An autonomous agent needs to be aware of not just the environment but also of itself. This self-monitoring can be used by the agent to recognize the onset of a problem in terms of its own characteristics rather than the specifics of the environment. In this paper, we propose the use of a hybrid model for such self-monitoring. The model is a combination of two approaches – a rule-based approach that uses heuristics to classify states and an instance-based approach that classifies states by retrieving the closest matching state from memory. The instance-based approach has a learning phase where it learns to associate a sequence of sensor states with a particular behavior. We construct this learning set from the results of heuristic classification, thus negating the need for annotating the ground truth. We demonstrate this approach using a model built on the ACT-R cognitive architecture that monitors the performance of an autonomous robotic agent navigating in the real world. This model learns the associations between the position, speed and odometer values, and three operational states – Moving, Looping and Stuck. We compare the hybrid model’s performance against the ground truth and the results from rule-based and instance-based models separately to demonstrate the benefits of combining approaches.