. It is shown that self-adapting control parameters does improve the performance of CDE in several instances of benchmark tests. Experimental evidence is presented that indicates that self-adaptive CDE compares favorably with other approaches in the literature.

Incorporating layout managers into an evolutionary programming algorithm to design graphical user interfaces

Designing graphical user interfaces (GUIs) is an arduous task that requires both functional and aesthetic considerations, often relying on documented style guides and design principles. Style guides and principles are mostly prescriptive in terms of what should be included and what should be avoided in interface design, but do not specify the how of the interface design process. Previous research has employed Genetic Algorithms to assist in the design process, but focused more on evolving colour schemes and ordering of user interface (UI) components than on the general layout of the interface. Components were essentially placed in a static grid which resulted in unappealing interfaces. This research seeks to evolve the placement of components on the screen through the use of layout managers. A user guides the evolution process by iteratively selecting promising interfaces from a collection of candidate interfaces. Various constraints are placed on the grouping of components to prevent inappropriate groupings in the UI layout and to reduce the number of selections that the designer has to make. Each candidate UI is encoded in a tree which made crossover operations inappropriate. This resulted in an Evolutionary Programming algorithm being used rather than a Genetic Algorithm. Various mutation operators are discussed. Through this evolutionary process, aesthetically pleasing and functional interfaces can then be created in a reasonable number of iterations.

Concurrent controller and Simulator Neural Network development for a snake-like robot in Evolutionary Robotics

Abstract Evolutionary Robotics (ER) is a field of study that has shown much promise in automating the development of robotic controllers and morphologies. The use of simulators as an alternative to real-world robots is often employed to reduce the time required to develop effective controllers in the ER process. However, the development of adequate simulators is often time-consuming and complex. Simulators are usually constructed from physics models and/or are based on empirically collected data. The vast majority of simulation approaches are based on physics models which can become complex and require specialised knowledge. Alternative simulation approaches that simplify and automate the modelling of real-world phenomena can provide certain advantages over traditional approaches. An alternative simulation approach, such as Artificial Neural Networks (ANNs) that model the real-world phenomena based on empirical data are relatively simple to construct and requires little specialised knowledge. ANN-based simulators are traditionally constructed before the ER process can begin and require the sampling of real-world experimental data. Disadvantages to the traditional approach to ANN-based simulator construction are that the simulator must be created before the ER process can be initiated and a large amount of behavioural data must be collected in order to accurately predict future behaviours. Previous research has successfully demonstrated that ANN-based simulators and controllers can be developed concurrently during the ER process for a simple differentially-steered wheeled robot, while the current research has demonstrated the concurrent approach using a complex snake-like robot. The viability of the concurrent approach was demonstrated on a real-world snake-like robot by performing trajectory planning tasks. Influential factors related to the success of the concurrent approach were also studied by investigating the effects various parameter settings had on success.

Neuroevolution of Inverted Pendulum Control: A Comparative Study of Simulation Techniques

The inverted pendulum control problem is a classical benchmark in control theory. Amongst the approaches to developing control programs for an inverted pendulum, the evolution of Artificial Neural Network (ANN) based controllers has received some attention. The authors have previously shown that Evolutionary Robotics (ER) can successfully be used to evolve inverted pendulum stabilization controllers in simulation and that these controllers can transfer successfully from simulation to real-world robotic hardware. During this process, use was made of robotic simulators constructed from empirically-collected data and based on ANNs. The current work aims to compare this method of simulator construction with the more traditional method of building robotic simulators based on physics equations governing the robotic system under consideration. In order to compare ANN-based and physics-based simulators in the evolution of inverted pendulum controllers, a real-world wheeled inverted pendulum robot was considered. Simulators based on ANNs as well as on a system of ordinary differential equations describing the dynamics of the robot were developed. These two simulation techniques were then compared by using each in the simulation-based evolution of controllers. During the evolution process, the effects of injecting different levels of noise into the simulation was furthermore studied. Encouraging results were obtained, with controllers evolved using ANN-based simulators and realistic levels of noise outperforming those evolved using the physics-based simulators.

Concurrent controller and Simulator Neural Network development for a differentially-steered robot in Evolutionary Robotics

Evolutionary Robotics (ER) strives for the automatic creation of robotic controllers and morphologies. The ER process is normally performed in simulation in order to reduce the time required and robot wear. Simulator development is a time consuming process which requires expert knowledge and must traditionally be completed before the ER process can commence. Traditional simulators have limited accuracy, can be computationally expensive and typically do not account for minor operational differences between physical robots.This research proposes the automatic creation of simulators concurrently with the normal ER process. The simulator is derived from an Artificial Neural Network (ANN) to remove the need for formulating an analytical model for the robot. The ANN simulator is improved concurrently with the ER process through real-world controller evaluations which continuously generate behavioural data. Simultaneously, the ER process is informed by the improving simulator to evolve better controllers which are periodically evaluated in the real-world. Hence, the concurrent processes provide further targeted behavioural data for simulator improvement.The concurrent and real-time creation of both controllers and ANN-based simulators is successfully demonstrated for a differentially-steered mobile robot. Various parameter settings in the proposed algorithm are investigated to determine factors pertinent to the success of the proposed approach. Controllers are developed for trajectory planning using Evolutionary Robotics and a differentially-steered mobile robot.A novel approach is proposed for the concurrent development of controllers and a simulator.Robot behaviours are simulated using Artificial Neural Networks.An extensive parameter comparison study of the proposed approach is conducted.

A Neural Network-based kinematic and light-perception simulator for simple robotic evolution

Current research reveals limited investigations into the use of Artificial Neural Networks (ANNs) as robot simulators. The noise-tolerance and generalization capabilities of ANNs, however, suggest that ANNs could be well-suited to this application. As a result of this observation, a novel technique has been identified wherein ANNs are used as robot simulators. ANNs were employed to simulate the motion dynamics of a mobile robot steered using differential steering, as well as the interaction of two light sensors onboard the robot with a light source in its vicinity. To test the performance of the developed simulators, these simulators were used to evolve a light-approaching robotic control structure in simulation, which was subsequently transferred to the real-world robot. Results indicate that the simulation-evolved controller transferred well from simulation to the real-world robot. It could thus be deduced that ANNs show definite promise as robot simulators.

Evolving Snake Robot Controllers Using Artificial Neural Networks as an Alternative to a Physics-Based Simulator

Traditional simulators can be complex, time-consuming and require specialized knowledge to develop while still being unable to adequately model reality. Artificial Neural Networks (ANNs) can be trained to simulate real-world robots and therefore serve as an alternative to traditional approaches of robot simulation during the Evolutionary Robotics (ER) process. ANN-based simulators require little specialized knowledge and can automatically incorporate many real-world peculiarities. This paper reports a simulator that consisted of ANNs which were trained to predict changes in the position of a real-world snakelike robot. Navigational behaviours were evolved in simulation and subsequently verified on the real-world robot. This paper demonstrated that ANNs are a viable alternative to traditional simulators for evolving controllers for snake-like robots.