Robust Perception for Underwater Robotics

Modern perception methods build semantic world models from raw sensor observations, facilitating high level learning and reasoning tasks of autonomous robots. A wide range of methods fall into this field: From low-level sensor-specific processing, the combination of simultaneous localization and mapping (SLAM) with semantic concepts, methods addressing dynamic scenes, up to object recognition and tracking. This talk introduces the notion of robust perception which explicitly takes into account the presence of outliers and false positives on all levels of processing, thus achieving high reliability needed for fully autonomous and/or critical applications.