Evolving neural network controllers to produce leg cycles for gait generation

The generation of gaits for hexapod locomotion controllers can be divided into two main parts: the cyclic action of a single leg (leg cycles) and the coordination of all legs to combine individual leg cycles to produce forward movement. In this paper, we use a genetic algorithm (GA) to evolve the structure of an artificial neural network (NN) that produces leg cycles in a hexapod robot. The movement of the robot´s leg is controlled by a horizontal servo and vertical servo. The servos are controlled by a NN that generates a cycle of pulses. With minimal restrictions on the structure of the NN a GA is used to find the parameters of neurons and the connections between them. The pulse sequences generated by the evolved NNs resulted in leg cycles that produced efficient forward movement.