We present a generative approach to behavioral organization which uses continuous dynamical systems for switching between multiple actions of an ant hropomorphic robot. The logical context of the possible actions is coded in matrices of parameters for a system of continuous differential equations. The sensor context is represented by a set of variable parameters depending on the sensor inputs. The switching between the behaviors is the result of nonlinear phase transitions in the solution of the underlying dynamical system. The stability of the overall system is guaranteed even for many different behaviors by keeping coupled behaviors on separated timescales. This is demonstrated on the anthropomorphic robot ARNOLD for the example of approaching a door. The robot searches for the door visually, recognizes it, points at it, and approaches it. The dynamical systems approach allows for a flexible change of a running behavioral sequence after unexpected perturbations brought about by dynamic changes in the environment.