Cognitive robots need to detect execution failures in runtime to prevent potential damages to their environments or objects in their interest. For this reason, robots monitor the execution to detect any inconsistencies. In this paper, we propose a hybrid monitoring system that processes different sensory data from different sources by using both model-based and model-free methods. Our proposed system continually monitors formulas, represented in Metric Temporal Logic (MTL), related to each action in order to detect failures. These formulas are analysed temporally according to the sensory data gathered during the execution to decide on a failure.