In co-operation with biologists of the university of Jena and mechanics of the university of the Duisburg the control algorithms of the quadruped, mammalian-like walking machine BISAM are developed. The main aspect is to take the insights of the biologist and the dynamic simulations of the mechanics to realise a versatile adaptive control architecture. The key element of this architecture is the usage of small functional units, called reflexes from their biological paragon, which are arranged in a hierarchical structure. These reflexes are realised by radial basis function neural networks trained online with reinforcement learning methods or by fuzzy controllers which are optimised using reinforcement learning. Current work deals with integration of the complete dynamic simulation in the design process of the reflex layers and the extension of the reflex system on higher layers. Further goals include the development of a full adaptive and flexible pattern generator allowing smooth, sensor based transition between gaits, the mechanical redesign of some parts of BISAM, and the construction of a small mammalian-like leg with fluidic muscle as actuators.