This paper presents a set of mechanisms allowing the detection of control architecture inconsistencies, in the context of autonomous mobile robotics. This approach integrates an observation level into the structure of the control architecture to monitor and analyze the internal state of the robot and detect inconsistencies. These data are processed and the most pertinent information are sent to the Human supervisor and/or to a decision level which can adjust the robot autonomy according to the identified problem. These mechanisms have been implemented using architectural concepts previously developed at LIRMM. Based on these developments, adaptive hybrid control architecture is currently under development.