The paper introduces taxonomy to classify these failures, providing a wide view of what might happen in the execution of cognitive tasks. To overcome and recover from most of the failures obtained in the taxonomy, the paper would introduce a specific Robust Planning Framework designed for cognitive robots. The framework blends processes like planning, reasoning, and learning to create the bigger picture of more reliable task execution. Detection, reasoning, and subsequent re-planning are used to deal with failures. Additionally, incrementally, new hypotheses can be learned by the robot about its experiences through the framework. The framework is tested on a Pioneer3DX robot, demonstrating its ability to handle temporary failures that would arise from the execution of specific actions during runtime. Preliminary results show promising outcomes and point toward the hypothesis learning approach being viable in failure scenarios.