Evolution of behaviors in autonomous robot using artificial neural network and genetic algorithm

Abstract

In distributed autonomous robot (agents) systems, each robot (predator or prey) must behave by itself according to its states and environments, and if necessary, must cooperate with other robots in order to carry out a given task. Therefore it is essential that each robot have both learning and evolution ability to adapt to dynamic environment. This paper proposes a pursuing system utilizing the artificial life concept where autonomous mobile robots emulate social behaviors of animals and insects and realize their group behaviors. Each robot contains sensors to perceive other robots in several directions and decides its behavior based on the information obtained by the sensors. In this paper, a neural network is used for behavior decision controller. The input of the neural network is decided by the existence of other robots and the distance to the other robots. The output determines the directions in which the robot moves. The connection weight values of this neural network are encoded as genes, and the fitness individuals are determined using a genetic algorithm. Here, the fitness values imply how much group behaviors fit adequately to the goal and can express group behaviors. The validity of the system is verified through simulation. Besides, in this paper, we could have observed the robots’ emergent behaviors during simulation.

Introduction

When facing difficult problems, we can often find the solution from creatures in nature. From this kind of attempts, an research area of artificial life (AL) came into being by Langton in 1987, in purpose of unifying hitherto individually achieved research results of living creatures and further activating those researches [1], [2], [3], [4].

In engineering aspects, the goal of the artificial life is to incarnate the unique behaviors or phenomena of living creatures in nature onto artifacts like computers. We expect the artificial life can provide a useful methodology for the autonomous mobile robot learning which is full of autonomy and creativity. One of basic concept of the artificial life is “emergence”. According to its mechanism, the emergence can be divided into several classes. Most of them determine behaviors and interaction rules of lower-level components. The local interaction between components generates global orders or behaviors [5], [6], [7], [8], [9], [10], [11], [12], [13].

The present paper develops a pursuit system that is mostly applied to the area of robot and robot in order to apply an emergent characteristic of the artificial life into machine learning. The pursuit system contains robots and preys. In the system, robots catch the preys escaping from them through evolution. For the simulator of the system, we have modeled the structure of the robot with the artificial neural network and evolved the structure of the neural network with the genetic algorithm. The prey is given only pre-knowledge of direction identification to move.

Numerous researches about the autonomous mobile robot control in the pursuit system have been studied. Nolfi and Foreano [23] simulated pursuit system with two robots (a predator and a prey) in real environments. They evolved both robots reciprocally with genetic algorithm. Yasuo and Shin [15] controlled the autonomous mobile robots using reinforcement learning. Kam-Chuen [18] evolved the autonomous mobile robots with the genetic algorithm. By II-Kwon [19], both fuzzy controller and genetic algorithm were used. The fuzzy function displayed robot’s position, and the genetic algorithm was used for learning. The reinforcement learning method is to develop robot’s behaviors by means of the interrelationship with environment and resulting reinforcement signals. It can guarantee learning and adaptability without precise pre-knowledge about environments. However, its critical weak point is the difficulty of learning when the rewards of taken actions are not instantly computed. Takayuki and Yasuo [14], [15] have a problem in applying to the open environments which is dynamically changing since once a robot’s role is decided, then it is fixed. By Jeong and Lee [19] and Pena-Reyes and Sipper [20] both robots and prey move randomly, and the genetic algorithm helps to prevent the collision between robots. However, robots were not intelligent to do learning for themselves. Also, since the gene structure was represented by fuzzy member functions, it took much time to resolve pursuing problems.

Therefore, in this paper, to resolve those problems, we apply the artificial life method in our system. We model the robot structure with the artificial neural network that is easy to model both robots and use the genetic algorithm for robot learning, which results in giving intelligence to robots in their pursuing movement toward the prey. The prey has only pre-knowledge to identify distance to avoid the robots. We have designed a virtual environment where 20 robots and a prey. Furthermore, various selection methods of the genetic algorithm are considered for simulation.

The rest of this paper is organized as follows. In Section 2, the related works of this paper and is described. Section 3 describes the evolution of autonomous robots in pursuit system. In Section 4, simulation and evaluation are described. Finally, conclusion is presented in Section 5.