Worst-Case Analysis of Complex Nonlinear Flight Control Designs Using Deep Q-Learning

What is it about?

This paper presents the use of Reinforcement Learning techniques for identifying potentially hidden design deficiencies in complex flight control systems. The application of learning-based methods for the purpose of penetration testing shows great potential in discovering adverse situations that go beyond the intuition of design and testing engineers. This is especially true for novel or highly augmented aircraft configurations for which knowledge about regions of adverse system behavior is still sparse.

Featured Image