Abstract
The utilization of a multi-objective approach for evolving artificial neural networks that act as the controllers for phototaxis and radio frequency (RF) localization behaviors of a virtual Khepera robot simulated in a 3D, physics-based environment is discussed in this chapter. It explains the comparison performances among the elitism without archive and elitism with archive used in the evolutionary multi-objective optimization (EMO) algorithm in an evolutionary robotics study. Furthermore, the controllers’ moving performances, tracking ability and robustness also have been demonstrated and tested with four different levels of environments. The experimentation results showed the controllers allowed the robots to navigate successfully, hence demonstrating the EMO algorithm can be practically used to automatically generate controllers for phototaxis and RF-localization behaviors, respectively. Understanding the underlying assumptions and theoretical constructs through the utilization of EMO will allow the robotics researchers to better design autonomous robot controllers that require minimal levels of human-designed elements.
TopIntroduction
Evolutionary Robotics (ER) is a methodology to develop a suitable control system for autonomous robots with minimal or without human intervention at all through evolutionary computation, to adapt itself to partially unknown or dynamic environments (Floreano, 2000b; Nelson, 2006; Pratihar, 2003). In other words, ER is defined as the synthesis of autonomous robots and/or controllers using artificial evolutionary methods (Teo, 2004a; Teo, 2005). ER is mainly seen as a strategy to develop more complex robot controllers (Floreano, 1996; Floreano, 2000b). Algorithms in ER frequently operate on a population of candidate controllers, initially selected from some random distributions (Alba, 2002; Urzelai, 2000). The evolutionary processes involve a set of operators, namely selection, crossover, mutation and other genetic algorithm (GA) operators (Coello, 2005a; Floreano, 2000a). Using different approaches of evolution, for example Genetic Algorithm, Genetic Programming, Co-evolution, and anytime learning, researchers strive for an algorithm that is able to train robots to perform their tasks without external supervision or help.
A number of studies have already been successfully conducted in evolutionary robotics for phototaxis, phonotaxis and obstacle avoidance tasks (Floreano, 2000; Teo, 2005; Floreano, 1996; Floreano, 2000a; Horchler, 2004; Reeve, 2005). In previous studies related to phototaxis tasks, the researchers used a fixed amount of hidden neurons in the neural network or just a two-layer neural network for their robot’s task (Floreano, 1996; Floreano, 2000b; Teo, 2004b). They have not emphasized on the relationship between the robot’s behavior and its corresponding hidden neurons. Additionally, to the best of our knowledge, there have not been any studies conducted yet in applying the multi-objective algorithm in evolving the robot controllers for phototaxis behavior. Besides, the literature also showed that other researchers have successfully synthesized some fitness functions to evolve the robots for the required behaviors (Floreano, 1996; Floreano, 2000a; Floreano, 2000b; Nolfi, 2000). However, the fitness functions used can be further improved or augmented in order to increase the robot’s ability in completing more complex tasks (Marco, 1996). In addition, research regarding radio frequency (RF) signal localization has yet to be studied in ER. The RF signal is defined as radio frequency signal (abbreviated RF, rf, or r.f.) (Gibson 2007; Pike, 2007; NOAA 2008). It is a term that refers to an alternating current having characteristics such that, if the current is an input to an antenna, an electromagnetic field is generated suitable for wireless broadcasting and/or communications used (Gibson 2007; Pike, 2007; NOAA 2008). The RF signal source has provided the capability for improvements in tracking, search and rescue efforts. As such, robots that are evolved with RF-localization behavior may potentially serve as an ideal SAR assistant (Gibson 2007; Pike, 2007; NOAA 2008).
Key Terms in this Chapter
Evolutionary Robotics: is a method that utilizes evolutionary computation to automatically synthesize controllers for autonomous robots. In other words, it refers to the application of artificial evolution to generate autonomous robots and/or their controllers with minimal or no direct input from humans.
Evolutionary Computation: refers to a subfield of artificial intelligence or computational intelligence that involves computational algorithms that are inspired by biological processes.
Radio Frequency Localization Behavior: is a kind of taxis behavior that occurs when an entity navigates in response to the propagation of radio frequency. This behavior is applied to the robotics area as an ideal Search and Rescue (SAR) approach.
Evolutionary Multi-objective Optimization: is known as multi-attribute or multi-criteria evolutionary optimization that involves simultaneous processes that optimize two or more conflicting objectives which may be subject to certain constraints.
Autonomous Mobile Robots: are robots that have the capability to navigate using their integrated sensors and wheel(s) independently using its onboard controllers as well as able to perform the requested tasks in an unstructured environment.
Artificial Neural Networks: is a computational model based on biological neural networks which consists of an interconnected group of artificial neurons that have been practically used in prediction, classification, control problem and approximation. In other words, ANNs are an adaptive system that changes its structure based on internal or external information that flows through the network during the learning phase. ANNs consist of a set of nodes (input, hidden and output neurons) connected by weights.
Phototaxis Behavior: is a kind of taxis behavior that occurs when a whole organism navigates or tracks in response to a light stimulus. The movement of the organism in the direction of light is defined as positive. Otherwise, it is negative. This is advantageous for phototrophic organisms to orient themselves towards light sources to receive energy for photosynthesis. This behavior is applied to the robotics area as a homing behavior.
Pareto Differential Evolutionary Algorithm: is a term that refers to hybridization of Evolutionary Multi-objective Optimization (EMO) into the Differential Evolution algorithm (DE).