Evolution of Neural Controllers for Simulated and Real Quadruped Robots

Evolutionary robotics is an approach that employs evolutionary computation to develop a controller for an autonomous robotic system. Evolutionary computing usually operates depending on a population of candidate controllers, initially selected from a random distribution. The population is iteratively modified according to the fitness function. In this paper, an automatic control system is designed for quadruped robots using an Evolutionary Neural Network (ENN) and the performance is measured in terms of the distance travelled by the robot from its origin. The evolved neural controllers are analyzed in the simulation environment and the results are implemented in a real quadruped robot. The comparison between the simulated and real robot shows the performance of the quadruped robot in terms of number of iterations over the distance covered in the desired direction. The developed ENN helps the robot to choose the best possible solution to achieve the maximum distance.