This article details the SFX-EH architecture for handling sensing failures in autonomous mobile robots. The SFX-EH uses novel extensions to the generate-and-test method to classify failures with only a partial causal model of the sensor/environment/task interactions for the robot. The generate-and-test methodology exploits the ability of the robot as a physically situated agent to actively test assumptions about the state of sensors, condition of the environment, and validity of task constraints. The SFX-EH uses the type of failure to determine the appropriate recovery strategy: reconfiguration of the logical sensor or logical behavior, recalibration of the sensor or actuator, and corrective actions. The system bypasses classification if all hypotheses lead to the same recovery strategy. Results of the SFX-EH running on two robots with different sensor suites and tasks are presented, demonstrating intelligent failure recovery within a modular, portable implementation.