Multi-Layered Model Based Diagnosis in Robots

The problem of fault diagnosis in the domains of robotics and autonomous systems are unique and interesting. A hidden internal fault affects not only other hardware components, as in any other machine, but also the different layers of abstraction and control in the sensethink-act cycle that the robot carries. We propose to implement a model-based diagnosis approach by utilizing the robot's controlarchitecture as a model to detect faults and to associate high-level abstractions with lowlevel hardware components. Furthermore, other approaches use behavioral models by representing the models with analytical equations, making it impractical to construct for very complex robots. We propose a structural model by representing only the dependencies in the model, which is significantly easier to construct. We demonstrate the feasibility of our approach by diagnosing high fidelity faults in a simulated Unmanned Aerial Vehicle (UAV) and a laboratory robot.