Reliable AI for Marine Robotics

Robot designers are increasingly relying on Artificial Intelligence methods to make robots more robust. AI, in contrast to classical control, allows robots to adapt better to changing or noisy condition (noise here may mean friction in movement, imprecision of localization, imprecision of actuation etc.). Unfortunately, at the same time, AI, reduces trustworthiness of robots, as it is very hard to predict their behavior ahead of time, and assure safety. This is particularly difficult for robots that are learning during operation.

This consortium, consisting of leading robotics, AI, and software safety research labs in Europe, will train research experts in methods for assuring safety for machine learning based controllers, including online-learning and transfer-learning controllers. It will be supported by large European robotics players in robotics market who are the receiving stakeholders for such safety assurance methods.