Neural networks robot controller trained with evolution strategies

摘要: Neural networks (NN) can be used as controllers in autonomous robots. The specific features of the navigation problem robotics make generation good training sets for NN difficult. An evolution strategy (ES) is introduced to learn weights instead learning method network. ES high performance reactive behavior and collision avoidance. No subjective information about "how accomplish task" has been included fitness function. learned behaviors are able solve different environments; therefore, process proven ability obtain a specialized behavior. All obtained have tested set environments capability generalization shown each A simulator based on mini-robot, Khepera,