Hierarchical neural network controller for legged robots

Biological systems provide great motivation by virtue of their striking performance as an indication of what is possible. Because robots face the same physical laws and environmental constraints as those faced by biological systems in similar tasks, the solutions they use may embrace similar principles. In this context, the recent development of the pneumatically driven rubber actuators [1] has been a significant advance in the realization of human-like robots. Rubber actuators are quite different from the electric and hydraulic actuators by their properties of elasticity, compactness and low inertia and hence are useful for not only industrial manipulators but also for rehabilitative prostheses. Because of their uniqueness their control schemes would require special consideration; however, very little work has been done till date in this area and only conventional control schemes have been employed. Since these actuators are functionally similar to animal muscles, the possibility exists of utilizing biologically evident motor control schemes for effecting manipulator movements as well. This paper reports the implementation of the 'equilibrium point control scheme' believed to effect musculo-skeletal control in humans and other 'higher' animals for the purposes of legged robot locomotion.