Marine Vessel Inspection as a Novel Field for Service Robotics: A Contribution to Systems, Control Methods and Semantic Perception Algorithms.

This cumulative thesis introduces a novel field for service robotics: the inspection of marine vessels using mobile inspection robots. In this thesis, three scientific contributions are provided and experimentally verified in the field of marine inspection, but are not limited to this type of application. The inspection scenario is merely a golden thread to combine the cumulative scientific results presented in this thesis. The first contribution is an adaptive, proprioceptive control approach for hybrid leg-wheel robots, such as the robot ASGUARD described in this thesis. The robot is able to deal with rough terrain and stairs, due to the control concept introduced in this thesis. The proposed system is a suitable platform to move inside the cargo holds of bulk carriers and to deliver visual data from inside the hold. Additionally, the proposed system also has stair climbing abilities, allowing the system to move between different decks. The robot adapts its gait pattern dynamically based on proprioceptive data received from the joint motors and based on the pitch and tilt angle of the robot’s body during locomotion. The second major contribution of the thesis is an independent ship inspection system, consisting of a magnetic wall climbing robot for bulkhead inspection, a particle filter based localization method, and a spatial content management system (SCMS) for spatial inspection data representation and organization. The system described in this work was evaluated in several laboratory experiments and field trials on two different marine vessels in close collaboration with ship surveyors. The third scientific contribution of the thesis is a novel approach to structural classification using semantic perception approaches. By these methods, a structured environment can be semantically annotated, based on the spatial relationships between spatial entities and spatial features. This method was verified in the domain of indoor perception (logistics and household environment), for soil sample classification, and for the classification of the structural parts of a marine vessel. The proposed method allows the description of the structural parts of a cargo hold in order to localize the inspection robot or any detected damage. The algorithms proposed in this thesis are based on unorganized 3D point clouds, generated by a LIDAR within a ship’s cargo hold. Two different semantic perception methods are proposed in this thesis. One approach is based on probabilistic constraint networks; the second approach is based on Fuzzy Description Logic and spatial reasoning using a spatial ontology about the environment. For each of the three mentioned key contributions of the thesis, the corresponding publications are provided in the Appendix.