A biologically motivated controller for a six-legged walking system

A network is proposed for the control of a six-legged, insect-like walking system. The controller is designed after its biological model, the stick insect. This requires using only decentralized structures and local rules for the controller itself and using artificial neural networks for the implementation, The control network contains internal recurrent connections, but important recurrent connections utilize the loop through the environment. We show that this extremely decentralized controller, in combination with the exploitation of the physical properties of the world, is not only much simpler but also more adaptive than classical controllers. The result is a highly robust system which can adapt to unpredictable external conditions with no need for specific reprogramming.