Motivated by the vision of Organic Computing, a concept is proposed towards building a more robust robot control architecture. An experimental setup is described that allows to develop, implement and test new approaches and their practical realizations on a walking robot demonstrator. These experiments demonstrate that normal walking of the robot can be distinguished from the walking with a defective leg or against an obstacle by observing appropriate sensor signals from the legs. Based on this, the properties of self-adaptation and self-reconfiguration in case of malfunctions of some parts or in case of obstacle avoidance can be realized.