Classifying and Recovering from Sensing Failures in Autonomous Mobile Robots

Proceedings of the AAAI Conference on Artificial Intelligence, 13

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 13

Mobile Robots Video

This paper presents a characterization of sensing failures in autonomous mobile robots, a methodology for classification and recovery, and a demonstration of this approach on a mobile robot performing landmark navigation. A sensing failure is any event leading to defective perception, including sensor malfunctions, software errors, environmental changes, and errant expectations. The approach demonstrated in this paper exploits the ability of the robot to interact with its environment to acquire additional information for classification (i.e., active perception). A Generate and Test strategy is used to generate hypotheses to explain the symptom resulting from the sensing failure. The recovery scheme replaces the affected sensing processes with an alternative logical sensor. The approach is implemented as the Sensor Fusion Effects Exception Handling (SFX-EH) architecture. The advantages of SFX-EN are that it requires only a partial causal model of sensing failure, the control scheme strives for a fast response, tests are constructed so as to prevent confounding from collaborating sensors which have also failed, and the logical sensor organization allows SFX-EH to be interfaced with the behavioral level of existing robot architectures.