Safe Autonomy in Marine Robotics

Abstract

The use of marine robots has become increasingly common in various disciplines and industries, such as oceanography, marine biology, archaeology, transportation, offshore energy, fisheries, and security. However, the marine environment is harsh and unstructured. Here, robots are exposed to currents, winds, and waves, and operations underwater are further complicated by challenges in communication and navigation, complex terrain, and obstacles, like rocks and fishing gear. As robots move into more challenging and extreme environments, e.g., under ice and in coastal areas, or as the system complexity increases, e.g., in multi-agent setups, new challenges arise. The ability to effectively understand and manage risks is crucial for safe and efficient operations of marine robots. Safety in marine robot autonomy can be improved in several ways. For example, one may design algorithms for supervisory risk control, using both deliberative and reactive approaches, enabling the robot to monitor and control the risk autonomously. Also, new algorithms, concepts and overall improvements to the control system may contribute to improved safety in marine robot autonomy. This thesis aims to contribute to the development of safer and more efficient autonomy in marine robotics by combining methods from control theory, artificial intelligence (AI) and risk science. The main contributions of this thesis are summarized as follows: - Comprehensive hazard identification and hazard analysis for operations of autonomous underwater vehicles (AUVs) under ice, in coastal areas, and networks of AUVs and autonomous surface vehicles (ASVs). - Development of online risk models based on the results of hazard analyses for modeling and quantifying the dynamic risk levels during operation. - Methods for supervisory risk control which use online risk models for decision-making by re-configuring relevant control parameters in a risk-optimal manner. - A methodology for developing risk maps, which allows for including aspects of risk into the robot's world model. - A method for risk-based path planning using the concept of risk maps, for achieving paths that balance the trade-off between different hazards and mission objectives. - Design of a switching controller for tracking of one or multiple AUVs with an ASV, which adaptively switches between modes for tracking, collision avoidance and standby depending on the situation. - Kalman filter design running onboard the ASV for state estimation of multiple AUVs. - Design and stability analysis of a novel observer for underwater navigation using the hybrid systems framework. - New operational concepts and lessons learned from field operations. In conclusion, combining methods from control theory, AI and risk science into algorithms for planning, decision-making, and control at different levels in the control architecture shows promise for developing safer and more efficient marine robotics and advancing safety in autonomous operations. This thesis has made a number of contributions and demonstrated their applicability through relevant case studies.