Efficient Hybrid Fault Detection for Autonomous Robots

The use of robots has increased significantly in the recent years; rapidly expending to numerous applications. Yet, these sophisticated and sometimes expensive machines are susceptible to faults that might endanger the robot or its surroundings (e.g., a crash of an Unmanned Aerial Vehicle (UAV)). To prevent such faults, the robot's operation needs to be monitored by Fault Detection (FD) algorithms. An autonomous robot, which is already engaged with heavy computational tasks, has to continuously apply FD on its own. Thus, the impact of a FD process on the robot's resources should be minimized. Unfortunately, the computational load of existing FD approaches, which may be very accurate, might be impractical for an autonomous robot. To solve this problem, we suggest to use a hybrid approach. A very efficient FD algorithm is applied continuously and is used to trigger a heavier, more accurate, FD approach that determines the occurrence of a fault. In this paper we focus on the efficient FD algorithm. We test the algorithm in several real-world and simulated domains and we show and discuss the promising results.