Evolving connection weights between sensors and actuators in robots

In this paper, an evolution strategy (ES) is introduced, to learn reactive behaviour in autonomous robots. An ES is used to learn high-performance reactive behaviour for navigation and collisions avoidance. The learned behaviour is able to solve the problem in a dynamic environment; so, the learning process has proven the ability to obtain generalised behaviours. The robot starts without information about the right associations between sensors and actuators, and, from this situation, the robot is able to learn, through experience, to reach the highest adaptability grade to the sensors information. No subjective information about "how to accomplish the task" is included in the fitness function. A mini-robot Khepera has been used to test the learned behaviour.