In this Section decentralized locomotion control inspired by the stick insect model is dealt with. In this model the locomotion pattern is the result of a set of local influences among leg controllers whose behavior is regulated by reflexes triggered by sensors. In this sense this scheme is totally different from the CPG, in which the generation of the driving signals does not rely on sensory feedback.

Two different implementations of decentralized locomotion control are presented here. The first is based on CNNs, the second on integrate-and-fire neurons. The two approaches share the idea of using nonlinear dynamical systems at the low level of locomotion control. This implementation strategy is also common to the CNN-based CPG presented in the previous Chapters. The capability to solve the complex problems of legged robot locomotion by means of distributed control in which the computational effort is divided between the self-organizing nonlinear units is a peculiar feature of this implementation strategy. Moreover, the intrinsic self-organization of the networks makes the system robust to parameter changes and faults.

4.1 CNN-Based Decentralized Control Model

The stick insect (Fig. 4.1) is a fascinating insect that slowly walks by adopting a large variety of reflexes. Its behavior is usually given as an example of an approach for locomotion control totally different from the CPG. The locomotion control is based only on reflexes.

Fig. 4.1: A photo of a stick insect.

In fact, experiments carried out on the stick insect by Cruse [Cruse et. al. (1998a)] led...