Reliable AI for Marine Robotics (REMARO)
“Meta-control for reliable autonomous operation of underwater robots”
The goal of this project is to develop meta-control solutions to improve the autonomy and reliability of robots for underwater operations by adapting at runtime their mission plan and control architecture simultaneously.
Meta-control is a combination of self-adaptive systems and cognitive control methods to develop supervisory modules for hybrid, AI based robot control architectures. The key idea is to bridge the three dimensions of an autonomous system: system, mission and environment, using semantic knowledge.
The project will explore model-based systems engineering (MBSE) and knowledge representation methods to capture that semantic knowledge about the robot’s control architecture, and ontological reasoning to diagnose the situation from perception data and drive the runtime adaptation. This will overcome the shortcomings of current approaches for robot control, e.g. based only on deep learning, that do not address the three dimensions.
The method will be demonstrated on a real underwater robot for oceanography applications. The developed solution will enable the robot to handle both component faults and task-related contingencies, e.g. particles occluding vision or losing grip on a part being replaced, while handling the execution of a complex mission plan involving different cognitive modules, robot’s skills, and physical resources.